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: receive information through an interface for receiving information from a user; analyze the received information using a generative artificial intelligence model to evaluate legal concerns, security impacts, and customer reactions related to the information; generate prompt information for sorting the information to the appropriate department based on the analysis results; and analyze the emotional state of the user using sentiment analysis technology, and generate prompt information for setting the priority of the information based on the emotional state.

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] In existing technologies, when enterprises or organizations receive large amounts of information from users (such as inquiries, complaints, feedback, business requests, etc.), they typically rely on manual methods to read, understand, and sort the information before transferring it to legal, security, customer service, or other relevant departments for processing. This manual sorting method has the following problems: First, manual judgment of information content is time-consuming and inefficient, easily leading to processing backlogs when information volume surges, resulting in delayed responses to user requests. Second, manual sorting heavily relies on the knowledge and experience of staff, making it prone to inaccurate sorting due to subjective judgment differences, causing information to fail to reach the truly capable departments in a timely manner. Third, existing sorting processes often do not fully consider the user's emotional state, failing to dynamically adjust information priority based on user anxiety, anger, or urgency, potentially causing important or urgent issues to be neglected, thus reducing user satisfaction. Furthermore, some existing rule-based automated sorting systems have limited capabilities in handling complex natural language expressions, implicit semantics, and multi-dimensional factors (such as legal risks, security impacts, and potential customer reactions), making it difficult to provide timely and accurate multi-dimensional evaluations of information. Therefore, there is an urgent need for a system that can use generative artificial intelligence models and sentiment analysis technology to intelligently analyze user information, automatically evaluate legal concerns, security impacts and customer reactions related to the information, and rationally set processing priorities based on the user's emotional state, thereby improving the efficiency and accuracy of information sorting and optimizing the user experience. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides an information processing system comprising a processor configured to: receive information via an interface for receiving information from a user; parse the received information using a generative artificial intelligence model to identify key elements in the information and evaluate legal concerns, security impacts, and customer reactions related to the information; generate prompts based on the parsing results to sort the information to appropriate departments, enabling the system to automatically or semi-automatically route information to legal, security, customer service, or other relevant business departments; and analyze the user's emotional state using sentiment analysis technology to identify whether the user is in a state of anger, anxiety, urgency, or other emotional states, generating prompts based on the emotional state to set the priority of the information, enabling the system to assign different processing priorities to different information, thereby prioritizing information with high emotional tension or urgency. Specifically, the processor can further utilize a generative artificial intelligence model to generate prompts based on the information content to automatically determine the relevant department, thereby automatically mapping information to the corresponding business department through semantic understanding of natural language content without the need for pre-defined fixed rules. Furthermore, the processor can generate prompts based on the user's emotional state to prioritize the sorting of information. This allows for multi-dimensional intelligent sorting and scheduling, taking into account both business attributes and emotional urgency during information sorting and routing. Through these technical means, the present invention enables automatic parsing of user information, multi-dimensional risk and impact assessment, and priority setting based on emotional state, effectively improving information processing efficiency and accuracy, reducing manual workload, and enhancing user experience.

[0005] "System" refers to a whole consisting of at least one processor and connected hardware and / or software components such as memory and communication interfaces, used to perform functions such as information reception, parsing, evaluation, prompt generation and information sorting.

[0006] A processor is an electronic component that can execute predetermined program instructions, perform calculations on input data, and output processing results. It can be a single physical processor, a multi-core processor, a processor cluster, or any combination including CPU, GPU, NPU, etc.

[0007] An "interface" refers to a software or hardware module used for data exchange between a user and the system or between the system and an external device, including but not limited to graphical user interfaces, application programming interfaces, network communication interfaces, message queue interfaces, etc. Data related to user information can be received and / or sent through the interface.

[0008] "User" refers to an individual or organizational entity that submits information, requests services, or interacts with the system through a terminal or other means, including end consumers, corporate clients, internal employees, and operational personnel.

[0009] "Information" refers to various types of content data submitted by users to the system through the interface, including but not limited to natural language or structured data in the form of inquiries, complaints, business requests, feedback, text descriptions, and attachment descriptions.

[0010] "Generative artificial intelligence models" refer to artificial intelligence models that are trained on large-scale data and can automatically generate text, vector representations or other output results based on input content, including but not limited to large language models, text generation models, and multimodal generation models.

[0011] "Parsing" refers to the process by which a processor analyzes and processes received information or related data to extract key elements, identify semantic relationships, and generate structured or semi-structured representations. This includes operations such as word segmentation, semantic understanding, entity recognition, intent recognition, and category determination.

[0012] "Legal concerns" refer to situations that, after analyzing information content, may constitute illegality, violation, or compliance risks based on applicable laws, regulations, and regulatory requirements. These include, but are not limited to, privacy protection risks, data compliance risks, and contract risks.

[0013] "Security impact" refers to risks or impacts related to information content that may arise at the level of information security or system security, including but not limited to security issues such as data leakage risk, unauthorized access risk, insufficient encryption measures, and lack of operational auditing.

[0014] "Customer reaction" refers to the attitudes, evaluations, and behavioral tendencies that potential or existing customers may have regarding the services, products, or processing methods involved in the information, including expected reactions such as positive acceptance, doubt, aversion, and dissatisfaction.

[0015] "Prompt information" refers to structured or unstructured data content automatically generated by the processor based on the parsing results, used to guide subsequent processing operations. This includes instruction information for departmental sorting, tagging information for priority setting, and auxiliary information for routing decisions.

[0016] "Sorting" refers to the process of classifying and allocating received information to one or more target objects according to predetermined rules or automatic judgment results based on prompt information or parsing results. In this invention, it mainly refers to dividing information and transferring it to the corresponding business departments or processing units.

[0017] "Appropriate department" refers to the business department or functional department that is determined to be the most capable or responsible for processing the information based on factors such as the information content, legal concerns, security impact, and customer feedback. Examples include legal departments, security departments, customer service departments, sales departments, and operations departments.

[0018] "Sentiment analysis technology" refers to the technology used to identify a user's emotional state from the information content, tone features, word intensity, or other relevant data submitted by the user, including but not limited to sentiment classification models and sentiment intensity assessment models based on machine learning or deep learning.

[0019] "Emotional state" refers to the result obtained after identifying the user's current emotional characteristics through sentiment analysis technology, including but not limited to emotional types such as anger, anxiety, tension, dissatisfaction, calmness, and positivity, as well as the corresponding emotional intensity.

[0020] "Priority" refers to the level or sorting mark used to indicate the order and importance of processing each piece of information when the system processes and schedules multiple pieces of information. It can be multi-level or a continuous value, and is used to control the processing order of information in the queue.

[0021] "Priority sorting" refers to the process of classifying, routing, and processing high-priority information by comprehensively considering its content, emotional state, and other evaluation results during information sorting, in order to ensure that urgent or important information receives a faster response. Attached Figure Description

[0022] Figure 1 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the first embodiment.

[0023] 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.

[0024] Figure 3 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the second embodiment.

[0025] 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.

[0026] Figure 5 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the third embodiment.

[0027] Figure 6 This is a conceptual diagram illustrating an example of the main functions of the data processing apparatus and head-mounted terminal according to the third embodiment.

[0028] Figure 7This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the fourth embodiment.

[0029] 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.

[0030] Figure 9 This represents an emotion map that maps multiple emotions.

[0031] Figure 10 This represents an emotion map that maps multiple emotions.

[0032] Figure 11 This is a sequence diagram illustrating the processing flow of the data processing system of the first embodiment.

[0033] Figure 12 This is a sequence diagram illustrating the processing flow of the data processing system in Application Example 1.

[0034] Figure 13 This is a sequence diagram illustrating the processing flow of the data processing system of the second embodiment.

[0035] Figure 14 This is a sequence diagram illustrating the processing flow of the data processing system in Application Example 2. Detailed Implementation

[0036] Hereinafter, an example of an implementation of the system to which the technology of this disclosure relates will be described with reference to the accompanying drawings.

[0037] First, let me explain the terminology used in the following instructions.

[0038] 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.

[0039] 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.

[0040] 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.

[0041] 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.

[0042] 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.

[0043] First Implementation Method Figure 1 An example of the configuration of the data processing system 10 according to the first embodiment is shown.

[0044] 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.

[0045] 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).

[0046] 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.

[0047] 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.

[0048] 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.

[0049] 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.

[0050] Figure 2 The diagram shows an example of the main functions of the data processing device 12 and the smart device 14.

[0051] 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.

[0052] 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).

[0053] 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.

[0054] 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.

[0055] 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."

