Information processing system

Abstract

The application provides an information processing system. An information processing system characterized by comprising: a processor; wherein the processor is configured to: use a prompt word for issuing an execution translation instruction to a generative artificial intelligence model to parse input content to generate information for foreigners and perform corresponding translation processing; use a prompt word for issuing an execution disaster prevention information translation instruction to a generative artificial intelligence model to parse input content to generate disaster prevention information and perform corresponding translation processing; and use a prompt word for issuing an information source evaluation benchmark instruction to a generative artificial intelligence model to parse information sources to select information sources with high credibility and classify the information sources.

Description

Technical Field

[0001] The technology disclosed herein relates to an information processing system. Background Technology

[0002] Japanese Patent Application Publication No. 2022-180282 discloses a method for controlling a role-based chatbot executed by at least one processor. The method includes the following steps: receiving a user's speech; adding the user's speech to a prompt word, the prompt word containing instruction statements associated with an explanation of the chatbot's role; encoding the prompt word; and inputting the encoded prompt word into a language model to generate a chatbot response to the user's speech.

[0003] The main challenges this invention addresses are as follows: While generative artificial intelligence (AI) technology has enabled the automatic generation and translation of multilingual texts, existing technologies still have several shortcomings when providing information to foreign users, such as those residing in Japan. Specifically, firstly, existing automatic translation systems typically only translate the input text directly, lacking contextual adaptation and expression adjustments to meet the information needs of foreigners, making it difficult to generate timely and accurate information that is easily understood by them. Secondly, disaster prevention information is often released in the user's native language, technical terms, and administrative language. Traditional translation methods struggle to promptly grasp the disaster prevention-related content input by users and provide information suitable for foreigners' understanding while maintaining accuracy, leading to difficulties for foreigners in obtaining crucial information in emergencies. Thirdly, for various information sources on the internet, existing systems often rely on manual screening or simple keyword matching, failing to fully utilize generative AI models to analyze information sources based on evaluation criteria, thereby automatically selecting and reasonably classifying high-credibility information sources, making it difficult for users to efficiently obtain reliable information. Furthermore, existing question-answering systems, when processing user questions, typically fail to fully integrate the semantic understanding capabilities of generative AI models for in-depth analysis and answer generation, resulting in insufficient accuracy and relevance in the responses. Furthermore, existing technologies generally lack mechanisms to incorporate user emotional states, failing to leverage generative AI for sentiment analysis of user input and dynamically adjust information content and presentation based on emotional states. This hinders the improvement of user experience and information acceptance. Therefore, constructing a system that utilizes prompts for generative AI models to instruct them to perform translation, information source analysis, question answer generation, and sentiment analysis, in order to provide more accurate, reliable, and user-friendly information services to foreigners, has become a pressing technical challenge in this field. Summary of the Invention To address the aforementioned issues, this invention provides an information processing system comprising a processor. The processor, through a specific processing flow and prompt word design, collaborates with a generative artificial intelligence model to parse and generate information from various input contents. Specifically, the processor is configured to: First, for general and everyday information intended for foreigners, the processor uses prompt words to instruct the generative artificial intelligence model to perform translation, parsing the original content input by the user or the original information acquired by the system. This parsing process enables the generative artificial intelligence model to understand the semantics, context, and target user attributes of the input content, thereby generating translations and explanatory content suitable for foreigners to understand, thus solving the problem of traditional translation merely performing literal conversion. Second, for disaster prevention information, the processor uses specially designed prompt words to instruct the generative artificial intelligence model to perform disaster prevention information translation, parsing disaster-related input content. This enables the generative artificial intelligence model to generate structured, clear, and easily understood disaster prevention information suitable for foreigners, while ensuring the accuracy of key information, thereby providing highly actionable guidance to foreigners in emergency situations. Furthermore, regarding the reliability of information sources, the processor utilizes prompt words to issue information source evaluation criteria instructions to the generative artificial intelligence model. It performs semantic parsing and feature extraction on candidate information sources and their related descriptions, enabling the generative artificial intelligence model to automatically evaluate and screen information sources based on predetermined evaluation criteria such as credibility, authority, update frequency, and language support. It selects information sources with high credibility and further classifies these information sources by theme or purpose, providing a structured foundation for subsequent information retrieval and display.

[0004] Based on the above, according to one embodiment of the present invention, the processor is further configured to: when a user asks a question, use prompt words to send instructions to the generative artificial intelligence model to parse the question content and generate an answer, perform detailed analysis of the user's input question text, identify the question intent, key entities, and contextual conditions, thereby driving the generative artificial intelligence model to generate answer content that matches the question, making the answer more accurate, specific, and executable, thereby improving the quality of the question-and-answer service. According to another embodiment of the present invention, the processor is further configured to: use prompt words to send instructions to the generative artificial intelligence model to parse the user's input, perform sentiment analysis on the user's input content, identify the user's emotional tendencies such as tension, anxiety, confusion, or calmness, and adjust the information provision strategy according to the analyzed emotional state, for example, providing more reassuring and explanatory explanatory text or disaster prevention guidance when the user is tense or anxious, and providing concise and direct information when the user only makes a general query, thereby realizing the dynamic optimization of information content and expression based on the user's emotions. By comprehensively applying the above structures and methods, the system of this invention can, within the same framework, utilize various prompt words designed for generative artificial intelligence models to uniformly manage and collaboratively process multiple tasks such as translation, disaster prevention information generation, information source screening and classification, question answer generation, and sentiment analysis. This effectively addresses the shortcomings of existing information services for foreigners in terms of accuracy, reliability, and user experience, significantly improving the system's service quality and adaptability for foreign users. "System" refers to a comprehensive device or combination of devices that utilizes hardware and software resources to work together to perform functions such as translation, disaster prevention information generation, information source analysis, question answer generation, and sentiment analysis. It may include servers, processors, memory, network interfaces, and program modules that interact with generative artificial intelligence models.

[0005] A processor is a piece of hardware or a combination thereof that is capable of executing computer program instructions, performing calculations and logic control on input data. It can be a central processing unit (CPU), a graphics processing unit (GPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination of any of the above processing units.

[0006] "Generative artificial intelligence models" refer to artificial intelligence models that are trained on large amounts of data based on machine learning and deep learning technologies, and can automatically generate new text content, translation results, or analysis results based on input text or other data. These include, but are not limited to, large language models, text generation models, and multimodal generation models.

[0007] "Cue words" refer to a text or structured instruction that is input into a generative artificial intelligence model to instruct or guide the model to perform a specific task (such as translation, information parsing, evaluation, or response generation), including information such as task description, role setting, input format constraints, and output requirements.

[0008] "Input content" refers to various types of data provided by users or external systems and input into the system of this invention for the generative artificial intelligence model to analyze and process. It is usually text data, including but not limited to news text, disaster prevention notices, user questions, user comments and other explanatory texts.

[0009] "Information for foreigners" refers to various types of information generated or converted by the system of this invention, primarily targeting foreign users and using language and expression methods suitable for foreigners to understand, including information on daily life, explanations of administrative procedures, tourism information, precautions, and related explanatory texts.

[0010] "Disaster prevention information" refers to information related to natural disasters or other emergencies, used to alert users to risks, explain the disaster situation, and guide users to take appropriate evacuation or self-protection measures, including disaster warnings, response steps, and precautions for disasters such as earthquakes, typhoons, and floods.

[0011] "Information source" refers to the original source or carrier of information content, including websites, databases, official announcements, media reports, online service platforms, and other content providers that can be accessed and analyzed by the system.

[0012] "Evaluation criteria for information sources" refers to a set of standards or rules used to judge and evaluate the reliability, authority, and usefulness of information sources, including but not limited to the authority of the publishing entity, the frequency of information updates, the accuracy of the content, language support, user evaluation, and whether it is an official or recognized institution.

[0013] "Highly credible information sources" refer to information sources that, based on the evaluation criteria of information sources, are identified by the system of this invention as reaching a predetermined threshold or level in terms of information accuracy, authority, and reliability, such as government agency websites, public institution platforms, and well-known authoritative media.

[0014] "Classification" refers to the process of grouping and categorizing information sources according to their content theme, purpose, target users, or information type, in order to form classification results with a certain structure hierarchy and category labels, which are used to facilitate user retrieval and browsing.

[0015] "Question" refers to the query information that a user enters into the system to obtain certain information or resolve a certain question. It usually appears in the form of natural language text and includes questions about life, disaster prevention, procedures, tourism, etc.

[0016] "Answer" refers to the response content generated by the system of the present invention based on the questions raised by users through a generative artificial intelligence model. It is used to answer user questions or provide relevant guidance information, and may include step instructions, precautions, reference links, etc.

[0017] "User input" refers to various types of text information submitted by users to the system through terminal devices, including question texts, feedback, declarative texts, dialogue content, and other inputs that can be used for sentiment analysis or content understanding.

[0018] "Emotion" refers to the emotional or attitudinal characteristics extracted from user input, including but not limited to emotional states such as tension, anxiety, confusion, calmness, satisfaction, and dissatisfaction, which are used to guide the system in adjusting the way information is presented and the style of content.

[0019] "Sentiment analysis" refers to the process by which the system of this invention uses a generative artificial intelligence model to analyze the semantic and emotional features of user input in order to identify the user's current emotional tendency or attitude state, and uses the results to adjust the subsequent information delivery strategy. Attached Figure Description

[0020] Figure 1 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the first embodiment.

[0021] Figure 2 This is a conceptual diagram illustrating an example of the main functions of the data processing apparatus and smart device according to the first embodiment.

[0022] Figure 3 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the second embodiment.

[0023] Figure 4 This is a conceptual diagram illustrating an example of the main functions of the data processing device and smart glasses according to the second embodiment.

[0024] Figure 5 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the third embodiment.

[0025] Figure 6 This is a conceptual diagram illustrating an example of the main functions of the data processing device and head-mounted terminal according to the third embodiment.

[0026] Figure 7 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the fourth embodiment.

[0027] Figure 8 This is a conceptual diagram illustrating an example of the main functions of the data processing device and robot according to the fourth embodiment.

[0028] Figure 9 This represents an emotion map that maps multiple emotions.

[0029] Figure 10 This represents an emotion map that maps multiple emotions.

[0030] Figure 11 This is a sequence diagram illustrating the processing flow of the data processing system of the first embodiment.

[0031] Figure 12 This is a sequence diagram illustrating the processing flow of the data processing system in Application Example 1.

[0032] Figure 13 This is a sequence diagram illustrating the processing flow of the data processing system of the second embodiment.

[0033] Figure 14 This is a sequence diagram illustrating the processing flow of the data processing system in Application Example 2. Detailed Implementation

[0034] Hereinafter, an example of an implementation of the system according to the present disclosure will be described with reference to the accompanying drawings.

[0035] First, let me explain the terminology used in the following instructions.

[0036] In the following embodiments, the processor (hereinafter referred to as "processor") with reference numerals may be a single computing device or a combination of multiple computing devices. Furthermore, the processor may be a single computing device or a combination of multiple computing devices. Examples of computing devices include CPU (Central Processing Unit), GPU (Graphics Processing Unit), GPGPU (General-Purpose computing on Graphics Processing Units), APU (Accelerated Processing Unit), etc.

[0037] In the following embodiments, RAM (Random Access Memory), as indicated in the figures, is a memory that temporarily stores information and is used as working memory by the processor.

[0038] In the following embodiments, the memory, as indicated by the reference numerals, is one or more non-volatile storage devices that store various programs and parameters. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), disks (e.g., hard disks), or magnetic tapes.

[0039] In the following embodiments, the communication I / F (Interface) with reference numerals is an interface that includes a communication processor and an antenna, etc. The communication I / F is responsible for communication between multiple computers. As an example of a communication specification applicable to the communication I / F, wireless communication specifications such as 5G (5th Generation Mobile Communication System), Wi-Fi (wireless fidelity) (registered trademark), or Bluetooth (registered trademark) can be listed.

[0040] In the following embodiments, "A and / or B" is synonymous with "at least one of A and B". That is, "A and / or B" means that it can be only A, only B, or a combination of A and B. Furthermore, in this specification, when "and / or" connects to express more than three items, the same interpretation as "A and / or B" applies.

[0041] First Implementation Method Figure 1 An example of the configuration of the data processing system 10 according to the first embodiment is shown.

[0042] like Figure 1 As shown, the data processing system 10 includes a data processing device 12 and an intelligent device 14. A server can be cited as an example of the data processing device 12.

[0043] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0044] The smart device 14 includes a computer 36, a receiving device 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. In addition, the receiving device 38, output device 40, camera 42, and communication I / F 44 are also connected to the bus 52.

[0045] The receiving device 38 includes a touchscreen 38A and a microphone 38B, and receives user input. The touchscreen 38A receives user input via touch by detecting contact with an indicator (e.g., a pen or finger). The microphone 38B receives user input via sound by detecting the user's voice. The control unit 46A in the processor 46 sends data representing the user input received by the touchscreen 38A and microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data representing the user input.

[0046] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting data in a form perceptible to the user 20 (e.g., sound and / or text). The display 40A displays visual information such as text and images according to instructions from the processor 46. The speaker 40B outputs sound according to instructions from the processor 46. The camera 42 is a small digital camera equipped with an optical system such as a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0047] Communication I / F44 is connected to network 54. Communication I / F44 and 26 are responsible for sending and receiving various information between processor 46 and processor 28 via network 54.

[0048] Figure 2 The diagram shows an example of the main functions of the data processing device 12 and the smart device 14.

[0049] like Figure 2 As shown, in the data processing apparatus 12, specific processing is performed by the processor 28. A specific processing program 56 is stored in the memory 32. The specific processing program 56 is an example of a "program" as understood in this disclosure. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0050] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. In the emotion inference function (emotion-specific function) using the emotion-specific model 59, various inferences and predictions related to the user's emotions are performed, including inferences and predictions of the user's emotions, but this is not limited to this example. Furthermore, emotion inference and prediction may also include, for example, emotion analysis (parsing).

[0051] In the smart device 14, the processor 46 performs the acceptance output processing. The memory 50 stores the acceptance output program 60. The acceptance output program 60 is used in conjunction with the data processing system 10 and the specific processing program 56. The processor 46 reads the acceptance output program 60 from the memory 50 and executes the read acceptance output program 60 on the RAM 48. The acceptance output processing is implemented by the processor 46 acting as the control unit 46A according to the acceptance output program 60 executed on the RAM 48. Furthermore, the smart device 14 has the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and these models can also be used to perform the same processing as the specific processing unit 290. The acceptance output processing is implemented by the processor 46 acting as the control unit 46A according to the acceptance output program 60 executed on the RAM 48.

[0052] Alternatively, other devices besides the data processing device 12 may also have the data generation model 58. For example, a server device (e.g., a generation server) may have the data generation model 58. In this case, the data processing device 12 obtains the processing results (prediction results, etc.) using the data generation model 58 by communicating with the server device that has the data generation model 58. Furthermore, the data processing device 12 may be a server device or a user-held terminal device (e.g., a mobile phone, robot, home appliance, etc.). Next, an example of the processing of the data processing system 10 of the first embodiment will be described.

[0053] Example 1 The flow of a specific process in Example 1 will be described. Each part of the system described below is implemented by the data processing device 12 and the smart device 14. Furthermore, the data processing device 12 is referred to as the "server," and the smart device 14 is referred to as the "terminal."

[0054] In existing technologies, when providing multilingual disaster-related and lifestyle-related information to foreign language users living in specific regions, a loosely combined architecture of "information retrieval system + independent machine translation service + general conversational AI service" is typically used. This type of architecture suffers from the following technical problems: (1) Fragmented information processing flow. The information capture module, translation module, information source evaluation module and dialogue question and answer module are independent of each other and lack unified control logic and data flow design. This results in the same information being repeatedly parsed and reassembled in different modules, which increases processing latency and resource consumption.

[0055] (2) Generative AI model invocation method is crude. Existing systems mostly ask questions to generative AI models directly in the form of natural language without structured design of prompts or differentiated control for different tasks such as "information source credibility assessment", "multi-turn question answering" and "emotion adaptive presentation". This results in unstable model behavior, difficult-to-reproduce results, and difficulty in meeting the technical requirements for security, reliability and controllability.

[0056] (3) Insufficient information source credibility and category management. For diverse information sources from the Internet, there is a lack of a technical framework for unified evaluation and hierarchical management using generative artificial intelligence models. It is difficult to perform fine-grained credibility calculation and category classification of information sources, making it impossible for the system to stably provide users with highly credible information in a multilingual environment.

[0057] (4) Lack of a system-level adaptive mechanism for users' emotional states. Even if existing systems use sentiment analysis algorithms, they are mostly limited to making simple tone adjustments to the dialogue content, without dynamically controlling system-level parameters such as "selection and presentation of translation data", "information source recommendation strategy" and "details of response content". Therefore, they cannot effectively alleviate the psychological burden and information comprehension burden on users in high-pressure scenarios such as disasters.

[0058] (5) Optimization of computational resource utilization and response latency is difficult. Due to the lack of a unified prompt generation mechanism and context reuse strategy, the system repeatedly calls large-scale generative artificial intelligence models between different tasks and does not perform structured caching of intermediate results such as "translation results", "list of trusted sources", and "context information", resulting in redundant operations in network communication, model inference and data access, making it difficult for the overall computational efficiency and response speed to meet the needs of a large number of users accessing the system simultaneously.

[0059] The technical problem to be solved by this invention is to provide a technical solution at the information processing system level. By uniformly designing prompt statements, data structures, and control flows for generative artificial intelligence models on the server side, information capture, machine translation, information source credibility assessment and classification, multi-turn question answering, and adaptive presentation based on emotional state can work collaboratively under the same computing architecture. This improves the data processing efficiency of computer systems in multilingual disaster information service scenarios, the controllability and stability of generative artificial intelligence model invocation, and the user interaction experience.

[0060] The specific processing performed by the specific processing unit 290 of the data processing apparatus 12 in Embodiment 1 is achieved by the following means.

[0061] In this invention, the server includes: a device for uniformly generating structured information data for use by various processing devices based on disaster-related information and life-related information obtained from an information transmission network; a device for generating control information to instruct machine translation processing devices to perform multilingual translation processing and to standardize the storage and indexing of translation results; a device for generating control information to instruct information evaluation processing devices to calculate the credibility of information sources based on evaluation benchmark data and to classify information sources according to category information; a device for generating prompt statements to drive generative artificial intelligence models to perform question parsing and answer generation based on question data sent by user terminals and the structured information data; a device for generating prompt statements to instruct generative artificial intelligence models to perform emotional state analysis based on input data from users as sentiment analysis objects and to dynamically adjust the display data presentation method and information source recommendation strategy based on the analysis results; and a device for transmitting the prompt statements, parsing object data, and evaluation benchmark data between the server and the generative artificial intelligence model through an interface processing device and receiving the parsing result data and generated text data output by the generative artificial intelligence model. This enables the formation of an end-to-end integrated computing process within the server, encompassing information capture, translation, trusted source assessment and classification, multi-turn question answering, and emotion-adaptive presentation. By structurally managing and distinguishing tasks in the invocation of generative artificial intelligence models, redundant data processing and model invocation times are reduced, improving overall data processing efficiency and response speed. Simultaneously, fine-grained computer system-level optimizations are achieved in information credibility control and user emotion adaptation, thereby improving the performance and reliability of computer technology in multilingual disaster information and life information service scenarios.

[0062] "System" refers to an overall technical solution consisting of at least one server, communication processing device, storage device, and interface processing device that interacts with generative artificial intelligence models, and which realizes the functions of information acquisition, processing, and output by executing program instructions.

[0063] A "server" is a computer device equipped with a processor and memory, capable of running programs for information acquisition, translation, evaluation, question answering, and sentiment analysis, and interacting with external devices and generative artificial intelligence models via a network.

[0064] "Communication processing device" refers to a hardware or software functional module used to transmit and receive data, parse protocols, and manage connections between a server and an information transmission network, thereby enabling the acquisition of data from external information sources and the output of display data to user terminals.

[0065] "Information transmission network" refers to the communication infrastructure used to transmit data between servers, external information sources and user terminals, including but not limited to the Internet, local area networks and mobile communication networks.

[0066] "Disaster-related information" refers to various data content related to the occurrence, prediction, early warning, response and recovery of natural or man-made disasters, including information such as event reports, early warning levels, evacuation instructions and response guidelines.

[0067] "Life-related information" refers to various types of data content related to residence, transportation, medical care, education, administrative procedures, and daily life services, used to support users' daily lives in a specific region.

[0068] "Information source" refers to various data providers or data providers that provide disaster-related and life-related information, including websites, databases, online service interfaces, and other data providers accessible through information transmission networks.

[0069] "Information data" refers to a collection of digital data obtained from information sources, existing in a structured or unstructured form, including text data, metadata, and content-related identification information.

[0070] "Machine translation processing device" refers to a computing device or software module used to perform automatic translation processing on input text according to control information, converting source language text into target language text, and may include translation models based on statistics or neural networks.

[0071] "Translation data" refers to text data generated by machine translation processing devices based on information data, expressed in one or more target languages, including the translation results of titles, body text, and related descriptions.

[0072] "Foreign language user" refers to a user whose native language or primary language is the target language and who needs to translate local language information from a specific region into the target language for comprehension.

[0073] "Information evaluation and processing device" refers to a computing device or software module used to receive attribute data and content data from information sources and perform credibility calculations, quality assessments, and classification decisions according to predetermined evaluation criteria.

[0074] "Attribute data" refers to various metadata used to describe the characteristics of information sources, including information source type, operating entity attributes, update frequency, historical performance, and access statistics.

[0075] "Content data" refers to the specific information content itself obtained from the information source, including text, titles, summaries, and other digital content related to the information topic.

[0076] "Evaluation benchmark data" refers to a set of data including rules, weights, thresholds, and judgment conditions used to evaluate the credibility and quality of information sources. It may include multi-dimensional standards based on source type, update frequency, and historical accuracy.

[0077] "Credibility" refers to the level of trust obtained after quantitatively evaluating an information source based on benchmark data. It is used to represent the comprehensive level of an information source in terms of accuracy, authority, and reliability.

[0078] "Categorization" refers to the process and result of classifying information sources into one or more predefined categories (such as disaster information, health information, education information, etc.) based on their content theme, purpose, or target audience.

[0079] "Classification information" refers to the identification data used to indicate the category to which an information source belongs, including the category name, category code, and descriptive information related to the category hierarchy.

[0080] "Consultation processing device" refers to a computing device or software module used to receive user question data, combine it with translation data and relevant data from classified information sources, generate parsed object data and construct prompt statements to drive a generative artificial intelligence model to generate answer data.

[0081] "User terminal" refers to a terminal device operated by a user for sending query data, receiving display data, and interacting with a server, including but not limited to smartphones, tablet computers, and personal computers.

[0082] "Question data" refers to natural language question text or its structured representation entered by the user on the user's terminal and sent to the server via the network, used to request the system to generate an answer.