[0056] Traditional case drafting and risk assessment largely rely on manual retrieval of regulations, information organization, and text writing, which presents the following technical problems: First, the information processing workflow is fragmented, with information processing devices used only as simple storage and display tools. This prevents the formation of an end-to-end automated pipeline within a single computer system, from data acquisition and text parsing to content generation, resulting in low efficiency in utilizing computing resources. Second, natural language processing technology and generative artificial intelligence models are not deeply integrated at the system level. There is a lack of dynamic prompting mechanisms based on structured features and risk assessment results, meaning the input to the generative model is only coarse-grained text, failing to fully reflect legal requirements, information protection requirements, and... The quality of model output is limited by multiple factors, including customer feedback; thirdly, the lack of a management and automatic selection mechanism for prompt statement templates for generative artificial intelligence models makes it difficult for existing systems to adjust the model input structure and output format in a timely manner according to the type of case and background points, resulting in the computer's inability to stably generate structured and approvable draft results; fourthly, the lack of a closed-loop data flow between user revisions and automatic system generation means that the information processing device cannot incorporate the user's corrections and additions on the terminal into the calculation process, thus failing to optimize the subsequent prompt statement construction logic based on the revision history, limiting the system's ability to adapt to model calling strategies and text structures in long-term operation.

[0057] In summary, how to construct a system at the computer technology level that can automatically collect heterogeneous case-related data and parse it using natural language, dynamically generate prompts based on structured evaluation results and call generative artificial intelligence models, and form a closed loop with user interaction, in order to improve the processing efficiency and output quality of information processing devices in case drafting scenarios, has become a technical challenge that needs to be addressed.

[0058] 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.

[0059] In this invention, the server includes: a processing unit for receiving purpose information and summary information about business matters and generating business matter identification information and business matter categories; a processing unit for retrieving historical business matter data and rule data from a storage unit based on the identification information and the categories, and automatically acquiring regulatory data and market data from external data sources, integrating multi-source data into business matter-related data; a processing unit for performing natural language processing operations such as segmentation, feature extraction, entity recognition, and relation extraction on the text data in the business matter-related data to generate structured data representing legal requirements, information protection requirements, and customer response requirements; and a processing unit for performing risk assessment and advantage assessment on the structured data to generate assessment data that distinguishes between risk factors and advantage factors and includes weight information. The system comprises: a processing unit for automatically selecting a target template from multiple prompt statement templates based on the business matter category and the structured data, filling the evaluation data and the structured data into the template, and constructing prompt statements and additional prompt statements containing elements to be emphasized and elements to be simplified; a processing unit for inputting the prompt statements and the additional prompt statements into a generative artificial intelligence model and obtaining natural language text containing the draft content of the business matter; a processing unit for associating and storing the draft content with business matter identification information and outputting it to the terminal through a communication unit; and a processing unit for receiving user modification information and append information returned by the terminal, merging and updating the draft content to form the final draft information, and associating and storing the update record with the business matter identification information. This enables a complete processing flow within a single computer system, from the automatic collection and analysis of case-related data, the generation of dynamic prompts based on structured evaluation results, to the invocation of generative artificial intelligence models to generate draft content and combine it with user modifications to form a closed-loop optimization. This improves the automation level and computing resource utilization efficiency of the server in complex text processing and case drafting tasks, enhances the structure and relevance of natural language generation results, and achieves performance and functional improvements to information processing technology itself.

[0060] "Information processing device" refers to a computer system consisting of one or more processing units, storage units, and communication units, used to execute program instructions and process input data related to business matters, such as receiving, retrieving, parsing, evaluating, generating, and storing such data.

[0061] "Terminal" refers to a user-side electronic device that interacts with an information processing device through a communication network, including but not limited to mobile communication devices, portable computing devices, or desktop computing devices, and is used to display drafted content and receive user input.

[0062] "Business matters" refer to objects that need to be analyzed and drafted in information processing devices, including projects, activities, services, or operational plans, which typically have attributes such as predetermined objectives, applicable rules, and related risks and opportunities.

[0063] "Business Item Identification Information" refers to identification data used to uniquely identify a business item in an information processing device, including but not limited to identification numbers, codes, or other identifiers that can be used to distinguish different business items.

[0064] "Business Item Category" refers to the category information obtained by classifying business items based on their type, domain, or application scenario. This category data is used to select the appropriate data processing rules and prompt statement templates.

[0065] "Historical business information" refers to structured or unstructured data related to previously processed business matters that are pre-stored in the storage device of the information processing device, including information such as previous draft content, evaluation results, and processing records.

[0066] "Rule information" refers to rule data related to business processing that is pre-stored in the information processing device, including business processing rules, internal management regulations, compliance guidelines, and assessment standards.

[0067] "Regulatory information" refers to the text or data of external management requirements or binding regulations related to business matters, including information such as laws, regulations, policies, regulatory guidelines and industry standards.

[0068] "Market information" refers to external data that reflects the target market environment, competitive situation, and customer behavior, including market reports, statistical data, competitor information, and user feedback.

[0069] "Business Item Related Information" refers to a data set associated with a specific business item, formed by an information processing device integrating multi-source data such as historical business item information, rule information, regulatory information, and market information based on business item identification information and business item category.

[0070] Natural Language Processing (NLP) technology refers to the techniques used to automatically analyze and process natural language text on computers, including methods such as word segmentation, sentence segmentation, part-of-speech tagging, entity recognition, relation extraction, keyword extraction, and semantic analysis.

[0071] "Structured information" refers to standardized data generated by information processing devices based on the results of natural language processing, represented by predefined fields or data structures, used to describe elements such as legal requirements, information protection requirements, and customer response requirements.

[0072] "Legal requirements" refer to the binding requirements extracted from regulatory and rule information that apply to business matters, including applicable laws, regulations, policies, corresponding obligations, prohibitions, and responsibilities.

[0073] "Information protection requirements" refer to the protection conditions and control measures related to data processing activities in business matters, including requirements for data confidentiality, integrity and availability, as well as corresponding security control strategies.

[0074] "Customer feedback requirements" refer to the requirements that should be considered in relation to business matters, including customer attitudes, evaluations, and behavioral tendencies, such as satisfaction with products or services, concerns, and expectations.

[0075] "Risk factors" refer to factors represented in structured and assessment information that may have an adverse impact on business matters, including compliance risks, information security risks, operational risks, and reputational risks.

[0076] “Advantageous factors” refer to factors represented in structured and evaluative information that may bring benefits or advantages to business matters, including market opportunities, competitive advantages, cost savings, and increased user recognition.

[0077] "Assessment information" refers to data generated by an information processing device after analyzing risk and advantage factors based on structured information. This data includes classification results and importance information and is used to reflect the overall risk and advantage status of a business matter.

[0078] "Prompt statement template" refers to a template-based data structure that is pre-stored in an information processing device and used to construct the input text for generative artificial intelligence models. It contains preset text frames and fillable placeholders for different business item categories.

[0079] "Prompt statements" refer to natural language text generated by an information processing device based on a prompt statement template, which is then filled with evaluation information and structured information and used as input to a generative artificial intelligence model to instruct the model to generate draft results that conform to the expected structure and content.

[0080] "Additional prompts" refer to supplementary natural language text generated by the information processing device based on the weights of risk and advantage factors in the assessment information, used to further instruct generative artificial intelligence models on the elements that should be emphasized or simplified in their description.

[0081] "Generative artificial intelligence model" refers to a content generation model trained based on machine learning technology that can be invoked by an information processing device. It is an artificial intelligence model that automatically generates natural language text or other forms of output based on input prompts and additional prompts.

[0082] "Draft content for business matters" refers to the natural language text content output by the generative artificial intelligence model after receiving prompts and additional prompts, which is used as a draft for business matter description, plan or report.

[0083] "Final draft information" refers to the final draft text information that is confirmed and stored by the information processing device after being integrated and updated with the modification and supplementary information entered by the user on the terminal, based on the draft content of the business matter, and used for business decision-making or approval.

[0084] "Update history" refers to the change history data generated and saved by the information processing device during the process of modifying and updating the draft content of business matters. It is used to represent the differences between the draft content of each version, the update time, and the corresponding user information.

[0085] In an embodiment of this invention, the server, as the core node of the information processing device, runs a case drafting support program, while the terminal, as a user interaction device, runs a display and editing interface. The user provides business-related input to the server through the terminal. The system of this invention can be deployed in a computing environment consisting of one or more physical servers. These servers can be rack-mounted servers or cloud computing instances, equipped with general-purpose processors (e.g., x86_64 CPUs), main memory (RAM), non-volatile memory (SSD), and network interface devices. The basic software environment on which the server runs may include an operating system (e.g., a Unix-like operating system), a database management system (e.g., relational database management software), application server software, and a runtime environment for natural language processing (e.g., a Python interpreter and its dependent libraries). The server executes the program instructions described in this invention within this environment, realizing functions such as data acquisition, natural language parsing, structured information generation, prompt statement construction, and generative artificial intelligence model invocation.

[0086] In implementing this invention, the server first stores a set of program modules for case processing in a storage medium. These modules include a data acquisition module, a text parsing module, a structured feature generation module, a risk and advantage assessment module, a prompt statement generation module, a generative artificial intelligence model interface module, a result management module, and a user revision management module. These modules are stored as executable instructions in non-volatile memory and are loaded into main memory for execution by the processor at runtime. The server maintains multiple logical data tables through a database management system, including a main business item table, a historical business item table, a rule information table, a regulatory information table, a market information table, a structured feature table, an assessment information table, a prompt statement template table, a draft content table, and a revision history table. Each data table includes predefined fields and indexes to efficiently support query, update, and correlation operations.

[0087] When performing text parsing, the server utilizes natural language processing (NLP) software libraries to perform specific data processing and operations. For example, the server loads word segmentation and syntactic analysis libraries within its Python runtime environment to perform sentence segmentation, word segmentation, part-of-speech tagging, named entity recognition, and dependency analysis on regulatory texts, policy interpretation documents, market reports, and customer review texts obtained from internal databases and external interfaces. The server constructs a sequence of lexical units for each sentence and attaches part-of-speech tags, syntactic dependency relationship tags, and entity type tags to each lexical unit. Simultaneously, based on predefined rules, the server marks sentences containing keywords such as "prohibited," "must," "ordered," and "fine," and those deemed to contain binding predicates and responsible entities, as candidate legal requirement sentences, thus providing input for subsequent risk and compliance point extraction.

[0088] The server employs a specific data structure when generating structured features: it represents each text record as a structure containing multiple fields, including text ID, original text, sentence list, entity list, triple list, and sentiment score. In the entity list, the server records entity types (e.g., regulation name, regulatory body, data category, customer group), the entity's position index in the original text, and its relationships with other entities; in the triple list, it records relationships consisting of subject, predicate, and object (e.g., "Service provider—must—not collect sensitive information without authorization") to further deduce legal obligations and potential risk points. The server can also incorporate statistical features, such as the frequency of a legal clause being cited in different documents or the proportion of a certain type of risk-related statement appearing in historical cases, appended numerically to the structured features.

[0089] When assessing risk and advantage factors, the server does not simply label them, but performs a specific numerical calculation process. Based on entity types and triple patterns in the structured features, the server applies a predefined set of rules to assign a base risk score or advantage score to each factor. For example, factors involving "cross-border transfer of sensitive personal information" are given a higher base risk weight, while factors such as "increased user trust" and "enhanced market differentiation capabilities" are given advantage scores. The server then adjusts the weights of each factor in the current case based on the assessment results of archived cases in the historical business transaction table, using statistical calculations (such as weighted averages or linear normalization) to ensure that the weight distribution reflects a combination of historical experience and the current context. Finally, the server represents each factor as a record containing fields such as "factor type," "source text ID," "base weight," "adjusted weight," and "related legal provisions or market indicators," storing this record in the assessment information table.

[0090] The server employs an explicit data-driven template mechanism when constructing prompt statements. It pre-stores multiple templates in a prompt statement template table, each represented as a text paragraph and containing multiple placeholders for inserting basic business information, legal requirements summaries, information protection points, customer feedback points, and a list of risks and advantages. The server selects one or more suitable templates based on the business category field and calls string processing functions to embed the top-ranked elements from the structured features and evaluation information into the corresponding placeholder positions, generating a complete natural language prompt statement. The server also generates additional prompt statements based on the weights in the evaluation information, explicitly instructing the generative AI model on the elements to be emphasized and those that can be simplified when generating draft content. For example, an additional prompt statement might look like this: "When drafting the proposal, please focus on a detailed analysis of compliance requirements involving the handling of sensitive personal information and cross-border transfers; general market opportunities can be briefly summarized." In one implementation, the server can generate the following example of a notification statement for a new product launch scenario: "You are a legal counsel responsible for compliance and business strategy. Please draft a comprehensive case proposal for the company regarding the 'launch of a new product in the Chinese market' based on the following information."

[0091] The aim of the case is to increase the acceptance of new products among the target customer group and maximize profit margins while ensuring compliance and data security.

[0092] Relevant laws and regulations include, but are not limited to, those concerning personal information protection, data security, consumer rights protection, and administrative regulations and local regulatory provisions applicable to the industry.

[0093] Information protection impacts the types of personal information this product will collect and process, including contact information, location information, and browsing history. Please analyze the potential impact on confidentiality, integrity, and availability, and propose technical and organizational measures to mitigate the risks.

[0094] Customer feedback and market environment: Referencing customer reviews and market share changes of similar products over the past year, identify key complaints and customer expectations, and assess the intensity of market competition.

[0095] The indictment you provide should include at least the following: 1. Case background and objective overview; 2. Analysis of main applicable laws and compliance requirements; 3. Information protection impact assessment (including risk classification and control measures); 4. Analysis of the target customer group's response and market trends; 5. Implementation recommendations and timeline after comprehensively considering legal, information security, and market factors. In another implementation, the server can generate the following example of a prompt statement for internal system modification scenarios: "Please write a proposal for the company's 'Customer Data Management System Upgrade Project' from the perspectives of both the information security compliance officer and the business manager, based on the following background:" - Project objectives: Improve query efficiency, add customer profiling and analysis functions, and comply with personal information protection and cross-border data transfer requirements; - Current problems: Slow system response, lax access control, and incomplete logging; - Regulatory environment: Regulatory authorities have recently tightened requirements for data minimization, access control, and audit trails.

[0096] Please emphasize the following: (1) Applicable legal provisions and regulatory requirements; (2) The positive and negative impacts of the upgrade on information protection; (3) Potential 'dangers' (such as compliance penalties, data breaches, and damage to customer trust) and corresponding mitigation measures; (4) Designs that can help the project gain 'advantages' (such as enhancing customer trust and supporting new business models).

[0097] Please use structured sections in your writing. The server uses explicit data interfaces and parameter configurations to interact with the generative AI model. It communicates with external model services via a network interface, employing an HTTP-based application layer protocol and transmitting requests and responses through encrypted connections. Each request includes a model identifier (e.g., the name of a text generation model), prompt text, additional prompts, and a set of control parameters. These parameters include a temperature parameter to control the diversity of generated content, a maximum number of tags to limit output length, and metadata specifying the output language and format. The server uses these parameter combinations to influence the probability distribution sampling behavior within the neural network, thereby altering the style and level of detail of the generated results without modifying the model weights.

[0098] In one embodiment, the server can employ a generative artificial intelligence model based on a transformer architecture. During the training phase, this model performs self-supervised learning on a large-scale corpus using a multi-layer self-attention network, optimizing its parameters using masked language modeling or autoregressive prediction tasks. The model internally contains multiple layers of encoding or decoding units, each layer including a multi-head self-attention sublayer and a feedforward neural network sublayer, connected by residual connections and normalization operations. During training, the model's weights are updated using gradient descent-like algorithms, and the error function can be a cross-entropy loss function, calculated from the difference between the predicted probability distribution of the next lexical unit and the true lexical unit distribution. In this embodiment, the server does not retrain the model during runtime; instead, it utilizes the trained weights and organizes the neural network's inference process by changing the input prompts and control parameters. However, because the server explicitly encodes structured features and weighted evaluation information into the prompts and highlights key elements through additional prompts, the neural network, when internally calculating attention to these markers, allocates more attention to lexical units corresponding to high-weight elements, thereby improving the coverage and accuracy of the output content regarding risk and compliance points.

[0099] The server's mechanism for generating prompts guided by structured features and weights significantly improves the efficiency of computational resource utilization and the relevance of output compared to simply inputting raw text into the model. Internally, the server first compresses, abstracts, and reorganizes large-scale text, ensuring that the generative AI model receives high-density prompts instead of redundant raw data. This preprocessing reduces the length of context that the model needs to consider during subsequent inference, while increasing the proportion of key information within the context. This reduces the ineffective overhead of attention calculations, improves inference speed, and lowers response time.

[0100] In this embodiment of the invention, the terminal performs interactive display and editing functions. The terminal can be a smart mobile device, portable computing device, or fixed computing device, equipped with a display screen, input device, and network communication module. After receiving the draft content returned by the server, the terminal renders the content into the display interface by paragraphs or sections, providing an editing area at the corresponding position of each section. When the user inputs modified text, the terminal records the changed content and its location, and sends the modification information along with business item identification information to the server via the network. In some embodiments, the terminal can locally buffer user editing behavior to reduce the communication load caused by frequent network requests, and report a batch of modification records at once when the user triggers a save or submit operation, thereby reducing the number of server-side processing operations and improving the overall system communication efficiency.