[0083] "Analysis object data" refers to the set of input data obtained by the server after integrating question data, translation data, information data, and relevant information from classified information sources, which is used for analysis and generation by generative artificial intelligence models.

[0084] "Generative artificial intelligence models" refer to artificial intelligence models trained based on deep learning technology that can automatically generate output results including natural language text based on input prompts and parsed object data, including large-scale language models.

[0085] "Prompt statements" refer to control text or structured instructions generated by a server and sent to a generative artificial intelligence model to instruct the model to perform specific tasks (including translation, information evaluation, question-and-answer generation, and sentiment analysis).

[0086] "Response data" refers to text data generated by generative artificial intelligence models based on prompts and parsed object data, and expressed in natural language in response to question data.

[0087] "Display data" refers to the output data extracted by the server from the response data and translation data, which has been formatted and adapted to the display needs of the user's terminal, including text content and its layout information.

[0088] "Emotional state" refers to the user's current psychological or emotional tendency determined by a generative artificial intelligence model based on user input data, such as tension, anxiety, calmness, or confusion.

[0089] "Emotional state analysis" refers to the process of using generative artificial intelligence models to analyze user input data and infer the user's emotional category and intensity.

[0090] "Interface processing device" refers to a communication interface module or software component used for data packaging, protocol conversion, request sending, and response receiving between a server and an external generative artificial intelligence model.

[0091] "Analysis result data" refers to the structured or semi-structured information output by generative artificial intelligence models, obtained by analyzing the input analysis object data, including sentiment judgment results, tags, and intermediate inference results.

[0092] "Generated text data" refers to natural language text automatically generated by generative artificial intelligence models based on prompts, including translation results, answers, summaries, and prompts.

[0093] The embodiments of the present invention will be described in conjunction with the system structure defined in notes 1 to 3. In the following description, the subject is limited to "server", "terminal", and "user".

[0094] I. Overall System Composition The server is equipped with a multi-core central processing unit, main memory, at least one non-volatile storage device, and a network interface controller. The server runs multiple application modules on an operating system (such as a Unix-like operating system), including: an information acquisition module, a translation management module, an information source evaluation module, a question-and-answer generation module, a sentiment analysis module, a prompt statement generation module, an interface communication module, and a data management module.

[0095] The server connects to external information transmission networks via a network interface, including public packet-switched networks and wireless access networks. The server also communicates with external generative artificial intelligence model services and machine translation services via an interface communication module. The generative artificial intelligence model employs a large-scale language model based on the Transformer architecture, and the machine translation service uses a neural machine translation model. The terminal is a smartphone, tablet, or personal computer running a web browser or local applications. Users access multilingual disaster and lifestyle information services provided by the server through their terminals.

[0096] II. Program Modules and Data Structures 1. Information Storage and Indexing Structure The server establishes a relational or key-value database on non-volatile storage for information data, translation data, and metadata from the information source. The server organizes data fields as follows: The server stores the following fields in the article data table: information source identifier, original language code, original title, original text, publication time, and topic tags. In the translation data table, the server stores the target language title, target language text, translation engine identifier, translation time, and quality score for each article and each target language. In the information source metadata table, the server stores the unique information source identifier, source type, operating entity category, update frequency, historical error rate, credibility score, and a category tag array.

[0097] Through the above separation design, the server can attach multiple language versions to the same article record, avoiding redundant storage of multilingual data, facilitating indexing and caching, and thus reducing access latency.

[0098] 2. Generative Artificial Intelligence Model Structure and Learning Methods When the server interacts with external generative AI model services, the model used employs a multi-layer Transformer encoder-decoder or autoregressive decoder architecture. During the training phase, the model utilizes a large amount of multilingual text corpus to perform supervised learning and self-supervised pre-training. Model parameters include multi-head self-attention weight matrices, feedforward network weight matrices, and positional encoding vectors.

[0099] The server describes the model training process in the following way: During the training phase, the server uses the cross-entropy loss function as the error function to measure the difference between the model's output distribution and the true word labels on the next word prediction task. The server updates the weight parameters using stochastic gradient descent or a variant optimization algorithm (such as the Adam optimization algorithm). During the training data preparation phase, the server performs data augmentation operations on the input text, such as word segmentation, sub-word encoding, and mask randomization, to improve the model's robustness in noisy environments. During the inference phase, the server fixes the above parameters and generates the output sequence using beam search or temperature sampling strategies.

[0100] Through the aforementioned clearly defined network architecture, loss function, and optimization algorithm settings, the server enables the generative artificial intelligence model to exhibit stable behavior in multi-task scenarios (translation assistance, information source evaluation, question answering, sentiment analysis), facilitating the design of prompt statements and output control.

[0101] III. Specific Implementation Forms of Server-Side Programs 1. Information Acquisition and Preprocessing Forms In the information acquisition module, the server uses a network communication library to send HTTP requests to pre-registered information source URLs. After receiving the response, the server uses a structured parsing library to parse the RSS or HTML content. The server extracts standard fields for each piece of information and performs text encoding standardization, special character cleanup, and sentence boundary annotation before writing it into the article data table.

[0102] The server determines whether information is new by setting unique constraints and hash fields in the database, thereby reducing duplicate storage and subsequent repeated translation calls. This specific data deduplication and field standardization process reduces invalid calls to the generative artificial intelligence model from the backend, directly resulting in reduced communication load and improved overall processing speed.

[0103] 2. Multilingual translation management model In the translation management module, the server reads untranslated original text entries from the article data table in batches. The server constructs translation tasks according to the target language list. For each task, the server generates control data instructing the machine translation processing unit to perform the translation, including source language code, target language code, text content, and constraint parameters (such as terminology preservation rules). The server sends a request to the machine translation service and receives the translation results.

[0104] When saving translation data, the server attaches a language code, translation engine identifier, and timestamp to each version of the translation. To save computational overhead on the generative AI model, the server will prioritize using this translation data directly, rather than having the generative AI model re-perform the translation. This "pre-translation + result caching" structure significantly reduces the number of model inferences and network round trips in scenarios with multiple question-and-answer sessions and multi-terminal access, thereby improving overall system performance.

[0105] In another implementation, the server can correct the machine translation results using a generative artificial intelligence model. The server constructs prompts and provides the original text and the machine translation result to the generative AI model, requesting it to only locally correct obviously unnatural or erroneous content in the machine translation, rather than rewriting the entire sentence. This "post-editing" strategy, combined with prompt constraints, can improve translation quality while maintaining speed.

[0106] 3. Information Source Credibility Assessment and Classification In the information source evaluation module, the server first retrieves the attribute data and sampled content data of candidate information sources from the information source metadata table. For each information source, the server constructs a parsed object text, concatenating information such as source type, historical update time, whether it belongs to a public institution, and the objectivity of past content samples into a descriptive text.

[0107] In the prompt statement generation module, the server generates structured prompt statements for the information source evaluation task. For example: Example of a system prompt statement: "You are a website evaluation assistant familiar with multi-source news and disaster prevention information. Your task is to determine the reliability of a given website based on its descriptive text, provide appropriate category tags (such as 'disaster information,' 'health,' 'education,' 'life guide,' etc.), and return a credibility score." Example of a user prompt statement: Please read the website instructions below and complete the following tasks: 1. Summarize the main purpose of this website in one sentence.

[0108] 2. Determine whether the website can serve as a reliable source of disaster information for foreigners living in a specific country, and answer 'yes' or 'no'.

[0109] 3. Assign 1 to 3 category tags to the website (e.g., disaster information, health, education, lifestyle guides, government agencies, media, etc.).

[0110] 4. Give a credibility rating of 1 to 5, where 5 indicates very reliable.

[0111] The website description is as follows: "This website is operated by a local public agency and is dedicated to publishing disaster warnings and evacuation information for the region, including earthquakes, typhoons, and floods. The information comes from official disaster prevention departments and is updated daily." The server combines the system prompts and user prompts into a message sequence and inputs it into the generative artificial intelligence model. Internally, the model uses a multi-head attention layer to calculate the weights of the relationships between keywords such as "operating entity," "update frequency," and "official disaster prevention department" and the evaluation task instructions, thereby outputting structured text. After receiving the results, the server parses out the category label array and credibility score, and writes them back to the information source metadata table.

[0112] Through the specific prompt statement design described above, the server can impose task-specific constraints on the same model, enabling the model to adopt a unified evaluation standard across different information sources. This structured prompt control and result parsing process transforms the assessment of information source credibility from subjective human judgment into a repeatable internal computer judgment process, resulting in improved consistency and increased automation.

[0113] IV. Question and Answer Generation and Emotion-Adaptive Presentation Format 1. Context-based question-answering generation Users input questions in text format on the terminal, which then sends the question data and interface language configuration to the server via the network. The server, in its data management module, records the questions in a question-and-answer log table and retrieves relevant translation data and trusted information source entries based on the question text. The server then organizes this content into parsed object data.

[0114] The server will provide "user question," "existing translated news summaries," and "summary of trusted information sources" as reference information in the prompts for the question-and-answer task. For example: Example of a system prompt statement: "You are a smart assistant to help foreigners living in a specific country. You need to answer questions related to local news, disaster prevention information, and daily life in concise and clear Chinese. Please prioritize answering based on the given reference information. If the reference information does not contain the answer, please state this clearly and provide general safety advice." Example of a user prompt statement: User issues: Where should I take refuge if an earthquake strikes the city? The following are some disaster prevention information websites registered in the system (for reference only): 1. Website Name: Meteorological Agency Official Website Note: Provides official observation and early warning information for earthquakes, tsunamis, typhoons, weather forecasts, etc.

[0115] 2. Website Name: Government Disaster Prevention Information Platform Note: Comprehensive disaster prevention measures and emergency information released by the government.

[0116] 3. Website Name: Public Media Disaster Prevention Special Page Note: Provides disaster prevention information and real-time reports to the public.

[0117] Based on the above information, please answer the user's question in Chinese, and try to indicate the official information channels that the user should prioritize confirming. The server feeds the aforementioned messages into a generative artificial intelligence model. The model utilizes its internal attention mechanism to assign higher weights to segments such as "earthquake," "evacuation," and "official information channels," generating targeted natural language responses. In this way, the server ensures that the model's output relies not only on statistical linguistic knowledge but also on credible references provided by the server, thereby reducing the probability of fabricated information and improving the controllability and accuracy of the responses.

[0118] 2. Emotional State Analysis and Dynamic Control In the sentiment analysis module, the server receives the user's most recent dialogue content and constructs prompts for the sentiment analysis task. The server adds explicit output format constraints to the prompts, such as requiring the model to output only three pieces of information: "emotion category," "intensity score," and "suggested tone," to reduce parsing complexity. The server uses the user's input text as the primary source of sentiment features, and the model internally extracts tone features (such as expressions like "very scared" or "don't know what to do") through multi-layer attention and feedforward networks.

[0119] After receiving the sentiment assessment results, the server performs algorithmic control over response generation and information display. For example, when the sentiment category is "anxiety" and the intensity score is high, the server will add an instruction to the question-and-answer generation prompt: "Please explain step by step when answering, and use reassuring language to avoid excessive technical jargon." At the same time, in the results output module, the server controls the terminal interface to display information in a concise mode, reducing the presentation of too many technical details at once, thereby reducing the cognitive burden on the user.

[0120] This dynamic control logic, driven by emotional state, enables the server to adapt to users in different psychological states through prompts and display strategies without changing model parameters, thus forming a technological improvement at the dialogue system level.

[0121] V. Terminal-side interaction and display form After receiving the display data returned by the server, the terminal performs interface rendering locally. The terminal adjusts the interface layout based on the language code, content type, and recommendation weight in the display data. For example, in disaster prevention mode, the terminal places disaster-related translation data at the top of the interface, lists information sources with high credibility scores in prominent positions, and provides one-click access links.

[0122] Users can change the display language on the terminal via a language selection component. When the language changes, the terminal only sends the language code and content identifier to the server. The server then directly returns the corresponding language version based on the cached translation data, without needing to re-invoke the generative AI model. This front-end and back-end collaborative caching and reuse strategy directly reduces model inference requests, lowering response time and network bandwidth consumption.

[0123] VI. Technical Effects and Causal Relationships Through meticulous design of information structure, prompts, and data flow, the server has achieved the following technical improvements: The server reduces redundant translations and duplicate evaluations through a unified data table design and unique identifier management, effectively reducing the consumption of storage and computing resources. By explicitly constructing prompts for different tasks (translation correction, trusted source evaluation, question-answer generation, sentiment analysis), the server constrains the behavior of generative AI models within a reproducible range, reducing uncertainty and irrelevant outputs.

[0124] By incorporating fine-grained attribute data and historical statistical data into the information source evaluation process, the server makes credibility calculation a process jointly driven by the model's internal attention weights and external evaluation benchmarks, thereby improving the accuracy of information source selection. Furthermore, by caching machine translation results and intermediate analysis results and reusing them in multi-turn dialogues, the server significantly reduces overall response time and increases server-side processing throughput.

[0125] The server uses the sentiment analysis results to control the granularity of information presentation and recommendation strategies, rather than simply changing the tone of the text. This achieves technical optimization of the internal behavior of the computer system, reducing the probability of misunderstandings and erroneous operations by users in high-pressure scenarios, which is reflected in a reduction in the overall system error rate and misreading rate.

[0126] VII. Other Implementation Forms and Variations In another implementation, the server can use different types of generative AI models. For example, a lightweight model with compressed parameters can be used to handle high-frequency, simple questions, while large-scale models are reserved for complex question-answering or sentiment analysis tasks. The server can also use a local vector indexing module to perform embedding representations and similarity searches on article and information source content, and then input the search results as parsed object data into the generative AI model, thereby further improving retrieval accuracy and answer relevance.

[0127] In another implementation, the server can manage prompt statement templates and parameterized variables separately, and dynamically adjust task instructions, output format constraints, and banned word lists through configuration files or databases, enabling the system to quickly adapt to different regions, regulations, and service scenarios without modifying the program code.

[0128] Through the various implementation forms and variations described above, the server, terminal, and user form a close technical collaboration in information acquisition, translation, trust source assessment, question and answer generation, and emotion adaptive presentation, thereby realizing the specific implementation of the system defined in notes 1 to 3.

[0129] use Figure 11 The processing flow is explained.

[0130] Step 1: The server reads the information source configuration and model service configuration from the configuration store.

[0131] Input: Configuration data such as a list of information source URLs, a list of target languages, generative artificial intelligence model service addresses, machine translation service addresses, and authentication keys in the configuration file or configuration database.

[0132] Based on the input, the server parses the type (RSS, HTML, API), update cycle, and corresponding region and topic tags of each information source, and loads these parsing results into a configuration structure in memory. Simultaneously, the server establishes a basic connection to the information transmission network and performs connectivity checks (such as sending heartbeat requests and verifying responses) on the generative AI model service and machine translation service.

[0133] Output: A collection of information source configuration objects in memory, service connection status flags, and updated scheduling parameters.

[0134] Step 2: The server retrieves raw information data from the information transmission network according to the scheduling parameters.

[0135] Input: A collection of information source configuration objects (including URL, source type, crawling method, and last crawl time), and the current system time.

[0136] Based on the current time and the configured update cycle, the server selects the target information source to be crawled. For RSS feeds, the server sends an HTTP GET request to retrieve XML content; for HTML feeds, the server retrieves the HTML text of the webpage; for API feeds, the server constructs an HTTP request with parameters to retrieve JSON data. The server performs a preliminary validity check on the response (status code, content length, encoding type).

[0137] Output: A list of raw response data, each entry including the source identifier, response type (XML / HTML / JSON), and raw content text.

[0138] Step 3: The server parses the raw response data and generates standardized information data.

[0139] Input: List of raw response data.

[0140] The server uses an XML parsing library to extract item nodes for RSS / XML data, and an HTML parsing library to extract title, body blocks, and time fields based on predefined CSS selectors or XPath for HTML data. For JSON data, fields are read directly by key name. The server performs encoding standardization (e.g., UTF-8), HTML tag removal, special character cleanup, and sentence segmentation on the parsed text. The server generates a unique identifier for each message (e.g., a hash based on the source URL and publication time) and compares it with existing records in the database to filter out new message entries.

[0141] Output: A standardized collection of deduplicated information data, each containing the source identifier, unique information ID, original language code, original title, original text, publication time, and topic tags.

[0142] Step 4: The server writes standardized information data into the storage system and creates an index.

[0143] Input: A standardized set of information data.

[0144] The server performs an insert operation on each piece of data in a relational database or document database, mapping fields to an article data table or document collection. The server builds an index based on the information ID, source identifier, and publication time to support subsequent fast retrieval by time, source, and topic. Simultaneously, the server marks untranslated information as "untranslated."

[0145] Output: Persistently stored article records and the updated index structure.

[0146] Step 5: The server performs multilingual translation processing on the untranslated information.

[0147] Input: A list of article records in the database that are in the "untranslated" state, and a list of target languages.

[0148] The server generates a translation task for each record and each target language combination, using the original title and body text as the text to be translated, along with source language code and target language code. The server invokes a machine translation service (e.g., a neural machine translation API), sending a request containing the text to be translated. The server receives the translation response and extracts the target language title and body text from it. The server writes the translation results for each target language into the translation data table and updates the translation status of the article record.

[0149] Output: A collection of translation data records (including language code, translation title, translation text, and translation time), and article records with the update status "translated".

[0150] Step 6: The server assesses and categorizes the credibility of information sources.

[0151] Input: Information source metadata (source type, operating entity category, update frequency, historical error rate, etc.) and sampled content data (representative information text extracted from the source).

[0152] The server concatenates the attributes and content of each information source into descriptive text, which is then used as part of the parsed object data. In the prompt generation module, the server generates system prompts and user prompts based on the information source evaluation task, explicitly writing the evaluation criteria (e.g., whether it is a public institution, whether the updates are timely, and whether the content is objective) into the prompts. The server combines the system prompts and user prompts into a message sequence and inputs it into the generative artificial intelligence model service. The model returns a natural language result text containing a usage overview, reliability status, classification labels, and credibility scores. The server parses this text, extracts structured data (label array, score values), and writes it into the information source metadata table.

[0153] Output: Updated information source metadata with credibility score and category label.

[0154] Step 7: Users select the language and information category through the terminal and send an information request.

[0155] Input: User actions on the terminal interface (language selection, category selection, filter criteria, etc.).

[0156] The terminal generates request parameters based on user selections, such as language code, category identifier (disaster, livelihood, etc.), time range, and pagination parameters. The terminal sends an information request message (e.g., an HTTP request) containing these parameters to the server over the network. The terminal then enters a waiting state locally, preparing to receive and render the display data returned by the server.

[0157] Output: A data packet sent from the terminal to the server as a request for information.

[0158] Step 8: The server retrieves and assembles multilingual display data based on user requests.

[0159] Input: Information request parameters from the terminal (language code, category, time range, etc.).

[0160] The server retrieves a list of target articles from the article data table based on category and time range, and then searches for the corresponding language code translation data in the translation data table. The server merges the article metadata (title, abstract, time, source identifier) ​​with the translated text, removing records with missing translations or insufficient credibility (e.g., records whose source credibility score is below a preset threshold). The server sorts the entries by publication time, trims them into a specified number of entries, and serializes these entries into a unified display data structure.

[0161] Output: A multilingual display dataset for the terminal, with each entry containing the translation title, translation summary, publication date, source name, source credibility, and article ID.

[0162] Step 9: The terminal receives and displays multilingual display data.

[0163] Input: A collection of multilingual display data returned by the server.

[0164] The terminal parses the returned data and creates UI elements (list items or cards) for each record, displaying the translated title, summary, and time on the screen using a font and layout adapted to the user's language. The terminal binds a click event to each record to request detailed content. During display, the terminal can set visual markers based on the source's credibility (e.g., using specific color labels for highly credible sources) to help users quickly identify information quality.

[0165] Output: A multilingual information list interface displayed on the terminal screen for users to browse and perform further operations.

[0166] Step 10: Users input questions through the terminal and send them to the server.

[0167] Input: The natural language question text entered by the user in the terminal Q&A interface and the current language setting.

[0168] The terminal encapsulates the user-input text, language code, and possible location information (such as city name) into question data and sends it over the network to the server's designated question-and-answer interface. Simultaneously, the terminal records the question locally for use in displaying the dialogue context.

[0169] Output: The query data message sent to the server.

[0170] Step 11: The server preprocesses the query data and retrieves relevant information.

[0171] Inputs: Question data (question text, language code, location, etc.) and article data, translation data, and trusted information source data from the server database.

[0172] The server first performs basic normalization on the question text (removing extra spaces and standardizing encoding), and determines the question topic (e.g., earthquake evacuation, medical care, administrative procedures, etc.) based on keywords or a simple classification model. The server retrieves records matching the topic and time window from the article data table, and extracts translated text summaries in the same language as the user from the translation data table. The server then filters the information source metadata to select a list of information sources relevant to the topic and with high credibility. Finally, the server organizes these search results into parsed object data, which serves as the context for subsequent generative artificial intelligence model calls.

[0173] Output: Parsed object data related to the user's question (including the question text, relevant translation summaries, and a list of trusted information sources).

[0174] Step 12: The server generates prompts for the question-answering task and invokes a generative artificial intelligence model.

[0175] Input: Parsed object data, predefined question-and-answer task prompt template.

[0176] The server uses prompt templates to populate system prompts and user prompts with system role descriptions (such as response style, language requirements, and whether based on reference information) and user questions and related reference information. An example of a system prompt constructed by the server is as follows: "You are a smart assistant to help foreigners living in a specific country. You need to answer questions related to local news, disaster prevention information, and daily life in concise and clear Chinese. Please prioritize answering based on the given reference information. If the reference information does not contain the answer, please state this clearly and provide general safety advice." The server-generated user prompt statement is as follows: User issues: Where should I take refuge if an earthquake strikes the city? The following are some disaster prevention information websites registered in the system (for reference only): 1. Website Name: Meteorological Agency Official Website Note: Provides official observation and early warning information for earthquakes, tsunamis, typhoons, weather forecasts, etc.

[0177] 2. Website Name: Government Disaster Prevention Information Platform Note: Comprehensive disaster prevention measures and emergency information released by the government.

[0178] 3. Website Name: Public Media Disaster Prevention Special Page Note: Provides disaster prevention information and real-time reports to the public.

[0179] Based on the above information, please answer the user's question in Chinese, and try to indicate the official information channels that the user should prioritize confirming. The server packages the aforementioned prompt and parsed object data into a request message and sends it to the generative AI model service through the interface communication module. The model performs internal attention calculations and sequence generation, returning a natural language response text.

[0180] Output: Answer data text containing the content of the answers to the user's questions.

[0181] Step 13: The server performs post-processing on the response data and returns it to the terminal.

[0182] Input: The text of the response data returned by the generative artificial intelligence model.