[0101] In this embodiment of the invention, the user is responsible for reviewing and revising the draft content generated by the server. The user checks the text segment by segment on the terminal and, based on professional judgment, corrects inaccurate statements, supplements missing operational procedures, or adds enterprise-specific control measures. The user's modifications are recorded by the server as a revision history. In subsequent embodiments, the server can perform statistical analysis on these revision histories, such as identifying which risk factors frequently need to be strengthened or weakened in the generated results, thereby automatically adjusting the weight or prompting method of the corresponding elements in future prompt generation processes. This adaptive strategy based on runtime feedback enables the server to continuously optimize the way it calls the generative artificial intelligence model, thereby achieving long-term improvement in the quality of natural language generation at the system level.

[0102] In various implementations, servers can optimize data storage structures and access patterns to improve overall processing performance. For example, servers can create composite indexes for frequently accessed structured feature tables and evaluation information tables, and establish partitioned tables for different business transaction categories to accelerate query and update operations. Servers can maintain a cache of some frequently accessed data in memory, such as the structured features and evaluation information of recently processed business transactions, to reduce latency caused by repeated disk reads. Servers can also distribute some computationally intensive tasks, such as large-scale text segmentation and syntactic analysis, to multi-processor or multi-node clusters for parallel execution via task queues to further improve system throughput.

[0103] The server, constructed using the aforementioned technologies, extends this invention beyond mere automation at the business process level, bringing substantial improvements to the computer's internal data structure design, algorithm flow, and model invocation strategies. By mapping natural language text to structured features, guiding neural network inference through a weighted prompt generation mechanism, and adjusting prompt generation rules in reverse based on user revision history, the server optimizes the computational steps and improves information utilization efficiency for complex language processing tasks. This optimization not only enhances the accuracy and completeness of drafted content but also delivers quantifiable technical benefits in processing speed, storage utilization, and network communication, demonstrating an improvement in computer technology itself.

[0104] use Figure 11 The processing flow is explained.

[0105] Step 1: The server receives basic information about the business transaction and generates identification information for the business transaction.

[0106] Input: Basic information about the business transaction sent by the terminal over the network, including fields such as business transaction name, business objective, business lines involved, expected implementation time, and overview of related products or services.

[0107] The server parses the request message sent by the terminal, reads each field from the JSON or form data, performs non-empty validation on required fields, format validation on time fields, and upper limit checks on text length. After passing the validation, the server calls the database sequence or identifier generation function to generate unique business item identifier information, and writes the basic information and this identifier information into the business item master table.

[0108] Output: The main record of the business transaction written to the database and the business transaction identifier information returned to the terminal.

[0109] Step 2: The server determines the category of the business transaction and records the classification result.

[0110] Input: Basic information and business item identification information of the business item stored in step 1.

[0111] Based on fields such as business objectives, business line, and product type, the server applies predefined classification rules or a simple multi-label classification algorithm to divide business items into one or more preset categories (e.g., "new product launch," "system upgrade," "marketing campaign," etc.). The server writes the determined business item category into the corresponding field of the main business item table and caches the category information in memory for use by subsequent processing modules.

[0112] Output: The main record of the business item, including the business item category field.

[0113] Step 3: The server retrieves internal historical business transaction information and rule information.

[0114] Input: Business item identification information and business item category.

[0115] The server performs a query operation on the historical business item table in the database, using the business item category as the primary key or index condition to retrieve historical records that are the same as or similar to the current business item category. Simultaneously, it reads the internal processing rules, enterprise standards, and compliance guidelines associated with that category from the rule information table. The server merges the query results to form an internal data subset, which is then stored in memory as a structure or list.

[0116] Output: An internal data set containing historical business transaction records and rule document content.

[0117] Step 4: The server obtains regulatory and market information from external data sources.

[0118] Input: Keywords extracted from business item categories and internal rule information (such as related industries, target regions, data types, etc.).

[0119] The server sends HTTP requests to one or more external data sources (such as regulatory information services and market data services) via a network interface. The request parameters include the business category, industry name, region name, and data processing-related keywords. The server receives JSON or HTML data from external data sources, including regulatory texts, policy interpretations, industry analysis reports, and customer feedback summaries. It uses a parser to extract the main text from the HTML page, reads the fields from the JSON structure, and organizes all text and numerical information into a unified data object format.

[0120] Output: An external dataset containing regulatory information text, market information text, and numerical indicators.

[0121] Step 5: The server integrates data from multiple sources to form information related to business transactions.

[0122] Input: The internal data set obtained in step 3 and the external data set obtained in step 4.

[0123] The server merges historical business transaction records, rule documents, regulatory texts, market reports, and customer review texts using a unified data structure. For example, the server assigns a document ID to each document, records its source type (internal / external), language, timestamp, and scope of application. During the integration process, the server removes duplicate or highly similar documents, deletes empty or incorrectly formatted entries, and packages all the organized records into a business transaction association information object associated with the business transaction identifier information, and writes it to the business transaction association information table.

[0124] Output: Records of business transaction association information stored in the database.

[0125] Step 6: The server performs natural language preprocessing on the information related to business transactions.

[0126] Input: Text data fields in business-related information (legal provisions, rule document paragraphs, market report text, customer reviews, etc.).

[0127] The server invokes sentence segmentation and word segmentation functions in the natural language processing engine to divide each long text into sentences and perform word segmentation on each sentence, generating a sequence of terms. The server records the document ID, sentence number, original text, word list, and language tag for each sentence. The server also performs basic text cleaning operations, such as removing extra whitespace, standardizing punctuation, and removing HTML tags.

[0128] Output: A collection of intermediate text data containing sentence segmentation and word segmentation results.

[0129] Step 7: The server performs feature extraction and structured encoding on the preprocessed text.

[0130] Input: The intermediate text data set generated in step 6.

[0131] The server invokes part-of-speech tagging and named entity recognition modules to tag the words in each sentence with part-of-speech tags, identifying named entities such as regulatory names, regulatory agencies, data types, business entities, amounts, and geographical locations. The server further uses dependency parsing tools to construct a dependency tree for each sentence, extracting semantic triples composed of subject, predicate, and object, such as "service provider—shall—take encryption measures." Through rule matching and dictionary retrieval, the server marks sentences containing binding predicates such as "must," "must not," "shall," and "prohibited" as legal requirement candidates; sentences involving terms such as "data collection," "data storage," "access control," and "log recording" as information protection candidates; and sentences containing sentiment words such as "satisfaction," "complaint," "concern," and "recommendation" as customer response candidates. The server encodes these candidate results, along with their entities, triples, and other information, into structured records and writes them into a structured feature table.

[0132] Output: A structured set of candidate legal requirements, information protection points, and customer response points.

[0133] Step 8: Server-side risk and advantage assessment of structured features.

[0134] Input: A set of structured features and past evaluation results recorded in the historical business transaction table.

[0135] The server assigns a basic risk score or a basic advantage score to each structured feature according to preset rules. For legal requirement candidates, the server sets a higher or lower basic risk weight based on information such as whether it involves mandatory obligations and the severity of penalties; for information protection candidates, the server sets different levels of risk weight based on the sensitivity of the data type and the scope of the handling behavior; for customer reaction candidates, the server sets an advantage or risk weight based on emotional polarity and frequency of occurrence. The server then reads the statistical distribution of element weights under similar situations from the historical business transaction table, and uses a weighted average or linear normalization algorithm to adjust the weight of the current case, thereby generating the final weight. The server writes the type, basic weight, adjusted weight, associated text ID, and source category of each element into the evaluation information table.

[0136] Output: A set of assessment information containing risk factors and favorable factors and their weights.

[0137] Step 9: The server selects a prompt statement template and fills in the structured points.

[0138] Input: Business item category, set of structured features, and set of evaluation information.

[0139] The server searches the prompt statement template table for one or more template records corresponding to the business matter category, reads the template text, which contains placeholder tags for inserting information. Based on the weights of elements in the structured feature set and the evaluation information set, the server selects the highest-weighted legal requirement items, information protection items, and customer response items, and maps them to placeholders in the template in a predefined order. Using string replacement or a template engine mechanism, the server populates the template with basic business matter information, summaries of important regulatory items, key information protection risks and control measures, major customer concerns, and market opportunities, generating a complete prompt statement text.

[0140] Output: The main prompt statement for generative artificial intelligence models.

[0141] Step 10: The server generates additional prompts based on the element weights.

[0142] Input: The adjusted weights of each element in the evaluation information set.

[0143] The server defines weight thresholds and sorting rules, selecting the highest-weighted set of elements from the evaluation information set as the key descriptive elements and the lower-weighted set as elements that can be described briefly. Based on these elements, the server constructs supplementary natural language text, clearly indicating the content that should be elaborated upon and the content that can be briefly summarized, such as "Please explain in detail the requirements related to sensitive personal information and cross-border data transfer; general marketing strategies can be briefly summarized." The server organizes these explanations into one or more consecutive supplementary prompts, logically corresponding to the main prompt.

[0144] Output: Additional prompts to guide generative AI models on their focus.

[0145] Step 11: The server invokes a generative artificial intelligence model to generate draft content for business matters.

[0146] Input: main prompt statement, additional prompt statements, and model call parameters (such as temperature, maximum output length, target language, etc.).

[0147] The server sends a request to the generative AI model service via a network interface. This request includes a combined text of the main and supplementary prompts, along with control parameters. Internally, the model encodes the input text using a transformer architecture, calculates the correlation between lexical units using a multi-head self-attention mechanism, and progressively generates the output sequence. The server receives the natural language text returned by the model, parses it into a string, and performs post-processing, including removing extra blank lines, standardizing paragraph separation, detecting potential formatting errors, and marking sensitive words. Finally, the server uses the processed text as the draft content for business matters, writing it along with business matter identification information into the draft content table.

[0148] Output: Draft text of business matters stored in the database.

[0149] Step 12: The server sends the draft content to the terminal and organizes the display structure into segments.

[0150] Input: Business item identification information and corresponding records in the draft content table.

[0151] The server queries the draft content table to obtain the complete draft text. Based on pre-defined sectioning rules (e.g., using heading tags or specific key phrases as delimiters), it divides the text into multiple logical paragraphs, such as "Background and Objectives," "Applicable Laws and Compliance Requirements," "Information Protection Impact Assessment," "Customer and Market Analysis," and "Comprehensive Recommendations and Timeline." The server organizes these paragraphs into structured data containing paragraph titles, content, and numbers, serializes it into JSON, and returns it to the terminal over the network.

[0152] Output: Segmented draft content structure for terminal display.

[0153] Step 13: The terminal receives and displays the drafted content, and provides an editing interface.

[0154] Input: JSON data returned by the server, containing paragraph titles and content.

[0155] The terminal parses the JSON object, generating a corresponding display area and editable text box or "Edit" button for each paragraph in the user interface. The terminal presents the paragraph titles and body text sequentially on the screen, allowing users to modify text, add supplementary explanations, or delete parts of content within the corresponding areas using the keyboard or touch. The terminal can maintain a local change list, recording the paragraph ID, original text fragment, and modified text fragment for each modification, facilitating subsequent unified submission.

[0156] Output: A user-visible interface for drafting content, and locally cached modification history data.

[0157] Step 14: Users review and edit the draft content and confirm its submission.

[0158] Input: The text of each paragraph displayed in the terminal interface.

[0159] Users read through the draft content section by section, revising any inaccuracies or inadequacies, such as adding specific regulatory clause numbers, adjusting risk level descriptions, and adding explanations of internal technical control measures. After completing the overall review, users click the "Save Changes" or "Submit Final Draft" button on their terminal, instructing the terminal to send all recorded changes to the server in batches.

[0160] Output: The user-modified text content, along with the paragraph ID and business item identifier corresponding to each modification, is packaged by the terminal and prepared to be sent to the server.

[0161] Step 15: The terminal sends batch modification information to the server.

[0162] Input: The user's locally cached set of modification records and business item identification information.

[0163] The terminal assembles the paragraph ID, original content summary, modified content, and timestamp of each modification record into structured data, packages it into a JSON request body, and sends it to the server's specified revision submission interface via an HTTP request through the network interface. After sending, the terminal waits for the server's response and displays a "submission successful" or "submission failed" message on the interface based on the server's returned status code and message content.

[0164] Output: Batch modification request messages that can be parsed by the server.

[0165] Step 16: The server merges user modifications to form the final draft information and records the update history.

[0166] Input: Batch modification request messages sent by the terminal and original draft content records in the database.

[0167] The server parses the request message, reads each modification record one by one, and locates the corresponding position in the original draft content using the business item identifier and paragraph ID. The server replaces or inserts the modified text into the corresponding paragraph, generating a new, complete draft text, and writes this text to the final draft information table, or marks it as the latest version in the original draft content record. Simultaneously, the server records each modification log in the revision history table, including the business item identifier, paragraph ID, summary before modification, summary after modification, user identifier, and timestamp. After merging, the server can update the business item status to "drafting completed" and return the processing result to the terminal.

[0168] Output: Updated final draft information record and corresponding revision history.

[0169] 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".

[0170] In the process of initiating and reviewing new business projects related to electronic trading services, the system typically needs to comprehensively analyze multi-dimensional information, including legal and compliance requirements, security and technical risks, market environment, and customer feedback. Currently, most systems only store or retrieve natural language text as simple unstructured data, lacking the ability to perform deep semantic analysis and structured modeling of text content, leading to the following technical problems: (1) On the server side, when processing long text inputs (such as project purpose descriptions, business background descriptions, etc.) in a unified manner, there is a lack of an effective natural language structuring mechanism. It is unable to automatically distinguish legal factors, technical factors, market factors and customer-related factors, resulting in low accuracy of subsequent automatic analysis and generation, and increasing the burden of manual intervention. (2) When using generative artificial intelligence models to generate project proposal documents, existing systems typically use simple, fixed prompts to directly input the entire original text into the model, without fully utilizing the structured elements obtained through natural language processing. This results in problems such as mismatch between the model-generated results and business points, and incomplete coverage of legal and security risks, thereby reducing the usability of automatically generated documents in engineering scenarios. (3) In the process of multiple rounds of interaction between users and the system, existing technologies often ignore the impact of users’ emotional state on information priority setting and content presentation. The server cannot dynamically adjust the weight and level of detail of content such as legal concerns, security impact, and customer reactions according to the user’s emotional changes at different stages (such as urgency, dissatisfaction, anxiety, etc.), which is not conducive to concentrating the model’s computing power on the most important risk points and decision points under limited computing resources. (4) Regarding the integration of the draft of the project proposal document and the user's modification opinions, the traditional solution is mostly a simple overwrite or manual editing. There is no mechanism to encode the existing draft, structured analysis results and user feedback into high-quality prompt statements and input them into the generative artificial intelligence model. As a result, the server cannot perform incremental and highly consistent revisions while maintaining the original document structure and reasonable content, which affects the automation level of the system and the consistency of the output documents.

[0171] Therefore, a technical solution is needed to improve the natural language information processing flow and model invocation strategy on the server side, enabling the server to: efficiently structure natural language item information from the terminal; automatically construct high-quality prompt statements based on structured elements; dynamically adjust the generation strategy in multi-round interactions by combining user emotional state and market environment; and use generative artificial intelligence models to generate and revise new business item proposal documents related to electronic transaction services, so as to improve computer technology from the underlying computing flow and data representation method, and achieve efficient and controllable processing of complex business texts.

[0172] 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.

[0173] In this invention, the server includes: a processing means for receiving natural language information containing project objectives, project overview, and constraints from a terminal via a communication interface, and structuring the natural language information into project analysis data; a processing means for performing word segmentation, sentence segmentation, text classification, and entity extraction on the text data in the project analysis data using a natural language processing program to generate a structured set of elements containing legal factors, technical factors, market factors, and user emotional states; a processing means for extracting legal factors as legal concerns, technical factors and security-related expressions as security impacts, and market factors and customer-related expressions as customer reactions and market environment from the structured set of elements; and a processing means for automatically generating project proposals based on the structured set of elements for input into a generative artificial intelligence model to generate new business projects related to electronic transaction services. The document includes: a processing method for generating a draft project proposal document by inputting prompt statements into the generative artificial intelligence model; a processing method for determining the user's emotional state based on natural language information and user feedback information using an emotion analysis program, generating prompt statements based on the emotional state to adjust the weights and priorities of legal concerns, security impacts, and customer reactions in the project proposal document, and revising the draft project proposal document using the generative artificial intelligence model; a processing method for constructing prompt statements based on user correction opinions and supplementary information received from the terminal and the draft project proposal document, and generating a final version of the project proposal document using the generative artificial intelligence model; and a processing method for sending the final version of the project proposal document to the terminal and storing the final version of the project proposal document in a storage device. This allows for the structured representation of unstructured natural language business information within the server, and the automatic construction of targeted prompts based on structured elements, improving the quality of input signals for generative artificial intelligence models. Simultaneously, by combining the weights and priorities of factors such as user emotional states, market environment concerns, security impacts, and customer reactions in a dynamic adjustment process, the computational process for generating and revising project proposal documents becomes more efficient and controllable at both the model inference and data processing levels. This, in turn, improves the overall processing performance and resource utilization efficiency of the computer system in electronic transaction service project initiation support scenarios, based on improvements to natural language processing and text generation processes.