[0183] The server performs length checks and security filtering on the response text (e.g., detecting obvious inappropriate language or content that does not conform to preset constraints), and performs simple trimming or adds explanatory notes if necessary. The server writes the response text and the question ID together into a question-and-answer log table for subsequent tracking and improvement. The server then encapsulates the response text into a response message and sends it to the terminal over the network.

[0184] Output: Terminal-oriented question-and-answer response data (natural language response text).

[0185] Step 14: The terminal displays the answer data and allows users to continue asking follow-up questions.

[0186] Input: Question and answer response data returned by the server.

[0187] The terminal adds a new answer bubble to the dialogue interface, displays the answer text, and updates the local dialogue history. The terminal retains context for input fields, allowing users to continue asking follow-up questions based on previous answers. If the user enters a new question, the terminal sends the new question along with the dialogue history identifier to the server, so that the server can construct prompts containing the historical rounds in subsequent steps, enabling multi-turn dialogue.

[0188] Output: The question-and-answer dialogue interface displayed on the terminal, along with possible data for the next round of questions.

[0189] Step 15: The server analyzes user input to determine their emotional state and adjusts the content presentation strategy accordingly.

[0190] Input: User's most recent input text (questions, feedback), current answer data, and user's language settings.

[0191] The server reads the N most recent user messages related to the current session from the question-and-answer log table and concatenates them into sentiment analysis text. The server constructs system prompts for the sentiment task, such as: "You are a sentiment analysis assistant. Simply output the user's current emotion category (e.g., calm, anxious, fearful), emotion intensity score (1-5), and suggested response tone (e.g., reassuring, concise, detailed explanation)." The server inserts the sentiment analysis text into the user prompts and sends it to the generative AI model. The model returns the emotion category and score. Based on the returned results, the server adjusts the content presentation strategy parameters, such as reducing the information density of individual responses and increasing step-by-step guidance when the emotion score is high, and prioritizing short sentences and easily understood expressions in multilingual display data.

[0192] Output: Sentiment assessment results (emotion category, intensity score, suggested tone) and updated content presentation strategy parameters.

[0193] Step 16: The server controls subsequent responses and display behavior based on the sentiment assessment results and strategy parameters.

[0194] Input: sentiment assessment result, prompt templates for the current and subsequent question-and-answer tasks, and display strategy parameters.

[0195] When constructing the next round of question-and-answer prompts, the server incorporates suggested tone and level of expression into the system prompts (e.g., adding instructions such as "Please answer in three steps, using reassuring language and avoiding excessive jargon"). When returning displayed data to the terminal, the server controls the number of list items, their sorting, and the prominence of highly reliable information sources based on policy parameters. Through this emotion-state-driven control, the server dynamically adjusts processing paths and output structures within the computer, thereby improving the overall interaction quality and fault tolerance of the system.

[0196] Output: Adjusted question-and-answer prompts, strategy-controlled display data, and the impact on the user's next round of interaction.

[0197] Application Example 1 The process flow corresponding to the specific processing in Use Case 1 will be described below. The various parts of the system described below are implemented by the data processing device 12 and the intelligent device 14. Furthermore, the data processing device 12 is referred to as the "server" and the intelligent device 14 is referred to as the "terminal".

[0198] In existing technologies, computer systems designed to provide multilingual information to foreign language users living in specific regions often employ a "pipeline-style" modular approach: one pipeline handles translation using fixed rules or traditional machine translation, another handles simple rule retrieval, another handles keyword-based recommendations, and yet another separately calls a generative artificial intelligence model for question answering. This type of system suffers from the following technical problems: First, there is a lack of a unified control mechanism between translation processing and generative AI models. Existing systems typically treat generative AI models as black boxes, directly inputting natural language only in question-and-answer scenarios. In scenarios such as translation, disaster prevention information generation, and information source evaluation, they still rely on traditional rules or single models. This results in the inability to uniformly configure processing strategies across different tasks, making it difficult for model capabilities to synergistically perform across multiple tasks. Consequently, the overall system processing flow is redundant and inefficient.

[0199] Secondly, the multilingual processing chain for time-sensitive data such as disaster-related information and news is often scattered across multiple independent services, lacking end-to-end integrated control encompassing "acquisition → translation → filtering → personalized selection → real-time push." ​​Especially in disaster prevention scenarios, automatically generating and pushing high-concurrency, multilingual information within a limited timeframe presents a significant technical challenge to computer system architecture and task orchestration mechanisms. Traditional systems often rely on manually configured push rules or simple keyword matching, failing to guarantee the timeliness and relevance of information in complex situations.

[0200] Third, the acquisition and evaluation of "high-quality information sources" often rely on pre-maintained whitelists or static rules, lacking a computer-based approach that utilizes generative artificial intelligence models to perform unified semantic analysis and automated credibility assessment of multi-source heterogeneous information. As a result, the system struggles to adaptively update evaluation criteria as content sources change, and is unable to efficiently extract and maintain a high-credibility information set from massive amounts of online information.

[0201] Fourth, personalized recommendations and push notifications are mostly based on simple statistical or collaborative filtering algorithms, only considering user click history, and failing to incorporate the advantages of generative AI models in content understanding and expression adjustment into a unified control process. Existing systems typically generate candidate content first using traditional recommendation modules, and then generate push notification text using fixed templates. There is a lack of a technical solution that utilizes a unified "prompt control mechanism" to input user interest indicators, emotional state, and content features into a generative AI model, thereby completing personalized expression and content reorganization within the model.

[0202] Fifth, regarding users' natural language queries and their sentiment states, most systems separate sentiment analysis and question-and-answer generation into two unrelated modules: the sentiment analysis module only outputs tags for simple interface adjustments, while the question-and-answer module processes content separately. There is a lack of a linkage mechanism based on unified prompts between the two, which prevents the system from dynamically adjusting the depth, tone, and information density of the answers according to the user's current sentiment state, thus reducing the quality of interaction and user experience.

[0203] Therefore, a new computer implementation is needed to enable servers to build unified control logic around generative artificial intelligence models: through prompts with structured control information, coordinate multiple tasks such as translation instructions, disaster prevention information generation, information source evaluation, personalized selection, real-time push, question and answer generation, and emotion adaptation, thereby improving the technical capabilities of existing computer systems at the software architecture and data processing flow levels, and enhancing the real-time performance, reliability, and personalization of multilingual information services.

[0204] The specific processing performed by the specific processing unit 290 of the data processing apparatus 12 in Application Example 1 is achieved by the following means.

[0205] In this invention, the server includes: a translation instruction means for generating control information for controlling a generative artificial intelligence model and inputting the control information as prompts into the generative artificial intelligence model to execute translation instructions on the input content; a disaster prevention translation means for parsing disaster-related data obtained through a disaster-related information acquisition interface using the translation instructions for the generative artificial intelligence model and generating multilingual disaster-related data using the control information; an information source evaluation means for parsing multiple information sources obtained through an information acquisition interface using the evaluation instructions for the generative artificial intelligence model and performing credibility evaluation and classification using the control information; a multi-level translation editing means for inputting first-language data obtained through an external information provision interface into a machine translation processing module to convert it into multiple second-language data, and using the editing instructions for the generative artificial intelligence model and the control information to summarize and adjust the expression of the converted data; and a translation editing means for, when sending information to a terminal device through a communication interface, based on user attribute data and user behavior... The system includes: a historical method for selecting and determining individual information to be sent from multilingual news data and multilingual disaster-related data; an information distribution method for automatically sending the data to be sent to terminal devices via a notification distribution infrastructure and presenting it in real time using the notification function on the terminal devices; a question-and-answer method for generating answer generation control information for a generative artificial intelligence model based on query data from users and information sources evaluated by the information source evaluation method, and inputting the answer generation control information as prompt statements into the generative artificial intelligence model to generate answer data; an information editing method for parsing key user interest information and user behavior history to generate interest and concern indicators for each user, and organizing and sorting candidate prompt data based on the interest and concern indicators; and an individual recommendation generation method for inputting the interest and concern indicators and the candidate prompt data as recommendation generation control information for the generative artificial intelligence model to generate guidance data for each user. This allows for the formation of a unified control framework centered on generative artificial intelligence models within the server. By using structured prompts, various tasks such as translation, disaster prevention information generation, information source evaluation, personalized selection, real-time push, question-and-answer generation, and emotion-adaptive recommendation can be collaboratively orchestrated. This reduces the data conversion and control overhead between multiple independent processing pipelines, improves the processing efficiency, response speed, and resource utilization of the computer system in multilingual information processing, real-time disaster prevention information distribution, and personalized services, and achieves substantial improvements to existing computer technology.

[0206] "Generative artificial intelligence models" refer to data processing models that utilize artificial intelligence algorithms such as deep learning, and are trained on large-scale data to automatically generate text information or other forms of output based on input prompts.

[0207] "Control information" refers to structured data used to instruct generative artificial intelligence models to perform specific tasks, including task type, constraints, target language, style requirements, etc., which are used to form prompt statements and serve as part of the model's input.

[0208] "Prompt statements" refer to natural language or structured instruction text composed of control information and related content data, which are input into generative artificial intelligence models to guide them in generating the desired output.

[0209] "Translation instruction means" refers to a functional module that runs on a server and generates translation-related control information for a generative artificial intelligence model. This control information, along with the content to be translated, is then input into the generative artificial intelligence model to perform translation processing.

[0210] "Disaster prevention translation methods" refers to a functional module that runs on a server, which acquires disaster-related information and uses translation-related control information to instruct a generative artificial intelligence model to perform multilingual conversion of the information, thereby generating multilingual disaster prevention information.

[0211] "Disaster-related information acquisition interface" refers to the software and hardware interface used to receive data related to natural disasters or emergencies from external information sources, including network communication ports, application programming interfaces and corresponding protocol processing units.

[0212] "Information Acquisition Interface" refers to the communication and protocol processing module used to automatically acquire raw data such as news, announcements, and guidelines from multiple external information sources.

[0213] "Information source evaluation method" refers to a functional module that runs on a server and uses evaluation instructions and control information for generative artificial intelligence models to perform credibility assessment and classification processing on multiple information sources obtained through the information acquisition interface.

[0214] "Credibility evaluation" refers to the process of quantifying or classifying the reliability of various information sources based on information such as source attributes, content consistency, and publication frequency.

[0215] "Multi-level translation and editing methods" refers to a functional module that includes at least two stages: machine translation processing and generative artificial intelligence model editing processing. It is used to first convert first language data into one or more second language data, and then use editing-related control information to instruct the generative artificial intelligence model to summarize, polish and adjust the expression of the conversion results.

[0216] The "machine translation processing module" refers to a program component that uses statistical or neural network methods to automatically convert text into language, and its output can be further edited or optimized by a generative artificial intelligence model.

[0217] "External information provider interface" refers to the communication interface and its protocol processing logic used to receive raw language information from external services or databases.

[0218] "Individual information selection method" refers to a functional module that runs on the server and is used to filter and determine the set of information to be sent to a specific user based on user attribute data and user behavior history from multilingual news data and multilingual disaster-related data.

[0219] "User attribute data" refers to basic characteristic information related to users, including but not limited to language preferences, geographical region, type of terminal used, registration category, etc., which are used to select personalized information.

[0220] "User behavior history" refers to user interaction data recorded during system operation, including click records, browsing duration, query content, feedback operations, etc., which are used to model user interests and preferences.

[0221] "Notification distribution infrastructure" refers to a set of network services and communication components used to reliably deliver notification messages generated by the server to multiple terminal devices, including message queues, push gateways, and corresponding protocol adaptation modules.

[0222] "Information distribution methods" refer to functional modules that run on servers and are used to push information to be sent to terminal devices in the form of notifications through notification distribution infrastructure, and trigger the terminal-side notification display mechanism.

[0223] "Terminal device" refers to a network-connected computing device carried or used by a user, including mobile phone terminals, tablet terminals, personal computers, etc.

[0224] "Question-answering methods" refer to functional modules that run on servers, receive user query data and information sources processed by information source evaluation methods, generate control information for answer generation, and instruct generative artificial intelligence models to generate answer data.

[0225] "Control information for answer generation" refers to the specific control information used to guide generative artificial intelligence models in generating answers to user queries. It typically includes query text, relevant background information, answer style, and length constraints.

[0226] "Key interest information" refers to explicit or implicit tagging information that reflects a user's interests and preferences, such as interest categories chosen by the user or topics of interest extracted from their behavioral history.

[0227] "Interest and Attention Indicators" are a set of quantitative indicators calculated based on key interest information and user behavior history, used to characterize the degree of attention users pay to different topics or categories of content.

[0228] "Information editing methods" refers to functional modules that run on a server and are used to filter, cluster, sort, and reorganize candidate information based on interest and concern indicators, thereby generating a set of prompts for specific users.

[0229] "Candidate prompt data" refers to a collection of multiple information records that may be displayed to users, selected from various information sources and system data during the information editing process.

[0230] "Individual recommendation generation methods" refers to a functional module that runs on a server and uses interest indicators and candidate suggestion data as control information for recommendation generation to input into a generative artificial intelligence model, thereby generating guidance data or recommended content for each user.

[0231] "Control information for recommendation generation" refers to control data specifically used to guide generative artificial intelligence models in generating personalized recommendation content, including information such as user interest indicators, candidate content summaries, display scenarios, and tone requirements.

[0232] In this embodiment of the invention, the server operates as the central computing node, and the terminal operates as the user-side human-computer interaction and display node, through which the user interacts with the server. The system of this invention can be deployed on general-purpose computing hardware and standard network infrastructure, such as: the server using a rack-mounted server or cloud computing instance equipped with a multi-core central processing unit (CPU) and a graphics processing unit (GPU), and the terminal using a smartphone, tablet computer, or personal computer, etc.

[0233] In one implementation, the server runs an application stack based on an operating system (such as a Unix-like operating system). The applications can be implemented in various programming languages, such as Python, Java, or JavaScript runtime environments on the server side, relational database management systems (such as SQL-based database software) and in-memory key-value databases (such as hash table-based caching software), push notifications using general push service infrastructure (such as push gateways based on the HTTP / 2 protocol and device token mechanism), machine translation using general machine translation service interfaces (such as cloud translation services based on neural machine translation), and generative artificial intelligence models can be deployed on dedicated inference services through provided APIs.

[0234] I. System Overall Structure and Data Structure The server maintains multiple logical data tables or data sets in the storage device, including but not limited to: 1. Original Information Table: The server stores the raw data obtained through the news and disaster information interfaces in the "Raw Information Table". This table may include the following fields: information identifier (primary key), information type (news / disaster), original language code, original title text, original body text, publication time, geographic region code, source identifier, preprocessing status flag, etc. The server designs the table structure to be normalized for efficient querying and indexing.

[0235] 2. Translation Information Form: The server maintains multilingual versions for each original message, linked to the original message table via foreign keys in the "Translation Information Table". This table may include: original message identifier, target language code, translation title, translation text, translation quality score, translation timestamp, etc. The server can create a composite index for this table to support fast retrieval by language and region.

[0236] 3. Information Source Table and Information Source Scoring Table: The server records the identifiers, categories, and basic reputation scores of each information source, such as websites and service interfaces, in the "Information Source Table"; and records the credibility scores, category tags, and topic vectors calculated for specific content instances in the "Information Source Scoring Table".

[0237] 4. User table and user interest table: The server maintains information such as user identifier, account status, language preference, geographical location (coarse-grained, such as city or administrative region), and terminal type in the "User Table"; and maintains the interest index vector corresponding to each user in the "User Interest Table", such as the weight value corresponding to different topics (disaster prevention, culture, life guide, etc.).

[0238] 5. Behavior Log Table: The server records user clicks, browsing duration, search terms, and feedback behavior on the terminal, with fields including: user identifier, content identifier, operation type, timestamp, and terminal category. The server updates user interest metrics through periodic batch processing or streaming processing of this table.

[0239] 6. Recommended candidate list and notification team list: The server stores the organized content identifiers and characteristics ready for recommendation in the "Recommendation Candidate Table"; and stores the information records that will be sent to the terminal through the push infrastructure in the "Notification Team List" to buffer high-concurrency push requests.

[0240] The server uses an object-oriented or modular application structure to encapsulate each functional logic into an independent module: translation instruction module, disaster prevention translation module, information source evaluation module, multi-level translation editing module, individual information selection module, information distribution module, question and answer module, information editing module, individual recommendation generation module, etc.

[0241] II. Generative Artificial Intelligence Models and Prompt Statement Control Structures In a preferred implementation, the server uses a large-scale language model based on the Transformer architecture as the generative AI model. This model has a multi-layered self-attention network structure, such as an encoder-decoder or decoder-only structure, with a parameter scale ranging from billions to tens of billions. During inference, the server deploys the model on GPU-enabled inference instances to reduce response time.

[0242] The server structures the "prompt statement" into a composite input containing the following parts: 1. Task Description Section: Clearly indicate the current task type of the model using natural language, such as "translation", "summarization", "answering questions", "sentiment analysis", "generating recommendation copy", etc.

[0243] 2. Constraints section: This includes the target language, word count limit, style requirements (formal / friendly / concise), and audience characteristics (foreigners living in a particular country).

[0244] 3. Contextual data segment: This includes user question text, relevant information summary, candidate content titles and summaries, and textual descriptions of user interest indicators.

[0245] 4. Output format description: Instructs the model to return results as paragraphs, columns, or a specific structure.

[0246] When constructing prompts, the server maps structured data (such as interest vectors and credibility scores) into natural language descriptions or labeled sequences, thereby injecting the numerical features inside the computer into the input of the generative artificial intelligence model, so that the model can automatically take these features into account when generating output.

[0247] The server can use the following translation-related prompts as an example: "You are a translation assistant working with foreigners living in Japan. Please translate the following Japanese disaster prevention information into simplified Chinese. The language should be concise and easy to understand, and all key instructions related to evacuation should be retained. The original Japanese text is as follows: '...'." For example, the server can use the following disaster prevention instructions to generate a prompt statement: "As a disaster prevention expert, please explain the basic steps to take during an earthquake to foreigners living in Japan using simplified Chinese, based on the official Japanese materials below. List the steps in bullet points and keep the explanation within 500 words. Materials content: '...". For example, the server can use the following question-and-answer related prompt statements: "You are a guide assistant for foreigners living in Japan. Answer users' questions in Simplified Chinese based on the references provided below."

[0248] User question: "I want to learn about Japanese culture." Reference: 『..." Please describe the main characteristics of Japanese culture point by point, and at the end, provide suggestions on where to further explore it. The server can use the following recommended example to generate relevant prompt statements: "You are a personalized content assistant. Below are some Chinese content summaries related to 'disaster prevention information.' Please generate a daily reminder for users who are particularly interested in disaster prevention, using a friendly and concise tone, and keeping it under 200 characters:" Content 1: ... Content 2: ...".

[0249] III. Translation Instructions and Multi-Level Translation Editing In one implementation, the server first invokes a general machine translation service to perform a preliminary conversion of the Japanese text. The server sends the original Japanese title and body text as source text to the neural machine translation service via an HTTP request. This service performs sequence-to-sequence neural network inference on a remote computing node and outputs the target language text. The server then writes this machine translation result as the "first-stage translation result" into the translation information table.

[0250] The server then uses a generative AI model to perform a "second-stage editing" of the result. The server explicitly states in the prompt that the task is "to correct and naturalize the machine translation result," and includes the machine translation output and necessary contextual information. Internally, the generative AI model uses a self-attention mechanism to simultaneously monitor multiple locations in the original machine-translated text, rewriting any literal or awkward expressions to improve readability and semantic accuracy.

[0251] The server uses this multi-level translation structure to distribute computational tasks to two types of models: the first stage handles high-throughput basic translation, while the second stage performs high-quality polishing on a smaller scale, which helps improve overall translation quality while ensuring speed. The server controls the frequency of calls to the two stages through a load balancing strategy. When the load is high, only the first stage of translation can be used, while the second stage of editing is enabled for high-priority information (such as disaster alerts for specific regions), thus achieving a trade-off between performance and quality.

[0252] IV. Disaster Prevention Translation and Multilingual Disaster Information Generation The server incorporates specific feature tags for disaster information into its disaster prevention translation module, such as disaster type (earthquake, typhoon, flood, etc.), alert level, shelter location, and emergency contact information. When parsing the data returned by the disaster information interface, the server maps these fields to structured features and indicates in a prompt statement, either in natural language or using markup, that "The following information contains evacuation instructions; please be sure to retain and highlight these parts."

[0253] By injecting domain-specific features, the server enables generative AI models to prioritize action instructions and safety information when generating multilingual disaster prevention messages, thereby reducing misunderstandings caused by translation ambiguity. The server also assigns higher priority fields to disaster prevention messages in the database, facilitating rapid filtering and weighted sorting during subsequent information selection and delivery phases.

[0254] V. Information Source Evaluation and Credibility Calculation In the information source evaluation module, the server combines rule features with the output of a generative artificial intelligence model to evaluate credibility. The server first assigns a basic score based on whether the source domain belongs to a predefined "basic trusted source set" (such as government agencies, public institutions, and large media outlets), and then calculates statistical features based on historical stability (false reporting rate, update frequency, and link reachability).

[0255] The server also constructs evaluation prompts for the generative artificial intelligence model, such as: "You are an information source evaluation assistant. Below is a text from a website. Please determine whether the content is highly reliable, based solely on the expression and information structure within the content, and provide a brief reason: '...'." The server parses the model's output credibility suggestions (e.g., "highly credible," "caution required," "unreliable") into discrete labels, and then maps them to numerical scores using rules. The server uses methods such as linear weighting or logistic regression to integrate content features, source features, and model suggestions into a unified credibility score. This score is written to an information source scoring table, and in recommendation and question-answering processes, data is selected only from the set of scores above a threshold, thereby improving the overall reliability of the system's output.

[0256] VI. Question and Answer Processing and Emotional Adaptation In the question-answering module, the server first uses a retrieval component (such as a full-text search engine based on an inverted index) to select several pieces of content with high semantic similarity to the user's question from a high-quality information set. When calculating similarity, the server can combine statistical features based on the bag-of-words model and semantic similarity based on vector embedding, projecting the topic vectors of the candidate documents and the vectors of the user's question into the same vector space for comparison.

[0257] The server then constructs question-and-answer prompts, merging the user's question with summaries of several candidate data points before inputting them into the generative artificial intelligence model. In one implementation, the server can control the model output to adopt a "point-by-point structure," making the generated results easier for the user to understand by adding instructions such as "Please answer using a numbered list of points" to the prompts.

[0258] In terms of emotion adaptation, the server analyzes features such as word choice, punctuation, and explicit emotional vocabulary in the user's input text, and uses a lightweight emotion classification model (such as a bidirectional LSTM or a small Transformer classifier) ​​to make an initial judgment on the emotion. The server then embeds the judgment result into a prompt statement, such as "The user is currently in an anxious state. Please answer in a reassuring and clear tone," thereby guiding the generative artificial intelligence model to adjust the tone and information density to achieve a more appropriate response in emergency scenarios (such as disasters).

[0259] VII. Individual Information Selection and Recommendation Generation In individual information selection modules, the server performs ranking using user interest metrics and content feature vectors. Each piece of content is represented as a topic vector and metadata features (publication time, region, type), and the user interest metrics are represented as weight vectors in the same vector space. A relevance score is then calculated using the vector dot product, and the final ranking score is obtained by combining a time decay factor and a content credibility score.

[0260] In the recommendation generation module, the server constructs prompts for top-scoring content, guiding the generative AI model to generate personalized recommendation text for that user. For example: "You are a personalized recommendation assistant. The user shows high interest in 'Disaster Prevention Information' and 'Japan Living Guide'. Please generate a recommendation description for this user based on the following three content summaries, explaining how each piece of content actually helps the user. The language should be simplified Chinese, and the description should be within 300 characters. Content summary: ..."

[0261] In this way, the server not only selects the content itself, but also uses generative artificial intelligence models to automatically reorganize and interpret the content, so that the recommendation results convey more structured information within a limited number of words, reducing the user's understanding cost.

[0262] VIII. Terminal Display and User Interaction In this embodiment of the invention, the terminal communicates with the server through network libraries and notification interfaces provided by the mobile or desktop operating system. Applications running on the terminal receive push messages from the server and display them using the system notification bar, pop-up prompts, or the application's in-app message center. After the user clicks on a notification, the terminal requests detailed content from the server and presents it within the application using a multilingual interface.

[0263] The terminal receives natural language questions input by the user in the question-and-answer interface and sends the question text, along with the user's identifier, language preference, and optional location information, to the server. After receiving the response data from the server, the terminal displays it in a dialogue format, allowing the user to continue adding questions and creating multiple rounds of conversation. Simultaneously, the terminal uploads the user's feedback on the response (e.g., "helpful / not helpful") to the server for optimization of subsequent model prompt templates and recommendation strategies.

[0264] IX. Technical Effects and Improvements in Computer Technology The server employs a unified framework of "prompt statements + control information" to drive generative AI models to perform various tasks such as translation, disaster prevention information generation, question answering, and recommendation. This avoids the problems of traditional systems where each business module interacts with the model independently and lacks a unified strategy. By constructing prompt statements in a structured manner, the server integrates multi-dimensional features such as task type, quality requirements, user interests, and emotional state into the model input. This allows the generative AI model to consider multiple constraints simultaneously during a single inference process, thereby reducing the number of repeated inferences and improving overall processing efficiency.

[0265] The server improves translation quality while maintaining high throughput through a multi-level translation structure (machine translation + generative AI polishing). Furthermore, by enabling a second-stage high-quality translation for high-priority disaster prevention information, the risk of mistranslation can be reduced in critical scenarios. By storing translation results of common expressions and recommended templates in a cache, the server can further reduce the frequency of calls to remote inference services and lower network communication load.

[0266] The server transforms internal numerical features such as interest metrics and credibility scores into natural language descriptions in the prompt statements, guiding the generative AI model to filter and interpret content. This allows the model to implicitly perform multi-objective optimization (balancing relevance, credibility, and expression style) during content generation—a technical effect difficult to achieve with traditional fixed-template systems. Furthermore, because the server can version manage and A / B test the prompt statement templates, system performance can be continuously optimized without altering model parameters.

[0267] Through the modular structure and data flow design described above, the server enables the invention to go beyond simply automating business processes. Instead, it improves the task orchestration, data representation, and model invocation strategies within the computer system by focusing on the coupling method between generative artificial intelligence models and traditional computing modules. This results in a comprehensive improvement in processing speed, accuracy, and resource utilization in the fields of multilingual information processing, disaster prevention information distribution, and personalized services.

[0268] use Figure 12 The processing flow is explained.

[0269] Step 1: The server receives raw data from an external information retrieval interface.

[0270] Input: Raw response data (usually JSON or XML text stream) from the news and disaster information interfaces.

[0271] The server parses network protocol layer data packets, extracts the HTTP response body, deserializes JSON / XML into an internal object structure, reads fields one by one (title, body, time, region, type, etc.), and maps these fields to the corresponding columns in the "original information table".

[0272] Output: A collection of raw information records written to the database. Each record contains an information identifier, information type, original language code, original title, original body text, timestamp, region code, and source identifier.

[0273] Step 2: The server preprocesses and standardizes the format of the raw data.

[0274] Input: The set of original information records generated in step 1.

[0275] The server reads the raw text field, performs data cleaning operations such as segmentation, removal of HTML tags, deletion of control characters, and standardization of punctuation, and checks the text length and character encoding; at the same time, the server adds preliminary tags based on the content category (news / disaster).

[0276] Output: Updated original information records, with text fields being cleaned and standardized strings, and the preprocessing status marked as "preprocessed".

[0277] Step 3: The server performs the first stage of machine translation processing.

[0278] Input: Preprocessed but untranslated raw information records, and a list of target languages ​​(e.g., en, zh-CN, etc.).

[0279] The server selects a number of records in batches, combines the original title and body text of each record into source text, constructs a translation request, calls the machine translation processing module interface, and packages and sends the source text and target language code together; the server receives the multilingual text results returned by the translation service.

[0280] Output: Preliminary translation results in multiple target languages ​​for each piece of original information, and these results are written into the translation title and translation text fields of the "Translation Information Table", while recording the translation time and initial quality mark.

[0281] Step 4: The server utilizes a generative artificial intelligence model to perform the second stage of translation polishing and quality improvement.

[0282] Input: The machine translation results generated in step 3, as well as the corresponding original text and task metadata (content type, domain tags, etc.).

[0283] The server constructs prompt statements for each target language, combines "machine translation results + original text fragments + task requirements (such as maintaining accurate disaster prevention instructions)" as model input, and generates translation-related control information; the server inputs the prompt statements into the generative artificial intelligence model and receives the polished text output by the model.

[0284] The server compares the content before and after the polishing, calculates the simple similarity and length change ratio to check for any abnormal or significant rewriting. If any abnormality is found, the system reverts to the results of the first stage.

[0285] Output: The improved translation title and body text are written to the "Translation Information Table" (either overwritten or newly added), and the translation quality is updated to "High Quality".

[0286] Step 5: The server performs feature extraction of disaster-related information and generation of multilingual disaster prevention information.

[0287] Input: Disaster-type information records from the original information table and their multilingual translations.

[0288] The server extracts structured features and generates feature vectors based on disaster fields (disaster type, alert level, region, time, etc.), and embeds these features in natural language in the prompt statements (e.g., "This is an earthquake warning, evacuation action should be emphasized"). The server inputs the featured prompt statements and the original Japanese text into the generative artificial intelligence model to request the generation of multilingual disaster prevention instructions "specifically for foreign language users living in a specific region".

[0289] Output: Multilingual disaster prevention instructions text containing action steps, evacuation advice, and precautions, stored in a translation information table or a separate disaster prevention instructions table, and marked as "Disaster Prevention Specialized Content".

[0290] Step 6: The server performs information source evaluation and credibility scoring.

[0291] Input: Various web pages or interface content captured through the information acquisition interface, as well as their source identifiers and historical statistical characteristics.

[0292] The server calculates a base score (whether it is a pre-registered trusted source), historical error rate, update frequency, and other statistical characteristics for each information source, and puts the content fragments into evaluation prompts, such as "Please analyze the professionalism and reliability of this content". The server inputs the prompts into a generative artificial intelligence model to obtain text suggestions such as "highly trustworthy / cautious / untrustworthy", and then maps the suggestions into numerical scores.

[0293] The server uses a weighted formula to combine the source attribute score, historical feature score, and model suggestion score into a total credibility score.

[0294] Output: Credibility score and category label written into the "Information Source Rating Table" for subsequent filtering and recommendation.

[0295] Step 7: The server constructs a set of superior information based on the evaluation results of the information sources.

[0296] Input: Credibility score from the information source rating table and records of original and translated information.

[0297] The server filters information sources and their content with scores higher than the threshold, marks them as "good information," and writes the relevant records into a separate "good information table" or adds the good information tag. The server clusters and indexes the good information by topic (such as disaster prevention, life guide, Japanese culture) to generate topic vectors and keyword indexes.

[0298] Output: A well-organized collection of information by topic and language, providing high-quality material for the question-and-answer and recommendation modules.

[0299] Step 8: Users enter natural language queries on the terminal and send them to the server.

[0300] Input: Question text entered by the user in the terminal application, such as "I want to learn about Japanese culture." or "What should I do during an earthquake?".

[0301] The terminal encapsulates the problem text along with the user identifier, language preference, and current location (if authorized) into a request message, which is then sent to the server over the network.

[0302] Output: The query request from the terminal to the server, including the query text and user context information.

[0303] Step 9: The server performs data retrieval related to the user's query.

[0304] Input: The user query request from step 8, and the text and index data from the good information set.

[0305] The server segments and vectorizes the query text, calculates the similarity to the topic vectors of high-quality information, and selects several of the most relevant materials through a combination of full-text search engine and vector search. The server generates a short summary or directly extracts key information fragments for each material.

[0306] Output: A list of relevant materials and their summaries related to the user's question, used as contextual input for generative artificial intelligence models.

[0307] Step 10: The server constructs question-and-answer prompts and calls a generative artificial intelligence model to generate the answers.

[0308] Input: User question text (step 8) and a list of relevant information summaries (step 9).

[0309] The server combines "user question + data summary + answer format constraints (e.g., 'please answer in points')" into a question-and-answer prompt statement according to a predefined template, and adds control information for answer generation (language, style, length limits, etc.). The prompt statement is then input into the generative artificial intelligence model. The server receives the output answer text and performs post-processing such as length truncation and sensitive content filtering.

[0310] Output: A structured response text for the user and an optional list of related resource links.

[0311] Step 11: The server sends the generated answer to the terminal, where it is displayed.

[0312] Input: The answer text and related links generated in step 10.

[0313] The server encapsulates the answer into a response message and sends it to the terminal; the terminal parses the response content, displays the answer text in the form of a dialog bubble on the Q&A interface, and displays related links in a list or button.

[0314] Output: The answer and clickable extended reading items are displayed on the terminal interface, allowing users to continue entering follow-up questions to form multiple rounds of dialogue.

[0315] Step 12: The server updates user behavior history and interest metrics.

[0316] Input: User behavior data uploaded by the terminal, such as which notification was clicked, reading time, evaluation of the answer, and selected interest tags.

[0317] The server writes behavior records to the behavior log table, and through batch processing or real-time streaming computing, it counts the frequency of user interaction and dwell time on each topic, and updates the weight of each topic in the interest vector; the server writes the new interest indicators back to the user interest table for subsequent personalized selection.

[0318] Output: Updated user interest metrics, used for individual information selection and recommendation generation.

[0319] Step 13: The server performs individual information selection to determine the target information to be sent to each user.

[0320] Inputs: Multilingual news and disaster prevention information records, user attribute data, and interest and concern indicators.

[0321] The server filters out content in irrelevant languages ​​or regions based on the user's language preferences and geographical region; the server calculates the relevance score between each piece of content and the user's interest vector, and combines it with factors such as publication time (time decay), content type (e.g., higher priority for disaster prevention), and information source credibility to form a comprehensive score, and selects the top few pieces as candidate messages to be sent according to the scores.

[0322] Output: A list of candidate messages generated for each user or user group, written to the notification team list.

[0323] Step 14: The server constructs recommendation and notification text and calls a generative artificial intelligence model to personalize the message.

[0324] Input: The list of candidate messages to send in step 13, and user interest metrics.

[0325] The server constructs recommendation prompts for each user or user group, combining "candidate content summary + user interest characteristics + style requirements in the prompt (such as 'friendly and concise')" into input, instructing the generative AI model to generate recommendation descriptions or notification content, such as "Today we have prepared three important pieces of information related to disaster prevention for you..."; the server receives the personalized copy output by the model and associates it with the content ID.

[0326] Output: A personalized notification title and body text for a specific user, ready to be sent to the terminal.

[0327] Step 15: The server pushes information to the terminal through the notification distribution infrastructure, and the terminal displays and interacts with it.

[0328] Input: Personalized notification content and the corresponding user terminal identifier (device token, etc.).

[0329] The server encapsulates the notification content into a push message, calls the notification distribution infrastructure interface, and distributes the message to the corresponding terminal. After receiving the push, the terminal displays the title and part of the body of the message in the system notification bar or pop-up window. When the user clicks on the notification, the terminal requests the server through the API to obtain the complete content and displays it within the application. Users can perform actions on the notification (view details, ignore, provide feedback), and these actions are returned to the server as behavioral data.

[0330] Output: Personalized notifications displayed in real time on the terminal, along with associated subsequent interaction events, provide a data source for server behavior analysis and interest updates.

[0331] Alternatively, an emotion engine for inferring user emotions can be combined. That is, the specific processing unit 290 can also use the emotion-specific model 59 to infer user emotions and perform specific processing using user emotions.

[0332] Example 2 The flow of a specific process in Example 2 will be described. Each part of the system described below is implemented by the data processing device 12 and the smart device 14. The data processing device 12 will be referred to as the "server," and the smart device 14 as the "terminal."

[0333] In existing technologies, information services for non-native language users typically rely on simple machine translation or keyword matching searches. This approach presents several technical problems: First, server-side processing of massive amounts of unstructured online document data largely depends on manual rules or static crawler scripts, lacking unified structured parsing and automatic classification capabilities. This results in a lack of semantic intermediate representation between the storage and retrieval layers, making it difficult to balance retrieval efficiency and accuracy. Second, existing information retrieval systems are generally based on fixed query syntax or keyword searches, failing to perform deep semantic parsing of user-inputted prompts in natural language. This prevents the accurate extraction of multi-dimensional retrieval parameters such as geographic information, information category, and usage conditions, leading to coarse query conditions generated by the server and a significant semantic gap between the returned results and the user's actual needs. Third, although generative AI models can generate relatively natural text, in traditional architectures, these models are often simply used as question-and-answer interfaces or translation interfaces, lacking deep coupling with structured information storage, indexing structures, and information source credibility evaluation mechanisms. This fails to effectively improve the overall information organization capabilities and response quality of the system while ensuring data controllability and auditability. Fourth, for highly sensitive and time-sensitive content such as disaster prevention information and local life information for foreigners, existing systems often lack technical solutions for systematically assessing and selecting the credibility of information sources. Servers cannot automatically filter information sources and associate them with classification results based on unified credibility evaluation standards, thus increasing the risk of generating erroneous or misleading information. Fifth, from a computer technology perspective, existing systems often have loose coupling between processing modules within the server: the processes of web crawling, natural language processing, classification indexing, generative AI invocation, and user interaction lack a unified interface design centered on "prompt statements," leading to frequent data conversion between different modules, low efficiency in utilizing computing resources, and poor system scalability and maintainability.

[0334] In summary, there is an urgent need for a comprehensive technical solution that can be implemented within the server: from network information collection, natural language parsing, and classification index construction, to the unified invocation of generative artificial intelligence models based on prompt statements and the return of both structured and natural language results to the user terminal. This solution aims to improve the data processing flow, storage structure, and inference invocation mechanism at the computer technology level, thereby enhancing the system's processing efficiency, retrieval accuracy, response quality, and scalability in multilingual and multi-scenario information services.

[0335] The specific processing performed by the specific processing unit 290 of the data processing apparatus 12 in Embodiment 2 is achieved by the following means.

[0336] In this invention, the server includes: a collection unit for automatically collecting document data from multiple information sources based on input containing evaluation criteria and search terms; a parsing unit for performing natural language processing on the document data, including language discrimination, word extraction, and evaluation estimation, to generate structured parsing results and, based on the parsing results, generating prompt statements for use by generative artificial intelligence models; a classification unit for classifying the document data into multiple categories such as life-related information, tourism-related information, and food-related information, and adding category identifiers and tags to the classification results; a storage unit for storing the classification results in a storage area for retrieval and constructing an index structure based on the category identifiers and tags; and a user input containing prompt statements from a user terminal, parsing the prompt statements to extract regional information, etc. The system includes a query unit that retrieves candidate information from the storage area based on information categories and usage conditions; a response generation unit that generates prompts that instruct the generative artificial intelligence model to perform response generation processing based on the candidate information and user input, and combines the response text obtained from the generative artificial intelligence model with structured candidate information to generate response data; a translation unit that parses the input content and obtains translation results for information and disaster prevention-related information for non-native language users by instructing the generative artificial intelligence model to perform translation processing; and a credibility management unit that evaluates information sources according to credibility evaluation criteria, parses and selects information sources by using prompts that instruct the generative artificial intelligence model to parse the evaluation criteria, and associates and stores the selection results with the classification results. This allows for the formation of an end-to-end processing chain within the server, using prompt statements as a unified interface. This organically integrates network information collection, semantic parsing, structured classification, index building, generative artificial intelligence model invocation, and user interaction. It not only reduces data format conversion and redundant calculations between modules, improving the efficiency of computing resource utilization and retrieval response speed, but also enhances the reliability and controllability of information services by combining credibility evaluation with the associated storage of classification results. Thus, it achieves overall optimization of multilingual and multi-scenario information processing capabilities at the computer technology level.

[0337] A "system" refers to a whole consisting of one or more computing devices and program modules running on them, which is a technical solution used to perform functions such as information acquisition, processing, storage, retrieval and output.

[0338] A “device” refers to a functional unit implemented in a server or terminal in the form of hardware, software or a combination thereof, used to perform specific data processing or control tasks.

[0339] A "server" is a computing device that provides data processing, storage, and service response functions in a network environment, including a processor, memory, and service programs running on it.

[0340] "User terminal" refers to a computing device that allows users to input and view output results, including but not limited to mobile terminals, desktop terminals, or network-based clients.

[0341] "Information obtained from the Internet" refers to text, images, or other forms of digital data obtained from remote information sources through communication networks.

[0342] "Information source" refers to the external data provider or data storage location that provides document data, including websites, databases, information service interfaces, etc.

[0343] "Document data" refers to a collection of text-based data that can be obtained from information sources and used for subsequent parsing and processing, including web page content, articles, comments, announcements, etc.

[0344] "Evaluation criteria" refers to a set of rules, parameters, or indicators used to assess the credibility, quality, or relevance of information sources.

[0345] "Search terms" refer to one or more textual expressions used to limit the scope of online information collection or database query, including keywords, phrases, or query statements.

[0346] Natural Language Processing (NLP) refers to the automated analysis and transformation process performed on natural language text, including language discrimination, word segmentation, part-of-speech tagging, entity recognition, and sentiment analysis.

[0347] "Language discrimination" refers to the process of automatically identifying the type of natural language to which the input text belongs.

[0348] "Word extraction" refers to the process of identifying and extracting keywords, entity nouns, or other important words from text.

[0349] "Evaluation estimation" refers to the process of making quantitative inferences about attitudes, ratings, or recommendation levels of the objects involved in a text based on its content.

[0350] "Structured parsing results" refer to the processing results data obtained from natural language processing, represented in the form of fields or labels. This data has a predefined data structure, which facilitates storage and retrieval.

[0351] "Generative AI models" refer to AI models that can automatically generate output content such as text based on input prompts, including large-scale language models based on deep learning.

[0352] "Prompt statements" refer to natural language or structured text expressions used to indicate the task type, input content, and output requirements to generative artificial intelligence models.

[0353] "Classification results" refer to the category information and associated attribute information obtained after classifying document data.

[0354] "Life-related information" refers to information categories related to everyday life scenarios such as living, shopping, medical care, and education.

[0355] "Tourism-related information" refers to information categories related to tourism scenarios such as travel, sightseeing attractions, and activity arrangements.

[0356] "Food-related information" refers to information categories related to dining venues, eating habits, and dish recommendations.

[0357] "Category identifier" refers to the identification information used to mark the category to which document data belongs. It can be the category name, category code, or other data that can distinguish the category.

[0358] "Tag information" refers to attribute information used to further describe the characteristics of document data, including but not limited to the target audience, scenarios, price range, service features, etc.

[0359] "Storage area for retrieval" refers to a logical area or data set in a storage device used to store classification results and support indexing and query operations.

[0360] "Index structure" refers to a logical data structure established to accelerate data retrieval, including inverted indexes, hash indexes, tree indexes, etc.

[0361] "User input" refers to natural language or structured data provided by users through user terminals in the form of text, speech-to-text, or other similar methods.

[0362] "Geographic information" refers to geographical location-related information extracted from user input or document data, including cities, regions, and countries.

[0363] "Information category" refers to the category type used when grouping or classifying information content, such as lifestyle, travel, food, etc.

[0364] "Usage conditions" refer to the information restrictions extracted from user input that affect retrieval or filtering, including budget, peer groups, time constraints, etc.

[0365] "Search criteria" refers to the set of parameters parsed from user input and used to select candidate information from the storage area.

[0366] "Candidate information" refers to a set of data selected from the storage area used for retrieval based on search criteria, which is used for further filtering or generating responses.

[0367] "Response data" refers to the result data generated by the server and returned to the user terminal, which includes response text and structured candidate information.

[0368] "Response text" refers to the natural language response output by a generative artificial intelligence model, which is used to display to the user.

[0369] "Translation processing" refers to the automated text transformation process that converts input content from one language to another.

[0370] "Information for non-native language users" refers to target information that is generated through translation or multilingual processing and is suitable for readers in non-native languages ​​to understand.

[0371] "Disaster prevention-related information" refers to information related to the prevention, early warning, response, and evacuation guidance of natural disasters or emergencies.

[0372] "Translation result" refers to the target language text content obtained by converting the source language into a target language through translation processing.

[0373] "Credibility" refers to the comprehensive evaluation of the reliability, authority, and accuracy of information sources or information content.

[0374] The "credibility management unit" refers to a functional unit used to evaluate, screen, and maintain credibility-related data of information sources or information content based on evaluation criteria.

[0375] A "parsing unit" refers to a functional unit in a server that performs natural language processing on document data or user input and generates structured parsing results.

[0376] A “classification unit” refers to a functional unit in a server used to classify document data according to the parsing results and attach identifiers and tags.

[0377] A "storage unit" refers to a data storage functional unit in a server used to store classification results and build indexes for retrieval.

[0378] A "query unit" refers to a functional unit in a server used to parse user input, generate search criteria, and retrieve candidate information from the storage area.

[0379] The "response generation unit" refers to a functional unit in the server used to generate prompts for generative artificial intelligence models based on candidate information and user input, and to construct response data.

[0380] A "translation unit" refers to a functional unit in a server that is used to call a generative artificial intelligence model to perform translation processing and output the translation results.

[0381] In this embodiment of the invention, the server undertakes the main data processing and control functions, while the terminal and user undertake the interaction and input / output functions, respectively. The following describes an embodiment that can realize all the functions of this invention, in conjunction with the specific composition of the hardware structure, software modules, data structure, and generative artificial intelligence model.