[0174] A “system” refers to a collection of devices consisting of at least one server, at least one terminal, and programs for performing information processing and communication control.

[0175] A "server" is an electronic computing device equipped with a processor and storage device, capable of sending and receiving data with a terminal via a network and performing natural language processing and generative artificial intelligence model reasoning.

[0176] "Terminal" refers to an information processing device operated by a user for inputting natural language information and receiving and displaying results such as project proposal documents, including but not limited to computer equipment, mobile communication equipment, or tablet devices.

[0177] A "processor" refers to a hardware computing unit within a server that executes program instructions to process data, control the collaborative work of various processing methods, and invoke generative artificial intelligence models and natural language processing programs.

[0178] A "communication interface" refers to a set of hardware and software interfaces used for data transmission between a server and a terminal, including the communication protocol stack and its drivers.

[0179] "Natural language information" refers to text data or its digital representation input by users in human natural language, including project objectives, project overview, constraints, and suggestions for improvement.

[0180] "Project analysis data" refers to a collection of structured or semi-structured data generated by a server based on natural language information for subsequent semantic processing.

[0181] "Natural Language Processing Program" refers to a software module or model used to perform operations such as word segmentation, sentence segmentation, text classification, entity extraction, and sentiment analysis on text data.

[0182] "Structured element set" refers to a set of elements such as legal factors, technical factors, market factors, and user emotional states that are extracted from natural language information through natural language processing programs and represented in the form of predetermined fields or tags.

[0183] "Legal factors" refer to content in natural language information or a set of structured elements that relates to laws, regulations, regulatory requirements, and compliance obligations.

[0184] "Technical factors" refer to elements in natural language information or a set of structured elements that are related to technical content such as system architecture, implementation methods, security mechanisms, and performance requirements.

[0185] "Market factors" refer to elements in natural language information or structured sets that are related to the market environment, such as the target user group, competitive situation, market demand, and charging model.

[0186] "User emotional state" refers to the emotional tendency of a user during communication, which is inferred from natural language information or user feedback through emotion analysis programs. This includes, but is not limited to, states such as urgency, anxiety, satisfaction, or dissatisfaction.

[0187] "Legal concerns" refer to points of concern that the server infers based on legal factors, which may involve legal, regulatory, or supervisory risks.

[0188] "Security impact" refers to the risks or effects that a server may have on information security, system security, or transaction security, as determined by technical factors and security-related expressions.

[0189] "Customer response" refers to the factors related to the attitudes, evaluations, preferences, and expected behaviors of end users or service recipients towards a product or service.

[0190] "Market environment" refers to the external market conditions surrounding a specific product or service, including the competitive landscape, demand trends, policy guidance, and characteristics of target users.

[0191] "Generative AI models" refer to AI models that can automatically generate text and other content based on input prompts, including language models based on deep learning and neural network structures.

[0192] "Prompt statements" refer to instructional or descriptive texts that are automatically constructed by the server based on a set of structured elements, project analysis data, and user feedback, and are used as input for generative artificial intelligence models.

[0193] "Project proposal document" refers to a document generated for a specific business project, which describes the project overview, legal and compliance analysis, technical solutions and security design, market environment and customer response analysis, as well as risk control and project advantage maximization strategies.

[0194] "Draft Project Proposal Document" refers to a version of a project proposal document generated by the server using a generative artificial intelligence model for the first time or at an intermediate stage, which can still be revised based on user feedback.

[0195] The "final version of the project proposal document" refers to the version of the project proposal document that has been generated and revised at least once based on user feedback and supplementary information, and is output and stored as a completed version.

[0196] "Emotion analysis program" refers to a software module or model used to analyze natural language information and user feedback to infer the user's emotional state.

[0197] "User feedback information" refers to natural language content such as modification suggestions, supplementary explanations, or evaluations entered by users while viewing project proposal documents.

[0198] "Weight" refers to the relative importance parameters assigned to different elements (including legal concerns, security implications, customer reactions, etc.) when generating or revising project proposal documents.

[0199] "Priority" refers to the order and importance of different elements or chapters in terms of processing sequence or level of detail in description when generating or revising project proposal documents.

[0200] A “structured project proposal document” refers to a project proposal document with a pre-defined chapter structure, in which each chapter includes content such as legal and compliance analysis, technical solutions and security design, market environment and customer response analysis, risk control and project advantage maximization strategies.

[0201] "Storage device" refers to a storage medium or storage system used to store information such as project analysis data, structured element sets, and project proposal documents for long-term or temporary storage.

[0202] In various embodiments of this invention, the server, terminal, and user each assume different roles. The server performs core computing and storage operations, the terminal handles human-computer interaction and data transmission and reception, and the user provides business statements and feedback. The system of this invention can be deployed in a data center or cloud computing platform, or it can be implemented as part of a local enterprise internal system.

[0203] In one embodiment, the server includes a processor, storage devices, and a communication interface. The server's processor can be a combination of a multi-core general-purpose processor and an accelerator, such as a computing platform composed of a multi-core general-purpose processor (e.g., a central processing unit based on a general-purpose instruction set) and a graphics processing unit (GPU). The server's storage devices can include solid-state drives, hard disk arrays, and memory, used to store natural language processing programs, generative artificial intelligence model parameters, project analysis data, and versions of project proposal documents. The server's operating system can be a UNIX-like system, such as a server operating system. The server can run scripting language-based backend application frameworks, such as Python-based web frameworks (e.g., Flask or FastAPI), or it can run a JavaScript-based runtime environment and its web framework.

[0204] In one implementation, the natural language processing software used by the server includes a word segmentation library, a tagging library, and a sentence vector encoder based on a pre-trained language model. For example, the server can use a Chinese word segmentation library to perform word segmentation, a natural language processing library to perform part-of-speech tagging and dependency parsing, and a pre-trained transformer model (such as a Chinese language model based on the Transformer architecture) to compute text vector representations. The server can also use deep learning frameworks, such as PyTorch or TensorFlow, to load and execute generative artificial intelligence models. The generative artificial intelligence model can be an autoregressive language model with a multi-layered self-attention structure, comprising word embedding layers, multi-layered Transformer encoding blocks, feedforward network layers, and an output layer. The server persists the model's weight parameters in a storage device.

[0205] In one embodiment, the terminal can be a personal computing device, a mobile communication device, or a tablet device. The terminal runs a browser application or a dedicated client application to present the input and result interfaces. The terminal exchanges data with the server via a secure communication protocol (e.g., HTTPS). After the user inputs a natural language description, the terminal encapsulates the text and auxiliary fields into a request message and sends it to the server. Upon receiving a draft or final version of the project proposal document from the server, the terminal displays it to the user chapter by chapter and provides input areas for modification comments in each chapter.

[0206] In one implementation, the user inputs natural language information via a terminal, including project purpose, project overview, target users, technical constraints, and known compliance requirements. The user can input text such as the following: "I want to develop a small-amount mobile payment application for freelancers. It needs to comply with laws related to personal information protection and anti-money laundering regulations. I hope the system has transaction risk control capabilities and is easy to use." After receiving the natural language information, the server performs a series of data processing and computation processes using a natural language processing program. First, the server performs word segmentation, sentence segmentation, and part-of-speech tagging, transforming the original string into word and sentence sequences. The server then uses an encoder based on a pre-trained model to map each sentence into a vector, with dimensions such as 256, 512, or 768. The server classifies the sentence vectors using a multi-class neural network, categorizing sentences into groups such as "law and compliance related," "technology and security related," and "market and customer related." This multi-class network can contain at least one fully connected layer and one softmax output layer. During training, the server minimizes the classification error using a cross-entropy loss function and updates the weight parameters using gradient descent or its variants (such as the Adam optimization algorithm).

[0207] After obtaining the sentence classification results, the server uses a named entity recognition model to extract entities such as legal names, technical terms, product categories, and target user categories from the text. This named entity recognition model can be a sequence labeling model based on a bidirectional encoder and conditional random fields. During training, the server learns the model parameters by maximizing the conditional probability of the labeled sequences. The server maps the extracted entities to a predefined set of fields, such as "legal factors," "technical factors," "market factors," and "user type," thereby constructing a structured set of elements. Internally, the server can store these elements in key-value pairs, for example, storing them in the form of a dictionary or table such as "legal factors: laws related to personal information protection, anti-money laundering regulatory requirements; technical factors: user data encryption, multi-factor authentication, transaction risk control; market factors: freelancers' need for small-amount payments, preference for ease of operation," etc.