[0382] I. Overall Hardware and Software Composition of the System Servers can be implemented using general-purpose computing devices, such as rack-mounted servers or cloud computing instances equipped with multi-core central processing units (CPUs, such as those based on x86 or ARM architectures), graphics processing units (GPUs, such as GPU accelerator cards supporting general-purpose computing), main memory, and persistent storage devices (such as solid-state drives). The underlying software environment on which the server runs can include: operating systems (such as Unix-like operating systems), database management systems (such as relational database management systems or search engine-based database management systems), runtime environments (such as interpreted language runtime environments), and container management platforms, etc.

[0383] The server can run multiple functional modules at the application layer. These modules can collaborate through inter-process communication or microservice interfaces, including: information acquisition module, natural language parsing module, classification and labeling module, storage and indexing module, query parsing module, response generation module, translation processing module, credibility management module, and generative artificial intelligence model invocation module. The server can also integrate natural language processing libraries (such as open-source libraries for word segmentation, part-of-speech tagging, entity recognition, etc.) and deep learning frameworks (such as frameworks based on tensor operations) to support the operation of neural network models.

[0384] The terminal can be a mobile terminal or a desktop terminal. Its hardware includes a processing unit, a display, input devices (touchscreen, keyboard, microphone, etc.), and a network interface. The terminal runs an operating system (such as a mobile operating system or a desktop operating system) and browser applications or native applications. The terminal establishes a connection with the server through a secure communication protocol, sends user input to the server, and receives response data generated by the server.

[0385] Users input natural language queries or requirement descriptions through the terminal and browse the text descriptions and structured information lists returned by the server. Users do not directly access the database or model, but interact indirectly with the server's backend modules through the terminal interface.

[0386] II. Server-generated programs and modular implementation forms In this invention, the server can be deployed as a program system with multiple logical modules working together, and each module can be divided according to its function as follows: 1. Information Collection Module The server performs web data scraping through an information collection module. This module can be implemented using an HTTP request library and an HTML parsing library, or it can employ a dedicated crawling framework. During scraping, the server reads parameters such as "evaluation criteria" and "search terms" from a configuration storage unit and accesses multiple pre-configured information sources on the network one by one. The server stores the acquired web page content as document data, along with metadata (URL, scraping time, site type, etc.), in temporary storage.

[0387] 2. Natural Language Parsing Module The server uses a natural language processing module to perform language discrimination, word extraction, and rating estimation on document data. This module can use the following software components: language recognition model, word segmentation and entity recognition model, and sentiment analysis and rating model.

[0388] In this module, the server transforms document data into internally standardized structured parsing results. These results may include: language type field, location entity field, category candidate field, sentiment score, keyword list, etc. The server organizes these fields into key-value pairs, providing input for subsequent classification and indexing.

[0389] 3. Category and Tag Module The server, through its classification and tagging module, determines the category and assigns tags to document data based on the parsing results. The server can use the following methods in this module: - Rule matching method: Based on the mapping relationship between the keyword set and the predefined category, perform preliminary classification of life-related information, travel-related information, food-related information, etc. - Statistical classification method: Use a trained classification model, input the feature vectors from the parsing results into the classifier to calculate the class probability; - Generative AI-assisted approach: Construct specific prompts, input document summaries into the generative AI model, and request the model to output category names or a set of labels.

[0390] The server ultimately writes category identifier and tag information fields to each record, forming the classification result.

[0391] 4. Storage and Indexing Module The server uses storage and indexing modules to write classification results into a storage area for retrieval. This storage area can be a combination of a relational database and a full-text search engine. The server builds index structures (such as inverted indexes and B+ tree indexes) for fields like city, category, and tag to improve query speed. Through the construction of these index structures, subsequent combined searches based on geographic information, information category, and usage conditions are more time-efficient than linear scans, thus reducing CPU and disk I / O load.

[0392] 5. Query parsing module The server processes user input from user terminals through a query parsing module. This module receives user input containing prompts, performs semantic parsing on the natural language content, and extracts search criteria such as geographic information (e.g., city, region), information category (lifestyle, tourism, food, etc.), and usage conditions (e.g., suitable for families, budget constraints, time restrictions, etc.). The server can utilize the same natural language parsing technology or construct a specialized intent recognition model to map the prompts to the query parameter space.

[0393] 6. Response Generation Module The server, through the response generation module and based on the search criteria obtained from the query parsing module, accesses the storage and indexing module to retrieve a set of candidate information records. The server then sorts the candidate records (e.g., by recommendation level, time, freshness, etc.), and constructs a prompt statement for the generative AI model by combining the summary of the candidate information with the user's prompt, sending it to the generative AI model invocation module. The server then combines the response text returned by the model with the structured fields selected from the candidate records to form the response data, which is returned to the terminal.

[0394] 7. Translation Processing Module When a user requests information intended for non-native speakers or disaster prevention-related information, the server, through its translation processing module, constructs specific prompts to instruct the generative AI model to translate or rewrite the input content with constraints. The server can specify parameters such as target language, style (e.g., concise and suitable for beginners), and content filtering requirements to ensure the readability and security of the translation results. The server stores the translation results in corresponding fields and can associate them with the original text.

[0395] 8. Trust Management Module The server, through a credibility management module, assesses the credibility of information sources based on predefined evaluation criteria. These criteria may include: information source type, historical accuracy, data update time, and whether it is an official organization. The server can use statistical methods (e.g., accumulated scores based on historical error correction records), rule-based methods (e.g., whitelists, blacklists), and generative artificial intelligence models to assist in the evaluation (by prompting the model to determine the credibility level of the information source with appropriate hints), and associate the evaluation results with the classification results. During retrieval, the server prioritizes data from information sources with higher credibility, thereby reducing the probability of misinformation propagation.

[0396] III. Structure and Learning Methods of Generative Artificial Intelligence Models In this invention, the server can employ a generative artificial intelligence model based on deep learning, which can be a multi-layer self-attention neural network. The model consists of an embedding layer, multiple encoding units, and decoding units, with each layer including a multi-head self-attention sublayer and a feedforward network sublayer. The model input is a sequence of prompt statements after word segmentation and encoding, and the output is a probability distribution sequence of the target text.

[0397] During the model training phase, the server uses training data including multilingual dialogue corpora, question-and-answer corpora, and domain-specific texts to perform supervised training or fine-tuning of the model parameters. The training process can employ the cross-entropy loss function, comparing the word probability distribution output by the model with the ground truth labeled sequences, calculating gradients through backpropagation, and updating model weights using optimization algorithms (such as adaptive learning rate algorithms). The server can also use data augmentation methods, such as generating different representations of the same semantic content, to enhance model robustness.

[0398] During the deployment phase, to improve inference efficiency, techniques such as model quantization, weight pruning, and hierarchical caching can be used to reduce computational load. When the server receives a prompt statement, it converts it into vector form according to a predefined encoding method and feeds it into the model to gradually generate the output sequence. The server can set the maximum output length, temperature parameters, and sampling strategies (such as top-k or top-p sampling) to control the quality and diversity of the generated results.

[0399] IV. Technical Details in Data Structures and Processing Flow In this implementation, the server employs a unified data record structure, with record fields including: document identifier, information source identifier, original text, parsing results (language type, keywords, sentiment score, etc.), category identifier, tag information, credibility score, and timestamp. Through this structured data record, the server maps natural language processing results to the index structure one-to-one, reducing data conversion overhead between different modules.

[0400] During the query parsing phase, the server maps the user-input prompts into a search condition object composed of multiple fields, including geographic fields, category fields, tag filtering conditions, and confidence thresholds. When accessing the index, the server performs multi-condition combined queries based on these fields, rather than performing direct string matching on the full text, significantly reducing the search space and improving retrieval speed.

[0401] During the response generation phase, the server first pre-screens and sorts candidate records, then encapsulates their summary information into prompts and submits them to the generative AI model, rather than allowing the model to freely generate responses directly from the original large-scale document. Because the model's input is limited to candidate information highly relevant to the user's intent, the generated content is more focused, and the server can validate and post-process the model's output within the candidate set, thereby reducing the proportion of illusory content and improving output accuracy.

[0402] V. Prompt Statements and Specific Interaction Examples Users can enter the following natural language prompts on the terminal: "Please tell me some recommended tourist attractions in Tokyo." "This is my first time in Japan. Could you recommend some restaurants in Tokyo that are friendly to foreigners?" Please describe a few tourist attractions in Kyoto that are suitable for families with children in English. Are there any restaurants in Osaka that cater to both vegetarians and non-vegetarians? "Please tell me the basic information I need to know about living in Tokyo long-term, such as housing, transportation, and hospitals." The terminal sends the above prompt as text to the server. The server extracts the city, category, and constraints in the query parsing module. For example, for "Please tell me recommended tourist attractions in Tokyo.", the server extracts the city as "Tokyo," the category as "tourism-related information," and the constraints as empty. The server then accesses the storage and indexing module, filters out records belonging to "Tokyo" and categorized as "tourism-related information," selects several records with high ratings and high credibility as candidate information, and constructs a new model input prompt through the response generation module, for example: "You are an assistant providing travel information about Japan to foreigners. Below is information on some attractions in Tokyo. Please select 3 to 5 of the most suitable ones to recommend to the user, and explain the characteristics and reasons for recommending each attraction in simplified Chinese: ... (candidate information summary) ... User question: Please tell me about recommended sightseeing spots in Tokyo." The server sends the prompt to a generative AI model, which outputs a natural and fluent response text. The server then packages this response text with structured fields from candidate information into response data and sends it back to the terminal. The terminal displays a text description and detailed information about each attraction, including its name, tags, and address, on the interface.

[0403] VI. Explanation of Technical Effects and Causal Relationship The server achieves the following technical effects through the aforementioned specific data structures and module divisions: 1. Improved accuracy and reliability Because the server performs structured parsing, classification, and credibility assessment after information collection, and then combines these results with the output of a generative artificial intelligence model, the output content is constrained within a candidate set of credible sources and relevant categories. Therefore, under the same hardware conditions, the server-generated results have a lower error rate and a lower proportion of fictitious content compared to traditional methods that directly call text generation models.

[0404] 2. Improved processing speed and computational efficiency By leveraging an index structure and pre-constructed parsing results, the server reduces the retrieval operation from full-text scanning to index queries, significantly decreasing I / O and CPU time. Simultaneously, by pruning the model input length, it provides only candidate information summaries matching the retrieval criteria to the generative AI model, reducing sequence length and computational load in model inference, thereby achieving shorter response times and increased throughput.

[0405] 3. Improved Data Management and Scalability The server uses a unified data record structure to transform multilingual, multi-source document data into unified parsing and classification results, which are then stored in an indexable storage area. Based on this, new categories and tags can be easily added simply by adding corresponding classification rules or training new classification models, without needing to reconstruct the entire storage structure. The server can also adjust evaluation criteria and credibility algorithms as needed, thereby flexibly controlling the introduction and removal of information sources.

[0406] 4. The fundamental difference from human manual processing In this invention, the server utilizes a generative artificial intelligence model and specific prompts to implement non-human-like combined processing rules: First, a multi-dimensional structured index is established through natural language parsing and classification; second, prompts are used to uniformly inject search results and user-defined semantics into the model, enabling the model to reorganize and interpret content on a finite candidate set. This processing method does not simply replace human translation or retrieval, but rather forms a new end-to-end technical approach through the coordination of internal computer data representation and algorithmic processes, achieving deep integration between model inference and structured data access.

[0407] VII. Alternative Implementation Methods and Extended Forms The server can employ different technical approaches in different implementations, yet still fall within the scope of the technical concept of this invention. For example: 1. In the natural language parsing module, the server can use a sequence labeling-based neural network model to replace traditional word segmentation and entity recognition tools, directly encoding documents into semantic tag sequences.

[0408] 2. In the classification and labeling module, the server can use a generative artificial intelligence model as a zero-shot classifier, simply by providing the category set through prompts, and the model outputs the category to which each record belongs, without the need to train a separate discriminant model.

[0409] 3. In the storage and indexing module, the server can use a graph database or vector database to encode the parsing results into embedded vectors and use similarity retrieval to replace or supplement the traditional inverted index structure.

[0410] 4. In the translation processing module, the server can use a specially trained multilingual translation model, or add language alignment loss to a general generative artificial intelligence model, so as to further improve the translation results in terms of terminology consistency and style control.

[0411] 5. In the credibility management module, the server can introduce a time decay factor to automatically reduce the credibility score of outdated information source records and prioritize showing users the most recently updated information.

[0412] Through the above-mentioned various implementation methods and alternative technical routes, the core idea of ​​this invention is still followed between the server, terminal and user: the server uses a generative artificial intelligence model and prompt statements as a unified interface to closely integrate network data collection, natural language parsing, structured classification, credibility assessment, index retrieval and response generation, thereby achieving systematic optimization of multilingual information service capabilities at the computer technology level.

[0413] use Figure 13 The processing flow is explained.

[0414] Step 1: The server obtains configuration information and prepares to collect parameters.

[0415] The server's input consists of a set of parameters stored in a configuration file or configuration database, including evaluation criteria (reliability rules, update time thresholds, etc.), search terms (such as "Tokyo sightseeing" and "Tokyo restaurant recommendations"), and a list of target information sources (multiple site addresses). The server parses and loads this configuration data, converting the text-based configuration into internal data structures (such as key-value pairs and lists), and establishes an "information source queue" and a "retrieval task queue" in memory. The output is a set of instantiated collection task objects, each containing the target URL, associated search terms, and evaluation criterion reference information. Specific actions include: the server reading the configuration file content and parsing it into a structure; the server performing preliminary screening of information sources based on the evaluation criteria, removing those marked as low-reliability or expired, and adding the remaining information sources to the collection list.

[0416] Step 2: The server collects document data from online information sources.

[0417] The server's input is the set of data collection task objects generated in step 1. The server executes HTTP / HTTPS requests for each task object, using network communication protocols to obtain data returned by the webpage or API. The server performs basic validation on the response content (status code, content length), extracting the HTML body or JSON text from valid responses and using it as the "raw document data." Data processing includes: the server using an HTML parsing library to remove script and style tags, extracting the body and title areas; and the server recording the crawl time, response header information, and origin server identifier. The output is a set of document data records, each containing fields such as raw text, title, origin server ID, and crawl time. Specific actions include: the server looping through the URL queue, retrying failed requests a set number of times, and logging failure information.

[0418] Step 3: The server performs natural language parsing on the document data.

[0419] The server's input is the document data record output from step 2. First, the server performs language discrimination on the main text of each record, inputting the text into a language recognition model to obtain language tags (such as "ja", "en", "zh"), which are then written into the record. Subsequently, the server calls word segmentation and entity recognition tools to segment the text, tag parts of speech, and extract entities, generating entity fields such as location names, attraction names, and shop names. Next, the server uses a sentiment analysis model or rating estimation model to calculate evaluation scores for sentences related to locations or services, using the sentiment score (such as a floating-point number between 0 and 1) as a recommendation candidate. The data processing procedure is as follows: the server converts the original continuous text into a labeled sequence (words, parts of speech, entity tags), extracts a keyword list and sentiment score, and encapsulates them into a "parsed result object". The output is a structured parsing result containing fields such as language type, keywords, entity list, and sentiment score. Specific actions include: the server calls a word segmentation function to traverse text segments; the server calls a sentiment analysis function to calculate an average sentiment score for each candidate location; and the server associates the parsing result with the original document using a unique ID.

[0420] Step 4: The server determines the category and assigns a label based on the parsing results.

[0421] The server's input consists of the parsed result object generated in step 3 and its associated document data. First, the server performs a preliminary classification using a rule table based on keywords and entity patterns in the information: if the text contains words like "attractions," "sightseeing," or "temple," it tends to be categorized as "tourism-related information"; if it contains words like "restaurant," "menu," or "reservation," it tends to be categorized as "food-related information"; if it contains words like "renting," "hospital," or "insurance," it tends to be categorized as "lifestyle-related information." Based on this, the server can send the text summary and parsing results to the generative AI model as prompts, such as: "Based on the following text, determine whether it belongs to 'lifestyle-related information,' 'tourism-related information,' or 'food-related information,' outputting only the category name: ...text...". The server receives the category labels output by the model and merges or weights them with the rule judgment results to determine the final category. Then, the server adds label information to the record based on the description in the text, such as "suitable for families," "suitable for children," "foreigner-friendly," "has an English menu," etc. The data processing process is as follows: the server maps the parsed features to the category space, generates a category identifier field, and constructs a label list field. The output is a "classification result record" with appended category identifiers and label information. The specific actions include: the server executing the rule matching algorithm; the server constructing a prompt statement to call the generative artificial intelligence model; and the server writing the category output by the model into a record.

[0422] Step 5: The server writes the classification results to the storage area and builds the index structure.

[0423] The server's input is the set of classification result records generated in step 4. The server writes each record to a database or search engine, storing fields such as document ID, city, country, category identifier, tag information, sentiment score, and credibility score in the corresponding tables or indexes. During data storage, the server builds or updates the index structure for each searchable field; for example, it builds a B+ tree index for the city field, an inverted index for the category and tag combination, and a word segmentation index for the full-text content. The data processing involves the server serializing the record structure in memory into a storage format, performing insert or update operations, and registering the corresponding key-value pairs in the index. The output is the state of the "storage area for retrieval" available for subsequent searching, including the updated main table and index tables. Specific actions include: the server performing insert / update SQL operations or search engine index update operations; and the server maintaining index metadata for fast querying.

[0424] Step 6: The user enters a prompt message in the terminal and sends it to the server.

[0425] The user's input is natural language query text, such as: "Please tell me recommended tourist attractions in Tokyo." or "This is my first time in Japan, please recommend some restaurants in Tokyo that are friendly to foreigners." The terminal receives the text entered by the user in the interface, uses it as a prompt, and adds meta-parameters such as the user's language preference and current location information. The data processing is as follows: the terminal reads the user input from the interface components, combines it into a request message structure (text, language code, optional location parameters), and sends it to the server via network protocol. The output is a request message sent from the terminal to the server, containing the prompt and relevant parameters. Specific actions include: the user types text in the input box and clicks send; the terminal performs basic validity checks on the input (non-empty length, character encoding) and calls the network transmission component to send the request.

[0426] Step 7: The server parses the user prompts and generates search criteria.

[0427] The server's input is the request message received from the terminal in step 6, the core content of which is the user's prompt statement. The server first performs language discrimination to confirm the language type of the prompt statement. Then, using intent recognition and slot filling models or rule extraction algorithms, it extracts geographic information (e.g., "Tokyo"), information categories (e.g., "tourist attractions," "restaurants"), and usage conditions (e.g., "first time in Japan," "suitable for foreigners") from the prompt statement. The server combines these extraction results into an internal "retrieval condition object," which includes city fields, category fields, tag constraints, and a confidence threshold. The data processing involves: the server performing word segmentation and semantic parsing on the prompt statement, mapping natural language to structured conditions. It may also construct internal auxiliary prompt statements to call a generative artificial intelligence model, for example: "Please extract the city name, category, and constraints from the following user question and return them in JSON format: ... user question...". After receiving the structured content output by the model, the server parses and writes it into the retrieval condition object. The output is a set of retrieval conditions used to access the storage area. Specific actions include: the server calling the intent recognition model; and the server performing field validation and default value filling on the model's returned results.

[0428] Step 8: The server accesses the storage area and selects candidate information based on the search criteria.

[0429] The server's input is the search criteria object generated in step 7. Using the city, category, tag conditions, and credibility threshold from this object, the server executes a query in the aforementioned storage area. The server quickly locates the set of records that meet the criteria using the index structure, calculates and sorts the comprehensive score of these records (which can be a weighted function composed of recommendation score, credibility score, and time factor). The data processing involves: the server executing a multi-condition combined query, filtering the returned records (e.g., removing those with scores below the threshold), and limiting the number of returned candidate entries (e.g., Top N). The output is a set of candidate information records, including fields such as name, description, category, tags, score, and credibility. Specific actions include: the server constructing the specific query statement; the server converting the query results from the database or search engine into an internal list structure; and the server performing a sorting algorithm on the list and slicing it to extract the top few entries.

[0430] Step 9: The server constructs prompts for generative artificial intelligence models and generates response text.

[0431] The server's input consists of the candidate information records obtained in step 8 and the original user prompt statements from step 6. Based on a predefined system template, the server embeds candidate information summaries (such as the name, main features, and tags of each location) into new prompt statements, explicitly instructing the generative AI model on its task. For example: "You are an assistant providing local information to foreigners. Based on the following list of candidate locations, select the 3 to 5 most suitable ones and explain the reasons for the recommendations to the user in Simplified Chinese. User question: ...; Candidate information: ...." The data processing involves the server formatting the structured candidate records into text, concatenating system instructions, user questions, and candidate information to form a complete prompt statement, and passing it as input to the generative AI model. The generative AI model performs internal neural network inference, propagating the encoded prompt statement vector through a multi-layer self-attention network, calculating the output probability distribution for each location, and progressively sampling to obtain the natural language response text. The server receives the model's output text as the response text. The output is the natural language answer to the user question. Specific actions include: the server calling the model inference interface; the server removing redundant prefixes and suffixes when receiving the output, and performing sensitive content filtering or length pruning when necessary.

[0432] Step 10: The server generates the final response data and returns it to the terminal.

[0433] The server's input consists of the response text generated in step 9 and the candidate information records obtained in step 8. The server combines the response text and structured candidate information into a unified response data structure, which includes: directly displayable natural language descriptions, location entries for list display (name, tags, address, rating, etc.), and necessary links or additional fields. The data processing involves the server simplifying the candidate records, removing unnecessary fields, and possibly adding serial numbers or category tags to each record for front-end interface display. The output is a response data object returned to the terminal, containing the response text string and an array of location entries. Specific actions include: the server serializing the response data; and the server sending the response to the requesting terminal via network protocols.

[0434] Step 11: The terminal receives and displays the response data returned by the server.

[0435] The terminal's input is the response data sent by the server in step 10. The terminal parses the response data structure, reads the response text and the list of location entries, and presents them in the text area and list area respectively, according to the terminal's interface layout. The data processing is as follows: the terminal renders the text content into multi-line text, converts the location entries into cards or list items, and may further pass the geographic location field to the map component for location marking display. The output is a visually presented recommendation information interface on the user's screen. Specific actions include: the terminal updates the interface components; the terminal sends new requests or displays more detailed information based on further user actions (such as clicking on a location).

[0436] Step 12: Users can ask further questions or modify their search requirements based on the displayed results.

[0437] User input consists of subjective judgments of the results displayed on the terminal and new natural language prompts, such as: "Please recommend only Tokyo attractions suitable for families with children," or "Please recommend accommodations with lower prices." The user enters new prompts through the terminal's input box, and the terminal repeats step 6, sending the new prompts to the server. The data processing involves the user adjusting their requirements based on the initial results, and the terminal converting the new input into a request. The output is a new request message, initiating a new round of server query and generation. Specific actions include: the user reading the results and deciding to adjust their filtering criteria; the user entering new text, and the terminal sending a new query request, thus forming a multi-round interactive process.