[0208] In one implementation, the server also runs a sentiment analysis program. The server inputs the user's natural language input and subsequent feedback text into a sentiment classification model. This model can be a bidirectional recurrent neural network or a Transformer-based classification model, whose output includes sentiment labels such as "neutral," "urgent," "dissatisfied," and "worried," along with corresponding confidence scores. The server trains the model using cross-entropy loss by inputting text vectors into a fully connected network and outputting sentiment category probabilities, enabling it to distinguish different sentiment states. During runtime, the server determines the current user's sentiment state based on the highest probability and assigns different weights to different factors (legal concerns, security impact, customer reaction) according to the sentiment category. For example, when the server detects "worried," it can increase the weights of security impact and legal concerns; when it detects "urgent," it can increase the priority of project implementation time and key risk components.

[0209] In one implementation, the server automatically generates prompts for the generative artificial intelligence model using the aforementioned set of structured elements. Instead of simply inputting the raw text as a whole into the model, the server organizes the information into sections based on the classification results and the structured elements. For example, the server constructs the following prompts: Role: You are a senior project consultant in the field of electronic payment services.

[0210] Project Objective: To develop a mobile payment application for small amounts targeting freelancers, requiring compliance with laws related to personal information protection and anti-money laundering regulations, with a focus on improving transaction security and risk control capabilities.

[0211] Identified legal issues: laws related to personal information protection, and anti-money laundering regulatory requirements.

[0212] Identified technical issues and security impacts include: user data encryption, multi-factor authentication login, transaction risk control models, account security risks, and privacy leakage risks.

[0213] Identified market environment and customer response: Freelancers have a strong demand for small-amount payment tools, there is already competition in the market from QR code payment products, users want low transaction fees, and users want simple operation processes.

[0214] Task: Based on the information above, please draft a project proposal document for a new electronic payment project using formal written language, including at least the following sections: 1. Project Overview 2. Legal and Compliance Analysis (explaining specific regulatory requirements and legal concerns) 3. Technical solutions and safety design (explain the key technical measures and their mitigation effects on safety) 4. Market Environment and Customer Response Analysis 5. Risk control measures and strategies for maximizing project advantages. The server encodes the prompt statement into a sequence of inputs for the model and performs forward propagation computation of the generative AI model on the GPU. During execution, the server performs multi-head self-attention computation, residual connections, and layer normalization on each Transformer layer to aggregate contextual information from different locations in the vector space. The server controls the diversity and stability of text generation by setting upper limits on generation length, temperature parameters, and sampling strategies (e.g., top-k or top-p sampling). After completing inference, the server obtains a draft project proposal document containing multiple chapters and sub-paragraphs.

[0215] In one implementation, the server performs post-processing on the generated project proposal document draft. The server can use simple rules to check if chapter headings are complete; if a chapter is missing, the server can insert a placeholder heading and add the content during subsequent revisions. The server stores the draft along with the corresponding structured elements and user session identifiers in a database management system, such as a relational database. In this way, the server achieves associated management of the natural language draft and structured metadata at the storage level. Compared to the traditional method of only storing the full text, this facilitates subsequent retrieval and targeted regeneration, improving query efficiency and the controllability of storage management.

[0216] In one implementation, users view a draft project proposal document on a terminal and enter specific modification comments under each section. For example, a user can enter: "Please add a description of the regulatory requirements for cross-border data transfers to the legal and compliance analysis."

[0217] Please add descriptions of database backup strategies and disaster recovery plans to the technical solution and security design. The terminal sends these feedback texts along with the project identifier to the server. The server then runs a natural language processing program to parse the feedback text into a structured revision request, specifying the target chapter, revision type, and key requirements. Simultaneously, the server runs a sentiment analysis program to determine the user's emotional state at the time of feedback and adjusts the emphasis on certain risk points in subsequent generation accordingly.

[0218] In one implementation, the server constructs a new prompt statement based on the original draft project proposal document, the set of structured elements, and the structured revision request. For example, the server can generate the following prompt statement: Role: You are an electronic payment project consultant, and you need to revise the following draft project proposal document based on user feedback.

[0219] Draft Project Proposal Document (Insert the full text of the original project proposal document here) User Feedback 1. In the 'Legal and Compliance Analysis' section, please add a description of the regulatory requirements for cross-border data transfers.

[0220] 2. In the 'Technical Solution and Security Design' section, please supplement the description of the database backup strategy and disaster recovery plan.

[0221] Task: Without altering the original chapter structure, revise and supplement the proposal document based on user feedback. Retain existing reasonable content, expand relevant chapters, and make the overall document more complete, professional, and clear. The server inputs the prompt into the generative AI model, performs reasoning again, and obtains the final version of the revised project proposal document. The server performs consistency checks on the final version, such as verifying whether the chapter order is maintained and whether key legal and technical elements are still presented. The server then sends the final version to the terminal and records it as the new version in the storage device to support version management and subsequent auditing needs.

[0222] In one implementation, the server, through the aforementioned structured parsing and prompt generation mechanism, significantly reduces input noise to the model compared to directly inputting raw, long text. This improves the model's coverage of key elements within a limited inference length. Because the server breaks down different categories of information into independent fields and explicitly labels them (such as legal factors, technical factors, and market factors), the model can receive high-value features in a more compact form. This enhances the consistency between the generated results and the key business points, reduces repetitive fragments and irrelevant content, and achieves a dual improvement in computational efficiency and generation quality.

[0223] In an alternative implementation, the server can employ different generative AI model structures. For example, the server can use an encoder-decoder sequence-to-sequence model, where the encoder receives an input sequence linearized from structured elements, and the decoder generates a project proposal document. Under this structure, the server can use attention weight analysis to identify which input elements contribute significantly to specific paragraphs, thereby adjusting the element representation during subsequent retraining or model tuning phases to enhance the influence of key elements.

[0224] In another alternative implementation, the server can compress a large teacher model into a smaller student model for online inference using knowledge distillation. During the training phase, the server uses the teacher model to generate the target distribution, while the student model learns by minimizing the difference from the teacher's output (e.g., through KL divergence). This reduces the parameter size while maintaining high generation quality, lowering the computational burden of each inference iteration. Consequently, when faced with concurrent requests from multiple users, the server can maintain acceptable response latency without significantly increasing hardware resources, thus improving overall system performance.

[0225] In one implementation, the server trains the generative AI model using training corpus comprised of historical project proposal documents, analysis reports written by human experts, and simulated generated samples. Training is performed using an autoregressive language modeling objective or a joint task (such as next sentence prediction or paragraph title prediction). During training, the server uses a loss function (e.g., cross-entropy loss) to measure the difference between the predicted output and the real text, and adjusts the network weights through backpropagation and gradient update algorithms. The server can apply training techniques such as learning rate decay, gradient clipping, and parameter regularization to stabilize the model convergence process and suppress overfitting. The server can also perform data augmentation during the training data construction phase, such as using different expressions for the same content, changing the chapter order, or adding noisy sentences, to improve the model's robustness to diverse inputs.

[0226] In one implementation, the server achieves unified management of text features and structured elements by designing specific data structures for project analysis data. For example, the server can record a master record for each project in the database and store sentence-level classification labels, entity recognition results, sentiment labels, and association weights in the association table. Before generating project proposal documents, the server can dynamically construct prompt statements based on these structured fields, thereby avoiding re-executing the entire natural language parsing process for each generation, reducing computational redundancy, processing latency, and resource consumption in multi-round revision scenarios.

[0227] In one implementation, the server applies the aforementioned technologies to the project initiation environment for new business projects related to electronic transaction services. By combining structured elements with generative artificial intelligence models, the server enables the system to automatically identify legal risk points and security vulnerabilities when processing complex business texts, clearly presenting them in the project proposal document. This facilitates subsequent configuration of risk control rules, access control policies, and system architecture by users. For example, when the server identifies elements involving "cross-border data transfer," it automatically adds descriptions of data localization requirements and cross-border transfer approval processes to the generated document, prompting the use of additional encrypted channels or access log auditing mechanisms. This provides concrete technical guidance for subsequent system deployment and equipment configuration at the document level, going beyond abstract business descriptions.

[0228] Through the aforementioned implementation models and alternative methods, the server internally performs structured modeling, feature extraction, sentiment weighting, and prompt generation on natural language information. Combined with deep network inference using generative artificial intelligence models, it achieves refined processing of complex business information. This approach is not a simple automated replacement of manually written documents, but rather introduces new structured processes at the internal computer technology levels, such as data representation, model input construction, and inference strategies. This improves the accuracy and speed of text generation, reduces unnecessary computation and communication load, and thus enhances the overall technical performance of the computer system in relevant application scenarios.