[0438] Application Example 2 The process flow corresponding to the specific processing in Use Case 2 will be described below. The various parts of the system described below are implemented by the data processing device 12 and the intelligent device 14. In addition, the data processing device 12 is referred to as the "server" and the intelligent device 14 is referred to as the "terminal".

[0439] In technologies related to foreign language user information services, rule engines or fixed templates are typically used to translate multilingual texts, convert disaster prevention information, perform simple keyword filtering on information sources, and provide tourism or lifestyle information to terminals based on pre-defined search rules. However, existing technologies generally suffer from the following problems: (1) In multilingual scenarios, the server side often implements translation, retrieval, recommendation and question answering modules separately. Each module is independent of the other and cannot call the generative artificial intelligence model based on a unified instruction structure. This results in a complex system architecture, redundant interfaces, increased processing latency and resource consumption, and difficulty in efficiently generating personalized information in dynamic scenarios.

[0440] (2) In the process of providing disaster prevention information, life information and tourism information, the server usually only performs simple retrieval based on location or keywords. It lacks intelligent analysis and maintenance of the credibility and classification structure of information sources, and cannot automatically select highly credible information from a large number of information sources and perform dynamic classification, thus affecting the accuracy and timeliness of foreign language users obtaining information.

[0441] (3) When using machine learning or recommendation algorithms, existing systems mostly only model interests or historical behaviors and fail to deeply couple with the natural language generation capabilities of generative artificial intelligence models. This makes it difficult for servers to complete the personalized screening of facility data and generate a structured and readable natural language description that is consistent with the screening results in a unified processing flow.

[0442] (4) In terms of human-computer interaction, traditional systems often only respond based on the user’s explicit input, without making full use of multimodal data such as images and audio to infer the user’s emotional state, and without feeding back the emotional analysis results to the information selection and text generation process. As a result, they cannot dynamically adjust the information content and expression according to the user’s tension, anxiety, excitement, etc., resulting in a poor user experience. Especially in disaster prevention information or emergency information scenarios, they cannot provide emotionally adaptive information services.

[0443] (5) In question-and-answer services, existing technologies typically call general question-and-answer models through fixed interfaces. There is a lack of a unified construction method that revolves around "prompt statements" and can flexibly embed contextual information (such as classification information, location data, and emotional state). The server has difficulty driving the generative artificial intelligence model to perform multiple tasks (translation, question and answer, description generation, emotion adaptation, etc.) under a consistent prompt statement framework, thus limiting the scalability of the system and the overall utilization efficiency of computing resources.

[0444] Therefore, it is necessary to provide a new system design and processing mechanism to improve the way the server calls generative artificial intelligence models, integrates and classifies multi-source information, and fuses emotion data and recommendation data from the perspective of computer technology. This would enable the server to efficiently execute a series of processing steps such as translation, retrieval, classification, recommendation, question answering, and emotion-adaptive information generation under a unified prompt statement-driven framework, thereby improving the overall system's processing performance and user experience in multilingual and multimodal scenarios.

[0445] The specific processing performed by the specific processing unit 290 of the data processing apparatus 12 in Application Example 2 is achieved by the following means.

[0446] In this invention, the server includes: a means of parsing and translating multilingual text information using prompts for a generative artificial intelligence model to generate translation information for foreign language users; a means of parsing and translating disaster-related information using prompts for a generative artificial intelligence model to generate disaster prevention information; an editing means of parsing evaluation criteria for multiple information sources using prompts for a generative artificial intelligence model, parsing multiple information sources, selecting them based on credibility, and generating categorized information; a selection means of retrieving surrounding facility data based on location data obtained from a location acquisition device and user attribute data, calculating the suitability of each facility data using a machine learning model, and selecting the facility data as tourism information or lifestyle information according to the categorized information; and generating tourism information or lifestyle information containing the selected information. The system includes: structured data, and based on the structured data and user preference data, generating prompts for a generative artificial intelligence model to instruct the generation of explanatory text; an explanatory generation method that generates natural language explanatory information by the generative artificial intelligence model; an emotion estimation method that infers the user's emotional state based on facial expression data and sound data obtained from an image acquisition device and a sound acquisition device; a control method that adjusts the selection criteria for sightseeing information or life information in the selection method and the expression mode of explanatory information in the explanatory generation method according to the estimated emotional state; and a processing method that receives user input, location data, facial expression data, and sound data sent from a terminal device, and comprehensively executes the translation method, the editing method, the selection method, the explanatory generation method, and the emotion estimation method, and sends the output information to the terminal device. This allows for the construction of a computational workflow within the server using prompt statements as a unified interface. On one hand, a unified prompt statement structure enables multi-task invocation of generative AI models, achieving deep integration of natural language processing functions such as translation, question answering, and description generation with machine learning recommendation and information source classification. This reduces data transfer and redundant computation between modules, thereby improving overall processing efficiency and system scalability. On the other hand, through emotion inference and control methods, emotional information from images and audio is fused with location data, interest data, and classification information on the server side. This allows the server to dynamically adjust information selection and text generation strategies on the same processing pipeline, enabling personalized information provision with emotion adaptation in multilingual and multimodal scenarios. This improves the response quality and user experience of the computing system in foreign language information services, disaster prevention information provision, and tourism / life services.

[0447] "Generative artificial intelligence models" refer to data processing models that are based on machine learning algorithms and trained on large-scale data, and can automatically generate natural language text, translation results or other content based on input text or other data.

[0448] "Prompt statements" refer to instructional text generated by the server and input into the generative artificial intelligence model. They are used to specify the type of task the model should perform, constrain the output format, and provide contextual information.

[0449] "Multilingual text information" refers to digital information containing character sequences expressed in two or more natural languages, including but not limited to news, announcements, documentation, and user-input text.

[0450] "Foreign Language User Information" refers to translated or reorganized explanatory information, guidance information, tourist information, or lifestyle information provided to users who use a language other than the local language.

[0451] "Disaster-related information" refers to textual data related to natural disasters, emergencies and their response measures, including but not limited to early warning information, evacuation guidelines, safety precautions and recovery status announcements.

[0452] "Disaster prevention information" refers to explanatory information generated based on disaster-related information, used to prompt users to take preventive or response measures, and is applicable to providing safety guidance to users before, during, and after a disaster.

[0453] "Information source" refers to digital resources that provide raw data or content, including but not limited to websites, databases, application programming interfaces (APIs), and collections of electronic documents.

[0454] "Evaluation benchmarks" refer to a set of standards or rules used to measure the credibility, relevance, and timeliness of information sources, including scoring rules, quality indicators, and filtering conditions.

[0455] "Classification information" refers to structured labeled data obtained by analyzing information sources or facility data and assigning them to predetermined categories for subsequent retrieval, recommendation, and display.

[0456] "Editing methods" refers to the combination of hardware and software functions that generate classified information by analyzing and processing information sources and their evaluation criteria, and performing operations such as selection, screening, reorganization and classification.

[0457] "Location acquisition device" refers to a component or system used to acquire the location information of a user or terminal in physical space, including satellite positioning modules, wireless network positioning modules, and base station-based positioning modules.

[0458] "Location data" refers to digital data representing geographical location output by a location acquisition device, including coordinate data, location identifiers, or area identifiers.

[0459] "User attribute data" refers to descriptive data related to users, including language preferences, interest preferences, age group, nationality, or historical behavior records.

[0460] "Nearby facilities data" refers to a collection of data related to physical locations near the user's current location, including the location, category, business information, rating data, and tags of the facilities.

[0461] "Facility data" refers to structured or semi-structured information about a specific facility, including its name, type, location, services offered, and user reviews.

[0462] A "machine learning model" is an algorithmic model that learns from training data to establish a mapping relationship between inputs and outputs, and is used to perform tasks such as matching degree calculation, classification, prediction, or recommendation.

[0463] "Fitness" refers to a quantitative indicator that represents the degree of matching between facility data and specific user attributes, interests, or contexts, and is used to sort or filter facilities.

[0464] "Tourism information" refers to descriptive data related to tourism activities, used to guide users in visiting and experiencing attractions and activities, including attraction descriptions, route suggestions, and activity arrangements.

[0465] "Life information" refers to service and guidance information related to users' daily life scenarios, including shopping, medical treatment, transportation, waste sorting, and public services.

[0466] "Selection methods" refers to filtering and sorting surrounding facility data based on location data, user attribute data, and classification information in order to determine the combination of hardware and software functions for target tourism or lifestyle information.

[0467] "Structured data" refers to data represented by a predetermined data structure that is easy for computer programs to store, process, and transmit, including key-value pairs, records, tables, and tokenized text.

[0468] "Preference data" refers to digital data that reflects a user's personal preferences and interests, including historical click records, rating records, favorite records, and explicit preference settings.

[0469] "Explanatory text" refers to natural language text content used to explain, describe, or guide users in understanding information or facilities.

[0470] "Natural language explanatory information" refers to a collection of user-oriented explanatory texts expressed in natural language, which has semantic coherence and readability.

[0471] "Explanation generation methods" refers to the combination of hardware and software functions that generate prompt statements based on structured data and preference data and call generative artificial intelligence models to obtain natural language explanation information.

[0472] "Image acquisition device" refers to imaging components used to capture image data of users or the environment, including cameras, image sensors and their driving software, etc.

[0473] "Sound acquisition device" refers to audio components used to collect user voice or environmental sound data, including microphones and their signal processing software.

[0474] "Face data" refers to digital data extracted from images captured by image acquisition devices that reflects the facial expression features of users.

[0475] "Sound data" refers to audio signal data that has been collected and digitized by a sound acquisition device and is used for speech recognition or emotion analysis.

[0476] "Emotional state" refers to the classification results of a user's current psychological or emotional state, including but not limited to emotional tags such as happy, sad, tense, anxious, and excited.

[0477] "Emotion estimation methods" refer to a combination of hardware and software functions that infer a user's emotional state based on facial expression data and voice data, using pattern recognition or machine learning algorithms.

[0478] "Control measures" refers to the combination of control functions that adjust the selection criteria and expression methods of the means of expression based on the emotional state, so as to achieve adaptive content and tone.

[0479] "Terminal device" refers to a human-computer interaction device carried or operated by a user for sending input data and receiving output information, including mobile communication terminals, wearable devices and other information processing terminals.

[0480] "User input" refers to interactive data provided by users through terminal devices, including text input, voice input, touch operation, and selection commands.

[0481] "Processing means" refers to the combination of hardware and software functions that execute data reception, translation scheduling, editing, selection, description generation, and sentiment inference in the server, and output the result data to the terminal device.

[0482] "Question-answering methods" refer to a combination of functions that generate prompts for generative artificial intelligence models to instruct them to parse the question content and generate answer data, and then use the generative artificial intelligence models to output the question answers.

[0483] "Emotion-adaptive means" refers to a combination of functions that generate prompts for generative artificial intelligence models, instructing them to determine the content and expression of information based on user input text and emotional state, and generate output information that conforms to the current emotional state.

[0484] In a preferred embodiment of the present invention, the server serves as the central processing unit, and the terminal serves as the user-side human-computer interaction device, through which the user interacts with the server. The server can be a computing device with a multi-core processor and a graphics processing unit, such as a computing platform including a general-purpose central processing unit and a graphics acceleration unit, with built-in large-capacity storage and a high-speed network interface. The terminal can be a mobile communication terminal, a tablet device, a wearable device, etc., and includes at least a display unit, an input unit, an image acquisition device, and a sound acquisition device.

[0485] The server employs an operating system and application services on the software side. The application layer can utilize scripting languages, such as interpreted languages, to implement business logic. Regarding machine learning frameworks, the server can use deep learning frameworks for neural network model training and inference, for example, using tensor computation frameworks and other tensor computation frameworks to implement different types of models. In computer vision processing, the server can use image processing libraries to perform face detection and feature point extraction. For speech recognition and speech synthesis, the server can convert speech to text by calling general speech recognition and general speech synthesis service interfaces. In translation and natural language processing, the server can call cloud services such as general translation service interfaces, general natural language processing service interfaces, and general large-scale generative network interfaces.

[0486] The server is configured with several functional modules, including a translation module, an information source editing module, a facility selection module, a description generation module, a sentiment estimation module, a control module, a question-and-answer module, and a unified scheduling module. Data is transferred between these modules via a message bus or internal function calls. The data is organized and stored internally within the server using structured data formats (such as key-value structures, nested list structures, etc.).

[0487] When implementing translation, the server receives multilingual text information from the terminal, including Japanese, English, and other languages, and combines this text with prompts indicating the translation task to form an input sequence. The prompts generated internally by the server may include task descriptions, output language constraints, terminology processing requirements, etc. For example, the server might use the following prompts: Please translate the following Japanese news into simplified Chinese, and add a brief explanation in parentheses when you encounter Japanese-specific terms: 'A large-scale earthquake occurred in Tokyo'. " The server concatenates the aforementioned prompt with the original text to form the input sequence for the generative AI model. It then uses a pre-trained large-scale language generation network to perform inference operations, yielding the target language text. This generative AI model can employ a self-attention network based on an encoder-decoder structure, containing multiple layers of self-attention and feedforward sublayers, and using a multi-head attention mechanism to model the input sequence. During inference, the server employs beam search or other decoding strategies to achieve a balance between accuracy and speed.

[0488] The server uses a similar approach in translating and processing disaster prevention information, but the difference lies in adding constraints on the conciseness and accuracy of safety guidance in the prompts, for example: Please translate the following Japanese disaster prevention information into simplified Chinese. The translation should be concise, clear, and suitable for foreigners who have just arrived in Japan: 'The evacuation point is 00 Elementary School. Please remain calm and take action.' In the information source editing process, the server retrieves metadata from multiple network resources or databases for candidate information sources, such as source address, update time, publishing entity, historical error records, and user reviews. The server encodes this metadata into feature vectors and combines them with text representing evaluation criteria (e.g., "released by government agency," "update frequency," "whether it has been repeatedly proven to contain errors") to construct prompt statements for the generative AI model. This allows the model to output a credibility score and classification label for each information source. Because the generative AI model has contextual understanding capabilities, the server can inject multiple information source records into the prompt statements, enabling the model to compare and evaluate multiple sources simultaneously in a single inference, thereby reducing the communication load and processing time caused by multiple calls.

[0489] During facility selection, the server maintains a facility data store, which includes facility identifiers, latitude and longitude, categories (e.g., attractions, restaurants, shopping, medical facilities), tags (e.g., "suitable for families," "open late at night"), rating statistics, and visit counts. The recommendation model built using a machine learning framework can employ a multilayer perceptron structure. Its inputs include user features (language preferences, interest tags, historical click vectors, etc.) and facility features (category vectors, tag vectors, geolocation embedding vectors, etc.). The server performs embedding mapping on user and facility features, transforming high-dimensional sparse features into low-dimensional dense vectors, and then calculates matching scores through several fully connected layers and non-linear activation functions. During the training phase, the server uses historical interaction data (e.g., user clicks, dwell time) as supervision signals. By defining a binary cross-entropy loss or ranking loss function, gradient descent-like optimization algorithms are used for weight updates, and learning rate decay and weight regularization can be employed to prevent overfitting.

[0490] During the inference phase, after receiving the current location and user attribute data from the terminal, the server first uses a spatial indexing structure (such as a spherical distance index or a grid partitioning structure) to quickly retrieve a set of candidate facilities within a radius from the facility data storage. Then, it calculates a user-facility matching score for each candidate facility and selects the top few facilities based on their scores. By pre-utilizing spatial indexing and vectorized operations, the server significantly reduces the number of facilities and feature combinations that need to be evaluated, thereby reducing computational load and improving response speed.

[0491] In the description generation process, the server converts the selected set of facilities into structured data, such as organizing it by category into "Restaurant List," "Attraction List," "Activity List," etc., and adds fields such as distance, price level, and tags to each facility. Based on this structured data and user preference data, the server generates prompts containing categories, quantity requirements, tone requirements, etc., and then calls a generative artificial intelligence model to generate natural language descriptions. The server can use prompts like the following: "The user is currently near Tokyo Tower and is interested in 'sightseeing' and 'food'."

[0492] The following are nearby locations selected based on the database and model. Please categorize and describe them in Simplified Chinese: Dining room: 1. XXX Ramen Shop – Tags: Ramen, Local Recommendation, Price: Medium 2. YYY Cafe – Tags: Desserts, Quiet Environment, Price: Moderate Attractions: 1. Tokyo Tower – Tags: Landmark, Night View 2. Zojoji Temple – Tags: temple, history Activity: 1. Weekend Market – Tags: Local Specialties, Weekend Exclusive Require: 1. Output by three categories: 'Restaurants', 'Attractions', and 'Activities'.

[0493] 2. Each location is highlighted with one or two sentences, making it suitable for foreigners visiting Japan for the first time.

[0494] 3. All output should use Simplified Chinese. In this way, the server enables the generative AI model to output explanatory text that corresponds one-to-one with the internal structured data, avoiding the problem of needing to manually write a lot of rules in the traditional template generation method. It also uses prompts to uniformly constrain the output format, which is beneficial for subsequent partitioned display on the terminal interface.

[0495] In the emotion inference process, the server receives image frames and audio clips sent by the terminal. The server uses an image processing library for face detection and keypoint recognition, normalizes the face regions, and then inputs them into an emotion classification network. This emotion classification network can be a convolutional neural network, containing several convolutional layers, pooling layers, and fully connected layers, outputting probability distributions representing different emotion categories. When training this network, the server uses a dataset of face images labeled with emotion tags, updates the network weights using a cross-entropy loss function and a backpropagation algorithm, and can use data augmentation techniques (such as random cropping, horizontal flipping, and brightness variations) to improve the model's robustness in real-world environments.

[0496] For audio data, the server can first extract acoustic features such as Mel-frequency cepstral coefficients, and then input them into a recurrent neural network or a one-dimensional convolutional network for emotion recognition. Finally, the visual and auditory emotion results are fused in a weighted manner to obtain a stable emotion state estimate. Since the server uniformly maps multimodal features to the emotion state space, it can improve the accuracy of emotion recognition without increasing the burden on the terminal and reduce misjudgments caused by the instability of a single modality.

[0497] In the control process, the server dynamically adjusts its facility selection and description generation strategies based on emotional states. For example, when the presumed emotion is "anxious" or "nervous," the server can reduce the recommendation weight of highly stimulating activities and increase the matching weight of facilities related to safety, disaster prevention, and comfort, explicitly requiring "gentle tone and reassuring suggestions" during description generation. Conversely, when the presumed emotion is "excited," the server can increase the weight of activity and entertainment facilities and add encouraging descriptions to the description text. Through this control mechanism, the server establishes a feedback loop between the recommendation model and the generative artificial intelligence model, achieving policy adaptation based on emotional states. This enables a non-traditional rule combination and decision-making process within the computer, different from manual human selection.

[0498] In question-answering processing, the server receives natural language question text sent by the terminal. Optionally, the server first uses traditional natural language processing tools to extract keywords and topic tags, then combines this with internal knowledge base retrieval results and the question context to construct a high-information-density prompt input generative AI model. For example, when a user enters "Please tell me about Japan's garbage sorting methods" through the terminal, the server can construct the following prompt: "You are a life consultant for foreigners living in Japan. Please answer the questions in Simplified Chinese."

[0499] Question: How is waste classified in Japan? Please provide typical examples of incinerable waste, non-combustible waste, and recyclable waste.

[0500] Background information: 1. In most Japanese cities, garbage is generally classified into incinerable waste, non-combustible waste, and recyclable waste.

[0501] 2. Recyclable waste includes plastic bottles, aluminum cans, glass bottles, etc., and must be disposed of on designated dates.

[0502] Based on the background information and common sense, please generate a clear and concise answer, presented in bullet points. In this way, the server enables generative AI models to fully utilize the existing structured knowledge and rules on the server side in question-answering tasks, thereby improving the accuracy and consistency of answers, while reducing unnecessary external retrieval requests and lowering network communication load.

[0503] In this invention, the terminal is primarily responsible for user interaction and multimodal data acquisition. The terminal acquires location data through a positioning module, obtains user input through a touch interface, keyboard, or microphone, and acquires image data through a camera. The terminal can perform basic data preprocessing locally, such as scaling image frames, sampling and compressing audio, and performing language annotation on text, reducing the amount of data uploaded and thus lowering network bandwidth usage. After receiving structured results and natural language descriptions from the server, the terminal displays sightseeing and lifestyle information on the interface in the form of lists, map markers, and cards, and reads the text aloud when necessary.

[0504] Users can access multilingual translations, disaster prevention information, tourism recommendations, daily life Q&A, and emotion-adaptive content simply by inputting natural language and performing basic operations on the terminal. Because the server drives a generative artificial intelligence model through unified prompts and integrates traditional rule processing, retrieval processing, recommendation processing, and emotion analysis processing with specific data structures and control flows, this invention does not merely digitize manual operations; rather, it introduces a new task organization method and model invocation mechanism within the computer system. This mechanism significantly reduces the waste of computing resources and communication load by minimizing redundant calls, hierarchically encoding input data, and uniformly scheduling, thereby improving overall processing speed. Simultaneously, through multimodal feature fusion and emotion-state-based control logic, it enhances the accuracy and adaptability of output results, reducing the probability of erroneous recommendations and misleading information.

[0505] Furthermore, this invention can be implemented in various modified forms. For example, the server can employ a locally deployed generative artificial intelligence model without relying on external cloud services, or it can use a hierarchical combination of lightweight and large models depending on resource availability; the facility selection model can be replaced from a multilayer perceptron to a graph neural network to better utilize the spatial and semantic relationships between facilities; the emotion inference model can be replaced from a convolutional network to a hybrid network incorporating an attention mechanism to improve sensitivity to subtle changes in facial expressions. These modifications all operate within a unified prompting and scheduling framework and do not affect the technical effects achieved by this invention through the integration of translation, editing, selection, description generation, and emotion inference methods.

[0506] use Figure 14 The processing flow is explained.

[0507] Step 1: The user initiates a request on the terminal.

[0508] Users can input natural language commands into the terminal interface, such as typing "Please tell me how Japan sorts its garbage," or turning on the location function and selecting "Nearby sightseeing recommendations," or asking "Is this product gluten-free?" via voice.

[0509] Input: The user's raw input data (text, voice, location request commands, etc.).

[0510] The terminal performs local preprocessing on voice input, sampling the voice into a digital audio stream, and converting the audio into a text string by calling a speech recognition service interface when necessary. The terminal also reads the latitude and longitude data of the current location from the location acquisition device, and collects facial expression images and environmental voice clips from the image acquisition device and sound acquisition device.

[0511] Output: A request data packet containing user text instructions, location information (latitude and longitude), optional image frames, and audio clips.

[0512] Step 2: The terminal sends a request for data to the server.

[0513] The terminal encapsulates the request data packet generated in step 1 into a structured message (containing fields such as "text_input", "location", "image_data", "audio_data", etc.) and sends it to the specified interface of the server via the communication module using a network protocol.

[0514] Inputs: User input data, location information, and multimodal sensing data.

[0515] The terminal encodes and compresses various types of data according to a predefined data structure, such as scaling image data to a smaller resolution and compressing audio sampling rates, thereby reducing the size of data packets.

[0516] Output: A stream of request messages transmitted to the server over the network.

[0517] Step 3: The server parses the request message and identifies the task.

[0518] The server receives request messages from the terminal, parses the message body at the application layer, and extracts text input, location information, image data, and audio data. Based on the text content and request type flags, the server determines whether the current request belongs to a task category such as translation, disaster prevention information generation, tourism recommendations, lifestyle Q&A, or product Q&A.

[0519] Input: Structured request message (including text, location, multimodal data, and task flags).

[0520] The server performs preprocessing on the text fields, such as word segmentation and keyword extraction, and performs pattern matching to determine whether the text contains question marks or key phrases (such as "how to get there", "translate", "recommend" etc.), thereby classifying the tasks into "translation tasks", "question and answer tasks", "recommendation tasks", etc.

[0521] Output: An internal task description object containing task category labels, normalized text, location vectors, and multimodal data references.

[0522] Step 4: The server performs preprocessing on the multimodal data for sentiment inference.

[0523] The server reads image and audio data from the task description object. If they exist, it proceeds to the emotion estimation process. The server uses image processing algorithms to perform face detection and face region cropping on the image, normalizing the face image into a fixed-size grayscale or color matrix. The server extracts short-time energy, spectral features, or Mel-frequency cepstral coefficients from the audio data.

[0524] Input: Image frame data, audio clips.

[0525] The server performs operations such as convolution filtering, threshold detection, and face bounding box localization on the input image, and performs time window segmentation and Fourier transform on the audio to obtain a feature matrix suitable for the emotion classification model.

[0526] Output: Face feature image and audio feature vector set.

[0527] Step 5: The server estimates the user's emotional state using a sentiment inference model.

[0528] The server inputs the facial feature image obtained in step 4 into a convolutional neural network and the audio feature vector into an acoustic emotion recognition network, obtaining visual and auditory emotion probability distributions respectively. The server then performs a weighted fusion of these two distributions to obtain the final emotion state label (such as "calm," "anxious," "excited," etc.) and its confidence level.

[0529] Input: Visual feature data, audio feature data.

[0530] The server performs matrix multiplication, convolution, nonlinear activation, and normalization operations inside the neural network. It calculates the probability vector of emotion categories through forward propagation and then combines the multimodal outputs through weighted averaging or Bayesian fusion algorithms.

[0531] Output: Emotional state label and corresponding confidence score.

[0532] Step 6: The server selects the data source and processing strategy based on the task category and emotional state.

[0533] The server reads the task category (such as Q&A, recommendation, translation) and emotional state, and determines the type of data source to be used (such as life knowledge base, facility database, disaster prevention information database) and filtering strategy (such as prioritizing information that reduces high risk of anxiety) from the internal rule table.

[0534] Input: Task category label, emotional state label.

[0535] The server internally performs rule matching and weight adjustment operations, mapping emotional states to a set of weight coefficients. These coefficients are used to modify feature weights in the recommendation model or modify query retrieval conditions, such as increasing the scores of facilities labeled "safe" and "comfortable".

[0536] Output: A processing strategy object containing pointers to the selected data source and the adjusted weight configuration.

[0537] Step 7: The server retrieves and filters basic data.

[0538] The server accesses the corresponding data storage module according to the processing strategy object: in the translation task, it reads the text to be translated; in the disaster prevention task, it retrieves the latest disaster notices; in the tourism recommendation task, it uses spatial indexes to retrieve candidate facilities near the current location from the facility database; and in the question-answering task, it retrieves entries matching the question topic from the knowledge base.

[0539] Inputs: Processing strategy objects, raw text, location data, knowledge base, and facility database.

[0540] The server executes database queries, full-text searches, or spatial range queries, compiles the records that meet the criteria into a list, and arranges them according to initial sorting rules (such as timestamps or base scores).

[0541] Output: A set of basic data for subsequent in-depth processing, such as a list of candidate facilities, a list of candidate knowledge fragments, and a set of text to be translated.

[0542] Step 8: The server calculates individualized matching scores using machine learning models.

[0543] For tourism or lifestyle information recommendation scenarios, the server combines user attribute features (language preferences, interest tags, historical behavior vectors) with candidate facility features (category, tags, location embedding, average rating), and inputs them into a pre-trained recommendation neural network to calculate the matching score of each facility for the current user.

[0544] Input: User feature vector, candidate facility feature vector.

[0545] The server performs embedding lookup, vector concatenation, and multilayer perceptron forward propagation in the recommender network. It obtains the output score through a series of linear transformations and nonlinear activations, and then recalibrates the score according to the emotional state weights.

[0546] Output: A list of matching scores for each facility.

[0547] Step 9: The server selects target information based on matching degree and classification information.

[0548] The server uses the matching score from step 8 to sort the candidate facilities and combines it with the classification information generated by the information source editing module (such as category tags like "attractions", "dining", "activities", "safety information") to select the highest-scoring items from each category as target sightseeing information or lifestyle information.

[0549] Input: a list of matching scores, a list of candidate facilities, and category information.

[0550] The server executes sorting and grouping extraction algorithms, such as selecting the Top-N records after binning by category, while removing facilities that have too much overlap with the user's historical usage records to ensure recommendation diversity.

[0551] Output: A structured collection of target information, containing a list of facilities or knowledge entries organized by category.

[0552] Step 10: The server generates structured data and prepares prompt statements.

[0553] The server transforms the target information selected in step 9 into a structured data format, including category fields, name fields, brief label fields, and auxiliary parameters (such as distance and price level). Simultaneously, the server combines user preference data and task context to construct corresponding prompts, which drive the generative artificial intelligence model to generate natural language output.

[0554] Input: target information set, user preference data, task context (e.g., "translate", "explain", "question and answer").

[0555] The server performs field extraction and text concatenation operations, inserting structured data into the prompt statement template in the form of entries. For example, the text in the template is organized in the format of "Restaurant: 1.…2.… Attractions: 1.…2.… Activities: 1.…".

[0556] Output: The complete text of the prompts for generative artificial intelligence models.

[0557] Step 11: The server invokes a generative artificial intelligence model to perform natural language generation or translation.

[0558] The server inputs the prompt obtained in step 10 into the generative artificial intelligence model. Based on its internal encoder-decoder structure, the model decodes the prompt into target text, such as translation results, question-and-answer results, or comprehensive explanatory text.

[0559] Input: Prompt text.

[0560] During model inference, the server performs tensor computation and attention weighting operations to model the task description, data entries, and output requirements in the prompt statements as a whole. Then, it uses decoding algorithms (such as greedy decoding or bundle search) to gradually generate word sequences in the target language.

[0561] Output: Natural language text output by the generative artificial intelligence model, such as translated disaster prevention information, tourist information, question-and-answer answers, or emotion-adaptive suggestions.

[0562] Step 12: The server performs post-processing and formatting on the generated results.

[0563] The server examines the text output from step 11, filtering and correcting obvious formatting errors, garbled characters, or sensitive content. It also segments the text and adds category headings or list markers according to the terminal's display requirements. The server can also adjust the tone and emphasis of sentences based on the user's mood.

[0564] Input: Original generated text, mood state, display configuration.

[0565] The server performs string processing (line breaks, sentence breaks, category annotation), keyword replacement, and length pruning to ensure that the output is visually readable and logically clear.

[0566] Output: The formatted final text result and its corresponding data structure, which is sent to the terminal.

[0567] Step 13: The server returns response data to the terminal.

[0568] The server packages the formatted text and necessary structured information into a response message and sends it to the requesting terminal via the network interface.

[0569] Input: Final text result, structured display data.

[0570] During this stage, the server performs serialization and network transmission operations to ensure that the message reliably reaches the terminal under the agreed protocol and security mechanism.

[0571] Output: A response message that can be used directly on the terminal.

[0572] Step 14: The terminal displays the results and provides an interactive entry point.

[0573] The terminal receives response messages from the server, parses the text content and structured fields, displays the sightseeing information by category on the list and map, displays the Q&A results in the dialogue area, and displays the translated disaster prevention information in a prominent format.

[0574] Input: Server response message.

[0575] The terminal performs layout rendering, text drawing, and interactive control binding on the graphical interface, distinguishes and displays different categories of information using tabs or collapsible panels, and calls local speech synthesis to read the text to the user when needed.

[0576] Output: Results presented in visual and / or audio format for direct user understanding and operation.

[0577] Step 15: Users take action based on the results and generate feedback.

[0578] After reading or listening to information provided by the server on their device, users can perform further actions, such as clicking on a recommended facility for navigation, adding follow-up questions to the answers, or providing a simple evaluation of the translation quality.

[0579] Input: The result information displayed and broadcast on the terminal.

[0580] User clicks, rating inputs, or new follow-up questions are collected by the terminal and repackaged into request data, which is then sent back to the server for subsequent model optimization and personalization updates.

[0581] Output: New user interaction data and feedback signals, providing additional context and training samples for the next round of request processing.

[0582] The specific processing unit 290 sends the result of the specific processing to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires sound representing user input regarding the result of the specific processing. The control unit 46A sends the sound data representing user input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the sound data.

[0583] Data generation model 58 is a so-called generative AI (Artificial Intelligence). Examples of data generation models 58 include ChatGPT (registered trademark) (accessible via the internet (URL: https: / / openai.com / blog / chatgpt)). Data generation model 58 is obtained through deep learning on a neural network. Input to data generation model 58 are prompt words containing instructions, and inference data such as sound data representing sound, text data representing text, and image data representing images (e.g., still image data or animation data). Data generation model 58 infers from the input inference data based on the instructions represented by the prompt words and outputs the inference result in one or more data forms, such as sound data, text data, and image data. Data generation model 58 includes, for example, text generation AI, image generation AI, and multimodal generation AI. Here, inference refers to, for example, analysis, classification, prediction, and / or induction. The specific processing unit 290 performs the aforementioned specific processing while using data generation model 58. The data generation model 58 can also be a model finely tuned to output inference results from prompts that do not contain instructions. In this case, the data generation model 58 can output inference results based on prompts that do not contain instructions. The data processing apparatus 12, etc., includes various data generation models 58, including AI other than the generation AI. AI other than the generation AI can be, for example, linear regression, logistic regression, decision trees, random forests, support vector machines (SVM), k-means clustering, convolutional neural networks (CNN), recurrent neural networks (RNN), generative adversarial networks (GAN), or Naive Bayes, and can perform various processes, but is not limited to this example. Furthermore, the AI ​​can also be an AI agent. Furthermore, when the processing of the above-mentioned parts is performed by AI, the processing can be performed partially or entirely by AI, but is not limited to this example. Furthermore, the processing performed by the AI ​​including the generation AI can be replaced by processing in the rule base, and the processing in the rule base can also be replaced by processing performed by the AI ​​including the generation AI.

[0584] Furthermore, the processing of the aforementioned data processing system 10 is performed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart device 14, but it can also be performed by both the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart device 14. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart device 14 or external devices, and the smart device 14 acquires or collects information required for processing from the data processing device 12 or external devices.

[0585] For example, the collection unit is implemented by the control unit 46A of the smart device 14 or the specific processing unit 290 of the data processing device 12. For example, the acquisition unit uses the camera 42 or communication I / F 44 of the smart device 14 to acquire step data, which is then processed by the specific processing unit 290 of the data processing device 12. For example, the analysis unit is implemented by the specific processing unit 290 of the data processing device 12, which analyzes the data from the collection unit and the acquisition unit. For example, the generation unit is implemented by the specific processing unit 290 of the data processing device 12, which uses a generation AI to generate a menu. For example, the serving unit is implemented by the output device 40 of the smart device 14 or the specific processing unit 290 of the data processing device 12, which provides the generated menu to the user. The correspondence between each unit and the device or control unit is not limited to the above examples and various changes can be made.

[0586] In the above embodiments, examples of specific processing by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart device 14.

[0587] Second Implementation Method Figure 3 An example of the configuration of the data processing system 210 according to the second embodiment is shown.

[0588] like Figure 3 As shown, the data processing system 210 includes a data processing device 12 and smart glasses 214. A server can be cited as an example of the data processing device 12.

[0589] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0590] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, and communication I / F 44 are also connected to the bus 52.

[0591] Microphone 238 receives instructions from user 20 by receiving sounds emitted by user 20. Microphone 238 captures sounds emitted by user 20 and converts the captured sounds into sound data, which is then output to processor 46. Speaker 240 outputs sound according to instructions from processor 46.

[0592] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and imaging elements such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, which captures images of the user 20's surroundings (e.g., the field of view defined by an angle equivalent to the field of vision of an average healthy person).

[0593] Communication I / F44 is connected to network 54. Communication I / F44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F44 and 26 is performed in a secure state.

[0594] Figure 4 This illustrates an example of the main functions of the data processing device 12 and the smart glasses 214. For example... Figure 4 As shown, in the data processing device 12, specific processing is performed by the processor 28. The specific processing program 56 is stored in the memory 32.

[0595] The specific processing program 56 is an example of a "program" involved in the technology of this disclosure. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0596] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. In the emotion inference function (emotion-specific function) using the emotion-specific model 59, various inferences and predictions related to the user's emotions are performed, including inferences and predictions of the user's emotions, but this is not limited to this example. Furthermore, emotion inference and prediction may also include, for example, emotion analysis (parsing).

[0597] In the smart glasses 214, the processor 46 performs reception and output processing. The memory 50 stores the reception and output program 60. The processor 46 reads the reception and output program 60 from the memory 50 and executes the read reception and output program 60 on the RAM 48. The reception and output processing is implemented by the processor 46 operating as a control unit 46A according to the reception and output program 60 executed on the RAM 48. Furthermore, the smart glasses 214 has the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and these models can also be used to perform the same processing as the specific processing unit 290.

[0598] Next, the specific processing of the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the smart glasses 214. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal".

[0599] Example 1 The process is the same as that of the specific process described in Embodiment 1 in the first embodiment above, so the description is omitted.

[0600] Application Example 1 The process is the same as that in the specific processing described in Application Example 1 of the first embodiment above, so the description is omitted.

[0601] Example 2 The process is the same as that of the specific process in Embodiment 2 described in the first embodiment above, so the description is omitted.

[0602] Application Example 2 The process is the same as that in the specific processing described in Application Example 2 of the first embodiment above, so the description is omitted.

[0603] The specific processing unit 290 sends the result of the specific processing to the smart glasses 214. In the smart glasses 214, the control unit 46A outputs the result of the specific processing to the speaker 240. The microphone 238 acquires sound input representing the user's input regarding the result of the specific processing. The control unit 46A sends the sound data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the sound data.

[0604] Data generation model 58 is a so-called generative AI (Artificial Intelligence). Examples of data generation models 58 include ChatGPT (registered trademark) (accessible via the internet (URL: https: / / openai.com / blog / chatgpt)). Data generation model 58 is obtained through deep learning on a neural network. Input to data generation model 58 are prompt words containing instructions, and inference data such as sound data representing sound, text data representing text, and image data representing images (e.g., still image data or animation data). Data generation model 58 infers from the input inference data based on the instructions represented by the prompt words and outputs the inference result in one or more data forms, such as sound data, text data, and image data. Data generation model 58 includes, for example, text generation AI, image generation AI, and multimodal generation AI. Here, inference refers to, for example, analysis, classification, prediction, and / or induction. The specific processing unit 290 performs the aforementioned specific processing while using data generation model 58. The data generation model 58 can also be a model finely tuned to output inference results from prompts that do not contain instructions. In this case, the data generation model 58 can output inference results based on prompts that do not contain instructions. The data processing apparatus 12, etc., includes various data generation models 58, including AI other than the generation AI. AI other than the generation AI can be, for example, linear regression, logistic regression, decision trees, random forests, support vector machines (SVM), k-means clustering, convolutional neural networks (CNN), recurrent neural networks (RNN), generative adversarial networks (GAN), or Naive Bayes, and can perform various processes, but is not limited to this example. Furthermore, the AI ​​can also be an AI agent. Furthermore, when the processing of the above-mentioned parts is performed by AI, the processing can be performed partially or entirely by AI, but is not limited to this example. Furthermore, the processing performed by the AI ​​including the generation AI can be replaced by processing in the rule base, and the processing in the rule base can also be replaced by processing performed by the AI ​​including the generation AI.

[0605] Furthermore, the processing of the aforementioned data processing system 10 is performed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart glasses 214, but it can also be performed by both the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart glasses 214. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart glasses 214 or external devices, and the smart glasses 214 acquires or collects information required for processing from the data processing device 12 or external devices.

[0606] For example, the collection unit is implemented by the control unit 46A of the smart glasses 214 or the specific processing unit 290 of the data processing device 12. For example, the acquisition unit uses the camera 42 or communication I / F 44 of the smart glasses 214 to acquire step data, which is then processed by the specific processing unit 290 of the data processing device 12. For example, the analysis unit is implemented by the specific processing unit 290 of the data processing device 12, which analyzes the data from the collection unit and the acquisition unit. For example, the generation unit is implemented by the specific processing unit 290 of the data processing device 12, which uses a generation AI to generate a menu. For example, the serving unit is implemented by the speaker 240 of the smart glasses 214 or the specific processing unit 290 of the data processing device 12, which provides the generated menu to the user. The correspondence between each unit and the device or control unit is not limited to the above examples and various changes can be made.

[0607] In the above embodiments, examples of specific processing by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart glasses 214.

[0608] Third Implementation Method Figure 5 An example of the configuration of the data processing system 310 according to the third embodiment is shown.

[0609] like Figure 5 As shown, the data processing system 310 includes a data processing device 12 and a head-mounted terminal 314. A server can be cited as an example of the data processing device 12.

[0610] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0611] The head-mounted terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, display 343, and communication I / F 44 are also connected to the bus 52.

[0612] Microphone 238 receives instructions from user 20 by receiving sounds emitted by user 20. Microphone 238 captures sounds emitted by user 20 and converts the captured sounds into sound data, which is then output to processor 46. Speaker 240 outputs sound according to instructions from processor 46.

[0613] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and imaging elements such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, which captures images of the user 20's surroundings (e.g., the field of view defined by an angle equivalent to the field of vision of an average healthy person).

[0614] Communication I / F44 is connected to network 54. Communication I / F44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F44 and 26 is performed in a secure state.

[0615] Figure 6 This illustrates an example of the main functions of the data processing device 12 and the head-mounted terminal 314. For example... Figure 6 As shown, in the data processing device 12, specific processing is performed by the processor 28. The specific processing program 56 is stored in the memory 32.

[0616] The specific processing program 56 is an example of a "program" involved in the technology of this disclosure. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0617] The memory 32 stores the data generation model 58 and the emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290.

[0618] In the head-mounted terminal 314, the processor 46 performs the acceptance / output processing. The memory 50 stores the acceptance / output program 60. The processor 46 reads the acceptance / output program 60 from the memory 50 and executes the read acceptance / output program 60 on the RAM 48. The acceptance / output processing is implemented by the processor 46 operating as a control unit 46A according to the acceptance / output program 60 executed on the RAM 48.

[0619] Next, the specific processing of the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the head-mounted terminal 314. In the following description, the data processing device 12 will be referred to as the "server" and the head-mounted terminal 314 will be referred to as the "terminal".

[0620] Example 1 The process is the same as that of the specific process described in Embodiment 1 in the first embodiment above, so the description is omitted.

[0621] Application Example 1 The process is the same as that in the specific processing described in Application Example 1 of the first embodiment above, so the description is omitted.

[0622] Example 2 The process is the same as that of the specific process in Embodiment 2 described in the first embodiment above, so the description is omitted.

[0623] Application Example 2 The process is the same as that in the specific processing described in Application Example 2 of the first embodiment above, so the description is omitted.

[0624] The specific processing unit 290 sends the result of the specific processing to the head-mounted terminal 314. In the head-mounted terminal 314, the control unit 46A outputs the result of the specific processing to the speaker 240 and the display 343. The microphone 238 acquires sound input representing the user's input regarding the result of the specific processing. The control unit 46A sends the sound data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the sound data.

[0625] Data generation model 58 is a so-called generative AI (Artificial Intelligence). Examples of data generation models 58 include ChatGPT (registered trademark) (accessible via the internet (URL: https: / / openai.com / blog / chatgpt)). Data generation model 58 is obtained through deep learning on a neural network. Input to data generation model 58 includes prompt words containing instructions, and inference data such as sound data representing sound, text data representing text, and image data representing images (e.g., still image data or animation data). Data generation model 58 infers the input inference data based on the instructions represented by the prompt words and outputs the inference result in one or more data forms such as sound data, text data, and image data. Data generation model 58 includes, for example, text generation AI, image generation AI, and multimodal generation AI. Here, inference refers to, for example, analysis, classification, prediction, and / or induction. The specific processing unit 290 performs the aforementioned specific processing while using data generation model 58. The data generation model 58 can also be a model finely tuned to output inference results from prompts that do not contain instructions. In this case, the data generation model 58 can output inference results based on prompts that do not contain instructions. The data processing apparatus 12, etc., includes various data generation models 58, including AI other than the generation AI. AI other than the generation AI can be, for example, linear regression, logistic regression, decision trees, random forests, support vector machines (SVM), k-means clustering, convolutional neural networks (CNN), recurrent neural networks (RNN), generative adversarial networks (GAN), or Naive Bayes, and can perform various processes, but is not limited to this example. Furthermore, the AI ​​can also be an AI agent. Furthermore, when the processing of the above-mentioned parts is performed by AI, the processing can be performed partially or entirely by AI, but is not limited to this example. Furthermore, the processing performed by the AI ​​including the generation AI can be replaced by processing in the rule base, and the processing in the rule base can also be replaced by processing performed by the AI ​​including the generation AI.

[0626] Furthermore, the processing of the aforementioned data processing system 10 is performed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the head-mounted terminal 314, but it can also be performed by both the specific processing unit 290 of the data processing device 12 and the control unit 46A of the head-mounted terminal 314. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the head-mounted terminal 314 or external devices, and the head-mounted terminal 314 acquires or collects information required for processing from the data processing device 12 or external devices.

[0627] For example, the collection unit is implemented by the control unit 46A of the head-mounted terminal 314 or the specific processing unit 290 of the data processing device 12. For example, the acquisition unit uses the camera 42 or communication I / F 44 of the head-mounted terminal 314 to acquire step data, which is then processed by the specific processing unit 290 of the data processing device 12. For example, the analysis unit is implemented by the specific processing unit 290 of the data processing device 12, which analyzes the data from the collection unit and the acquisition unit. For example, the generation unit is implemented by the specific processing unit 290 of the data processing device 12, which uses a generation AI to generate a menu. For example, the serving unit is implemented by the speaker 240 and display 343 of the head-mounted terminal 314 or the specific processing unit 290 of the data processing device 12, which provides the generated menu to the user. The correspondence between each unit and the device or control unit is not limited to the above examples and various changes can be made.

[0628] In the above embodiments, examples of specific processing by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the head-mounted terminal 314.

[0629] Fourth Implementation Method Figure 7 An example of the configuration of the data processing system 410 according to the fourth embodiment is shown.

[0630] like Figure 7 As shown, the data processing system 410 includes a data processing device 12 and a robot 414. A server can be cited as an example of the data processing device 12.

[0631] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0632] Robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a controlled object 443. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, controlled object 443, and communication I / F 44 are also connected to the bus 52.

[0633] Microphone 238 receives instructions from user 20 by receiving sounds emitted by user 20. Microphone 238 captures sounds emitted by user 20 and converts the captured sounds into sound data, which is then output to processor 46. Speaker 240 outputs sound according to instructions from processor 46.

[0634] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and imaging elements such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, to photograph the area around robot 414 (e.g., the field of view defined by a perspective equivalent to the field of vision of an average healthy person).

[0635] Communication I / F44 is connected to network 54. Communication I / F44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F44 and 26 is performed in a secure state.

[0636] The controlled object 443 includes a display device, LEDs (light-emitting diodes) for the eyes, and motors for driving the arms, hands, and feet. The posture or movement of the robot 414 is controlled by controlling the motors in the arms, hands, and feet. Some of the robot 414's emotions can be expressed by controlling these motors. In addition, the robot 414's facial expressions can also be expressed by controlling the illumination state of the LEDs in its eyes.

[0637] Figure 8 This illustrates an example of the main functions of the data processing device 12 and the robot 414. For example... Figure 8 As shown, in the data processing device 12, specific processing is performed by the processor 28. The specific processing program 56 is stored in the memory 32.

[0638] The specific processing program 56 is an example of a "program" involved in the technology of this disclosure. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0639] The memory 32 stores the data generation model 58 and the emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290.

[0640] In robot 414, the processor 46 performs the acceptance and output processing. The memory 50 stores the acceptance and output program 60. The processor 46 reads the acceptance and output program 60 from the memory 50 and executes the read acceptance and output program 60 on RAM 48. The acceptance and output processing is implemented by the processor 46 acting as the control unit 46A according to the acceptance and output program 60 executed on RAM 48.

[0641] Next, the specific processing of the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the robot 414. In the following description, the data processing device 12 will be referred to as the "server" and the robot 414 will be referred to as the "terminal".

[0642] Example 1 The process is the same as that of the specific process described in Embodiment 1 in the first embodiment above, so the description is omitted.

[0643] Application Example 1 The process is the same as that in the specific processing described in Application Example 1 of the first embodiment above, so the description is omitted.

[0644] Example 2 The process is the same as that of the specific process in Embodiment 2 described in the first embodiment above, so the description is omitted.

[0645] Application Example 2 The process is the same as that in the specific processing described in Application Example 2 of the first embodiment above, so the description is omitted.

[0646] The specific processing unit 290 sends the result of the specific processing to the robot 414. In the robot 414, the control unit 46A outputs the result of the specific processing to the speaker 240 and the controlled object 443. The microphone 238 acquires sound input representing the result of the specific processing. The control unit 46A sends the sound data representing the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the sound data.

[0647] Data generation model 58 is a so-called generative AI (Artificial Intelligence). Examples of data generation models 58 include ChatGPT (registered trademark) (accessible via the internet (URL: https: / / openai.com / blog / chatgpt)). Data generation model 58 is obtained through deep learning on a neural network. Input to data generation model 58 are prompt words containing instructions, and inference data such as sound data representing sound, text data representing text, and image data representing images (e.g., still image data or animation data). Data generation model 58 infers from the input inference data based on the instructions represented by the prompt words and outputs the inference result in one or more data forms, such as sound data, text data, and image data. Data generation model 58 includes, for example, text generation AI, image generation AI, and multimodal generation AI. Here, inference refers to, for example, analysis, classification, prediction, and / or induction. The specific processing unit 290 performs the aforementioned specific processing while using data generation model 58. The data generation model 58 can also be a model finely tuned to output inference results from prompts that do not contain instructions. In this case, the data generation model 58 can output inference results based on prompts that do not contain instructions. The data processing apparatus 12, etc., includes various data generation models 58, including AI other than the generation AI. AI other than the generation AI can be, for example, linear regression, logistic regression, decision trees, random forests, support vector machines (SVM), k-means clustering, convolutional neural networks (CNN), recurrent neural networks (RNN), generative adversarial networks (GAN), or Naive Bayes, and can perform various processes, but is not limited to this example. Furthermore, the AI ​​can also be an AI agent. Furthermore, when the processing of the above-mentioned parts is performed by AI, the processing can be performed partially or entirely by AI, but is not limited to this example. Furthermore, the processing performed by the AI ​​including the generation AI can be replaced by processing in the rule base, and the processing in the rule base can also be replaced by processing performed by the AI ​​including the generation AI.

[0648] Furthermore, the processing of the aforementioned data processing system 10 is performed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the robot 414, but it can also be performed by both the specific processing unit 290 of the data processing device 12 and the control unit 46A of the robot 414. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the robot 414 or external devices, and the robot 414 acquires or collects information required for processing from the data processing device 12 or external devices.

[0649] For example, the collection unit is implemented by the control unit 46A of the robot 414 or the specific processing unit 290 of the data processing device 12. For example, the acquisition unit uses the camera 42 or communication I / F 44 of the robot 414 to acquire step data, which is then processed by the specific processing unit 290 of the data processing device 12. For example, the analysis unit is implemented by the specific processing unit 290 of the data processing device 12, which analyzes the data from the collection unit and the acquisition unit. For example, the generation unit is implemented by the specific processing unit 290 of the data processing device 12, which uses a generation AI to generate a menu. For example, the serving unit is implemented by the speaker 240 of the robot 414 and the control object 443 or the specific processing unit 290 of the data processing device 12, which provides the generated menu to the user. The correspondence between each unit and the device or control unit is not limited to the above examples and various changes can be made.

[0650] In the above embodiments, examples of specific processing by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the robot 414.

[0651] Furthermore, the emotion-specific model 59, acting as an emotion engine, can determine a user's emotion based on a specific mapping. Specifically, the emotion-specific model 59 can determine a user's emotion based on an emotion graph that serves as a specific mapping (see [reference]). Figure 9 The emotion-specific model 59 can also determine the robot's emotion, and the specific processing unit 290 performs specific processing based on the robot's emotions.

[0652] Figure 9 This is a diagram representing an emotion map 400 that maps multiple emotions. In the emotion map 400, emotions are arranged in concentric circles radiating from the center. The closer to the center of the concentric circles, the more primitive the emotion is. On the outer side of the concentric circles, emotions representing states or behaviors arising from mood are arranged. Emotions are concepts that include feelings and mental states. Emotions generated by reactions occurring in the brain are arranged roughly to the left of the concentric circles. Emotions derived from situational judgments are arranged roughly to the right of the concentric circles. Emotions generated by reactions occurring in the brain and derived from situational judgments are arranged roughly above and below the concentric circles. Furthermore, "pleasant" emotions are arranged above the concentric circles, and "unpleasant" emotions are arranged below them. Thus, in the emotion map 400, multiple emotions are mapped based on the structure that generates emotions, and emotions that are likely to occur simultaneously are mapped close to each other.

[0653] These emotions are distributed at the three o'clock position of the emotion map 400, typically fluctuating between peace and anxiety. In the right half of the emotion map 400, situational awareness dominates over internal sensation, thus resulting in an impression of calm.

[0654] The inner side of the emotion map 400 represents the inner state, while the outer side represents behavior. Therefore, the further outward you are from the emotion map 400, the more visible the emotion becomes (manifested in behavior).

[0655] Here, human emotions are based on various balances such as posture and blood sugar levels. When these balances deviate from an ideal state, it indicates an unpleasant state; when they approach the ideal state, it indicates a pleasant state. Emotions in robots, cars, motorcycles, etc., can also be created in the following way: based on various balances such as posture and remaining battery power, when these balances deviate from an ideal state, it indicates an unpleasant state; when they approach the ideal state, it indicates a pleasant state. Emotion maps can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on a Brain Physiological Signal Analysis System for Voice Emotion Recognition and Emotion, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). In the left half of the emotion map, emotions belonging to the sensory-dominated region, called "response," are arranged. Furthermore, in the right half of the emotion map, emotions belonging to the situational cognition-dominated region, called "situation," are arranged.

[0656] In the emotion map, two types of emotions that promote learning are defined. One is a negative emotion on the situational side, in the middle or peripheral region of "repentance" or "reflection." This occurs when the robot experiences negative emotions such as "I don't want to experience this feeling again" or "I don't want to be blamed again." The other is a positive emotion on the response side, near the "desire" region. This occurs when there are positive feelings such as "wanting more" or "wanting to know more."

[0657] The emotion-specific model 59 inputs user input into a pre-trained neural network to obtain emotion values ​​representing each emotion shown in the emotion map 400, thereby determining the user's emotion. This neural network is pre-trained based on multiple training data sets, which are combinations of user input and emotion values ​​representing each emotion shown in the emotion map 400. Furthermore, this neural network... Figure 10 As shown in the sentiment graph 900, it was trained in a way that sentiments that are configured close to each other have similar values. Figure 10 The text shows examples of emotions such as "peace of mind", "stability", and "reassurance" that have similar emotion values.

[0658] The above description focuses on the functions of the data processing device 12, but the system of this disclosure is not necessarily installed on a server. The system of this disclosure can also be installed as a general information processing system. This disclosure can also be installed, for example, as a software program running on a personal computer, an application running on a smartphone, etc. The method of this disclosure can also be provided to users in the form of SaaS (Software as a Service).

[0659] In the above embodiments, an example of a specific process being performed by a single computer 22 is given. However, the technology disclosed herein is not limited to this, and the specific process can also be distributed among multiple computers, including computer 22. For example, the data generation model 58 can be located on an external device of the data processing apparatus 12, where data is generated based on the input data.

[0660] In the above embodiments, examples of storing a specific processing program 56 in the memory 32 have been described, but the technology disclosed herein is not limited thereto. For example, the specific processing program 56 may also be stored in a portable computer-readable non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed into the computer 22 of the data processing device 12. The processor 28 executes specific processing according to the specific processing program 56.

[0661] Alternatively, a specific processing program 56 may be pre-stored in a storage device such as a server connected to the data processing device 12 via the network 54, and the specific processing program 56 may be downloaded and installed on the computer 22 according to the requirements of the data processing device 12.

[0662] In addition, it is not necessary to store all the specific processing program 56 in the storage device such as the server connected to the data processing device 12 via the network 54 or in the memory 32; a portion of the specific processing program 56 may be stored in advance.

[0663] As hardware resources for performing specific processes, various processors, as shown below, can be used. For example, a CPU can be listed as a processor, which functions as a general-purpose processor that performs specific processes by executing software, i.e., a program. Furthermore, processors can be listed as special-purpose circuits such as FPGAs (Field-Programmable Gate Arrays), PLDs (Programmable Logic Devices), or ASICs (Application-Specific Integrated Circuits), which are processors with circuitry specifically designed to perform specific processes. Each processor has built-in or connected memory, and each processor executes specific processes using that memory.

[0664] The hardware resources for performing a specific process can consist of one of these various processors, or a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Alternatively, the hardware resources for performing a specific process can be a single processor.

[0665] As an example of a single processor, there are two approaches: First, a processor is composed of a combination of one or more CPUs and software, which functions as a hardware resource to perform a specific process; second, as represented by a SoC (System-on-a-chip), a processor is used to implement the functionality of the entire system, which includes multiple hardware resources for performing a specific process, using a single IC (Integrated Circuit) chip. In this way, the specific process is implemented by using one or more of the aforementioned processors as hardware resources.

[0666] Furthermore, the hardware architecture of these various processors, more specifically, can utilize circuits that combine semiconductor elements and other circuit components. Moreover, the specific process described above is just one example. Therefore, without departing from the main point, unnecessary steps can certainly be deleted, new steps added, or the processing order changed.

[0667] The descriptions and illustrations above are detailed explanations of a portion of the technology disclosed herein, and are merely one example of the technology disclosed herein. For example, the above descriptions of the structure, function, effect, and results are just one example of the structure, function, effect, and results of a portion of the technology disclosed herein. Therefore, without departing from the spirit of the technology disclosed herein, unnecessary parts may be deleted, new elements added, or replacements may be made to the descriptions and illustrations above. Furthermore, to avoid confusion and facilitate understanding of a portion of the technology disclosed herein, explanations of common technical knowledge that do not require special explanation under the premise of being able to implement the technology disclosed herein have been omitted from the descriptions and illustrations above.

[0668] All documents, patent applications and technical specifications set forth in this specification are incorporated herein by reference to the same extent that each document, patent application and technical specification is specifically and individually described therein and referenced by reference.

[0669] In addition, the following notes are provided in response to the above explanation.

[0670] Example 1 (Note 1) An information processing system, characterized in that it comprises: A device for acquiring control information related to disasters and life-related information from an information transmission network based on an instruction communication processing device, and for acquiring information data from a predetermined information source; A device for generating control information for instructing a machine translation processing device to perform translation processing in response to the information data, enabling the machine translation processing device to generate translation data in multiple languages, and storing or formatting the translation data for use by foreign language users. An apparatus for providing attribute data and content data of information sources to an information evaluation and processing device, generating control information to calculate the credibility of each information source based on evaluation benchmark data, and classifying the information sources according to the credibility and classification information. A device for providing a consultation processing device with question data from a user terminal, as well as the translation data and the classified information source related data as parsing object data, generating prompt statements to instruct a generative artificial intelligence model to perform question content parsing and answer text generation, and using the prompt statements to drive the generative artificial intelligence model to generate answer data; A device for extracting display data corresponding to the user's selected language and area of ​​interest from the answer data and the translation data, and sending the display data to the user terminal for output processing; An apparatus for taking input data from a user as the parsing object, generating prompt statements to instruct a generative artificial intelligence model to perform emotional state analysis, and controlling the presentation format or content of the displayed data according to the emotional state. And an apparatus for sending and receiving the prompt statements, the parsed object data, and the evaluation benchmark data between the generative artificial intelligence model and the interface processing device, and for obtaining the parsing result data and generated text data output by the generative artificial intelligence model.

[0671] (Note 2) The information processing system according to Appendix 1 is characterized in that, The system is configured to receive question data sent by a user terminal, select relevant information corresponding to the question data, wherein the relevant information includes at least a portion of the information data, the translation data, and the classified information sources, generate parsing object data with the attached relevant information, and generate a prompt statement containing instructions based on the parsing object data and the question content, which is then input into the generative artificial intelligence model so that the generative artificial intelligence model generates answer data in natural language form for the question content, and output the answer data to the processing device of the user terminal.

[0672] (Note 3) The information processing system according to Appendix 1 is characterized in that, A control processing device for generating prompt statements based on input text data or speech recognition data obtained from a user terminal, instructing the generative artificial intelligence model to perform sentiment classification processing and attitude estimation processing, and dynamically changing the expression level, level of detail, or recommended information source type of the prompt translation data according to the sentiment judgment result output by the generative artificial intelligence model.

[0673] Application Example 1 (Note 1) An information processing system, characterized in that it comprises: Translation instruction means for generating control information for controlling a generative artificial intelligence model in a generative information processing device and inputting the control information into the generative artificial intelligence model to provide translation instructions for the input content; A disaster prevention translation method for parsing disaster-related information obtained from a disaster-related information acquisition interface using control information based on translation instructions for the generative artificial intelligence model and generating multilingual disaster-related information; Information source evaluation means for parsing and performing credibility evaluation and classification of multiple information sources obtained from the information acquisition interface using evaluation instructions for the generative artificial intelligence model; This is used to input first language information obtained from an external information provider interface into a machine translation processing module to convert it into multiple second language information, and to use control information to summarize and adjust the expression of the converted information using editing instructions for the generative artificial intelligence model; When sending information to a terminal device via a communication interface, an individual information selection method is used to select and determine the information to be sent from the multilingual news information and the multilingual disaster-related information based on user attribute information and user behavior history. This is an information distribution method used to automatically send the information to be sent to the terminal device through a notification distribution infrastructure, and to present it in real time using the notification function on the terminal device. A question-and-answer method for generating answer generation control information for a generative artificial intelligence model based on query information from users and information sources evaluated by the aforementioned information source evaluation method, and inputting the answer generation control information into the generative artificial intelligence model to generate answer information; This is an information editing method used to analyze key user interest information and user behavior history to generate interest and concern indicators for each user, and to organize and sort the prompt information based on the interest and concern indicators. This method uses the interest and concern indicators and the information to be prompted as input control information for recommending the generative artificial intelligence model, in order to generate individual recommendation information for each user.

[0674] (Note 2) The information processing system according to Appendix 1 is characterized in that the system further includes: A means for generating responses, which combines natural language query information obtained from the user with relevant information selected by the information source evaluation method to form prompt information for input into the generative artificial intelligence model, and generates a prompt statement containing the prompt information and inputs it into the generative artificial intelligence model to generate response information corresponding to the query content.

[0675] (Note 3) The information processing system according to Appendix 1 is characterized in that the system further includes: An emotion adaptation mechanism for generating sentiment analysis prompts for a generative artificial intelligence model based on text information obtained from the user and usage information obtained from the terminal device, inputting them into the generative artificial intelligence model to infer the user's emotional state, and changing the content or expression of the prompts according to the user's emotional state and the interest and concern indicators.

[0676] Example 2 (Note 1) An information processing system, characterized in that it comprises: A device for retrieving information from the Internet, the device being configured to collect document data from multiple information sources based on input including evaluation criteria and search terms; A device for performing natural language processing on the document data, the device being configured to perform processing including language discrimination, word extraction and evaluation estimation to obtain structured parsing results and generate prompt statements for inputting the parsing results into a generative artificial intelligence model; An apparatus for classifying document data based on the parsing results, the apparatus being configured to divide the document data into multiple categories including life-related information, travel-related information, and food-related information, and to attach identification information as category information and tag information as attribute information to the classification results; A means for storing the classification results into a storage area for retrieval, the means being configured to build an index based on the identification information and the tag information, thereby storing and retrieving the classification results; An apparatus for processing user input from a user terminal, the apparatus being configured to acquire a prompt statement containing user input, parse the prompt statement to extract search conditions representing geographic information, information category and usage conditions, and acquire candidate information from a storage area for retrieval based on the search conditions; An apparatus for generating response data based on the candidate information and the user input, the apparatus being configured to generate a prompt statement for instructing a generative artificial intelligence model to perform response generation processing, and to generate response data provided to the user terminal based on the response text obtained from the generative artificial intelligence model; An apparatus for acquiring information for non-native language users and disaster prevention-related information, the apparatus being configured to parse input content using prompts that instruct a generative artificial intelligence model to perform translation processing, and to obtain translation results through the translation processing; An apparatus for selecting information sources, configured to evaluate the information sources according to an evaluation criterion representing credibility, parse and select the information sources using prompts that instruct a generative artificial intelligence model to parse the evaluation criterion, and store the selection results in association with the classification results.

[0677] (Note 2) According to the information processing system described in Appendix 1, the device for processing user input from a user terminal is further configured to: extract the question object, information category, and constraints from the user input containing question content, and generate a prompt statement to instruct a generative artificial intelligence model to generate answer text based on candidate information obtained from the storage area for retrieval and the question content, thereby providing the answer text output by the generative artificial intelligence model to the user terminal.

[0678] (Note 3) According to the information processing system described in Appendix 1, the device for processing user input from a user terminal is further configured to: estimate the user's emotional state by parsing the expressive content in the user input, select information to be prompted based on the emotional state and the classification result, generate a prompt statement for instructing a generative artificial intelligence model to generate response text based on the emotional state and the information to be prompted, thereby prompting the user with the response text.

[0679] Application Example 2 (Note 1) An information processing system, characterized in that it comprises: This is a means of parsing and translating multilingual text information using prompts for generative artificial intelligence models to generate translations for foreign language users. Translation methods are used to parse and translate disaster-related information using prompts for generative artificial intelligence models in order to generate disaster prevention information. Editing techniques for parsing evaluation benchmarks of multiple information sources using prompts for generative artificial intelligence models, parsing multiple information sources, selecting them based on credibility, and generating categorized information. This tool is used to retrieve surrounding facility data based on location data obtained from a location acquisition device and user attribute data, and to calculate the suitability of each facility data using a machine learning model. Based on the classification information, the facility data is used as a selection method for choosing between sightseeing information or lifestyle information. Used to generate structured data containing the selected sightseeing or lifestyle information, and based on the structured data and user preference data to generate prompt statements for the generative artificial intelligence model to instruct the generation of explanatory text, and the explanatory generation means for generating natural language explanatory information by the generative artificial intelligence model. An emotion estimation method for inferring a user’s emotional state based on facial expression data and sound data obtained from an image acquisition device and a sound acquisition device. Control means for adjusting the selection criteria for sightseeing information or life information in the selection means and the expression mode of explanatory information in the explanatory generation means based on the presumed emotional state. Processing means for receiving user input, location data, facial expression data, and voice data sent from a terminal device, and comprehensively executing the translation means, editing means, selection means, description generation means, and emotion estimation means, and sending the output information to the terminal device.

[0680] (Note 2) The information processing system according to Appendix 1 is characterized in that it further includes: This is a question-and-answer method used to generate prompts for a generative artificial intelligence model, instructing the parsing of question content and the generation of answer data, wherein the generative artificial intelligence model generates answer data for the question.

[0681] (Note 3) The information processing system according to Appendix 1 is characterized in that it further includes: This is used to generate prompts for a generative artificial intelligence model, indicating an emotion adaptation method that determines the information content and expression corresponding to the emotion based on the input text from the user and the emotion state inferred by the emotion inference means, and generates emotion adaptation output information by the generative artificial intelligence model.

Claims

1. An information processing system, characterized in that, include: processor; The processor is configured as follows: The input content is parsed using prompt words that give instructions to the generative AI model to perform translation, in order to generate information for foreigners and perform the corresponding translation process. The input content is parsed using prompt words that give instructions to the generative artificial intelligence model to perform disaster prevention information translation, so as to generate disaster prevention information and perform corresponding translation processing. Information sources are parsed using prompt words that provide evaluation criteria for parsing information sources to generative artificial intelligence models, in order to select and classify information sources with high credibility.

2. The information processing system according to claim 1, characterized in that, The processor is also configured to: The input question is parsed using prompt words that instruct the generative AI model to parse the question content and generate an appropriate answer.

3. The information processing system according to claim 1, characterized in that, The processor is also configured to: Using prompt words that give instructions to the generative artificial intelligence model to parse user input, the model performs sentiment analysis on the user input and provides information corresponding to the sentiment based on the parsed user sentiment.

Images