[0229] use Figure 12 The processing flow is explained.

[0230] Step 1: Users input project-related natural language information using a terminal and send it to the server.

[0231] Users open the project input interface on the terminal and enter natural language text such as project purpose, project summary, target users, technical requirements and compliance requirements in the text input box. For example, "I want to develop a small-amount mobile payment application for freelancers. It needs to meet the relevant laws on personal information protection and anti-money laundering regulations. I hope the system has transaction risk control capabilities and is easy to operate."

[0232] The terminal combines the fields entered by the user into a request object in its local memory and validates the required fields. If the fields are complete, the terminal serializes the request object into a text format (such as a JSON string) and encapsulates it into a request message through a network protocol.

[0233] The terminal sends a request message to the server's designated interface address through a secure communication channel.

[0234] The input for this step is a natural language description of the item and related parameters entered by the user on the terminal, and the output is a structured request message sent from the terminal to the server.

[0235] Step 2: The server receives and parses the request messages sent by the terminal, and generates initial project analysis data.

[0236] The server receives request messages from the terminal through the communication interface and caches the message data in memory.

[0237] The server uses a parser to decode the message content, converting the text format (such as a JSON string) into an internal data structure, such as a key-value map or record object, and extracting fields such as "project purpose", "project summary", "target users" and "known compliance requirements".

[0238] The server performs basic cleaning processing on the extracted natural language text, including removing extra spaces, standardizing the encoding format, and removing control characters, thereby obtaining standardized raw text.

[0239] The server packages these cleaned fields into project analysis data objects and establishes an association with the session identifier or project identifier in the storage device.

[0240] The input for this step is the request message sent by the terminal. After the server processes the data, such as decoding and extracting fields, the output is a project analysis data object that can be used for subsequent natural language processing.

[0241] Step 3: The server uses natural language processing to perform text segmentation and sentence division on the project analysis data, generating sentence-level text units.

[0242] The server reads long text fields, such as project summary and project purpose, from the project analysis data object and sends them to the word segmentation module.

[0243] The server uses a word segmentation algorithm to divide a continuous stream of characters into word sequences, and then uses punctuation marks and a machine learning model to jointly determine sentence boundaries, thus dividing the text into several sentences.

[0244] The server internally assigns a unique number to each sentence and binds and stores the sentence with its corresponding field and word sequence to form a list of sentence-level text units.

[0245] The input for this step is a project analysis data object containing the original text. The server performs data operations such as word segmentation and sentence splitting, and outputs a collection of sentence-level text units with sentence numbers and word sequences.

[0246] Step 4: The server vectorizes and classifies sentence-level text units to identify categories such as law, technology, and market.

[0247] The server extracts sentences one by one from the set of sentence-level text units and feeds them into a pre-trained language model encoder, which maps each sentence into a vector representation of a fixed dimension.

[0248] The server inputs sentence vectors into a multi-class classification network, and uses fully connected layers and normalization functions to calculate the probability distribution of each category (law and compliance, technology and security, market and customer, etc.).

[0249] The server assigns a category label to each sentence based on the highest probability and writes the label back to the corresponding sentence record.

[0250] The input for this step is a set of sentence-level text units. The server performs data operations such as vectorization and classification, and outputs a set of sentences with category labels for subsequent feature extraction.

[0251] Step 5: The server performs entity extraction and feature classification based on the labeled sentence set, generating a structured feature set.

[0252] For sentences tagged as legal and compliance, the server invokes a named entity recognition program to identify entities such as legal names, compliance terms, and regulatory agencies, and categorizes them into the legal factors field.

[0253] For sentences categorized as technical and security, the server identifies technical terms (such as encryption, authentication, risk control, backup, etc.) and phrases related to security risks, and categorizes them into the technical factors and security impact fields.

[0254] For sentences tagged with market and customer categories, the server identifies information such as target user groups, competitive environment, pricing tendencies, and user preferences, and categorizes them into the market factors and customer response fields.

[0255] The server organizes these extracted entities and phrases into a structured set of elements, where each element is associated with the original sentence number and category label.

[0256] The input for this step is a set of sentences with category labels. The server processes the data through entity recognition and classification, and outputs a set of structured elements containing items such as legal factors, technical factors, market factors, customer reactions, and security impacts.

[0257] Step 6: The server uses a sentiment analysis program to analyze users' natural language information and feedback, determine users' emotional states, and generate sentiment weights.

[0258] The server extracts the latest text fragments from the project analysis data objects and subsequent user feedback texts, and inputs them into the sentiment classification model.

[0259] The server uses text vectorization and an emotion classification network to calculate the probability of each emotion label (such as neutral, urgent, worried, dissatisfied, etc.) and selects the label with the highest probability as the current user's emotional state.

[0260] The server searches a pre-set weight table based on emotional state. For example, in the "worried" emotional state, the processing weight of content related to safety impact and legal concerns is increased, and in the "urgent" emotional state, the priority of key risks and time-sensitive parts is increased.

[0261] The server records the emotional state and its corresponding weight parameters into a structured element set, which is used for subsequent prompt statement construction and generation strategy adjustment.

[0262] The input for this step is natural language text and existing structured features. The server performs data operations such as emotion classification and weight matching, and outputs an enhanced set of structured features labeled with emotion states and weight parameters.

[0263] Step 7: The server generates prompts for generative artificial intelligence models based on a set of structured elements and sentiment weights.

[0264] The server reads fields from the structured feature set, including project objectives, legal factors, technical factors, security implications, market factors, customer reactions, and sentiment weights.

[0265] The server fills these fields into specific locations according to a predefined template, constructs prompts with fixed chapters and explanatory instructions, such as explicitly listing "Project Objectives," "Identified Legal Issues," "Identified Technical Issues and Security Impacts," and "Identified Market Environment and Customer Responses" in the text, and adds a "Task Description" section to guide the generation structure and tone.

[0266] The server dynamically adjusts the emphasis of certain parts in the prompt statement based on sentiment weight, such as adding phrases like "Please emphasize the risk control measures" and "Please explain the security plan in detail."

[0267] The input for this step is a set of structured elements with emotional weights. The server processes the data through template filling and weight adjustment, and outputs a structured and clearly defined prompt text.

[0268] Step 8: The server will input the prompts into the generative artificial intelligence model to generate a draft project proposal document.

[0269] The server converts the prompt statement into a sequence of input tokens for the model, which is then mapped to word or sub-word indices by a tokenizer to form an input vector.

[0270] The server performs forward inference of the generative artificial intelligence model on computing-accelerated hardware. At each time step, it calculates the probability distribution of the next label based on the context and model parameters, and selects the output label through sampling or greedy strategies to gradually generate a complete document.

[0271] The server continues to generate until it reaches the preset length or encounters an end marker, and then concatenates the generated results into a readable draft project proposal document, which contains multiple chapters and paragraphs.

[0272] The input for this step is a prompt statement generated by the server. The server performs data operations such as forward computation of neural networks and probability sampling, and outputs a structured project proposal document draft text.

[0273] Step 9: The server checks and stores the draft project proposal document for formatting, and then sends it to the terminal.

[0274] The server parses the generated document draft and checks whether the expected chapter titles are complete. If some chapters are missing, the server can insert default titles or prompt text to ensure the integrity of the document structure.

[0275] The server stores the draft document along with project identifiers, structured elements, and version information in the database for subsequent revisions and tracking.

[0276] The server packages the draft document into a response message and sends it to the requesting terminal via the communication interface.

[0277] The input for this step is the raw document text output by the generative artificial intelligence model. The server processes the data through format validation and persistent storage, and the output is a draft record in the database and a document draft response sent to the terminal.

[0278] Step 10: The terminal receives and displays the draft project proposal document, and the user inputs revision comments.

[0279] The terminal receives a response message containing a draft document from the server, parses the message, divides the document content into chapters, and displays it on the interface.

[0280] The terminal provides text input areas and operation buttons under each chapter, allowing users to input modification opinions and supplementary explanations chapter by chapter.

[0281] Users can enter their natural language revision suggestions on the terminal, such as "Add explanation of regulatory requirements for cross-border data transmission" or "Supplement database backup and disaster recovery strategies," in the corresponding chapters as needed.

[0282] The input for this step is the draft project proposal document returned by the server. The terminal generates an editable display interface through parsing and presentation operations, and the output is the revision comments text entered by the user on the terminal.

[0283] Step 11: The terminal sends the user's revision comments to the server, which then parses the comments and updates the set of structured elements.

[0284] The terminal combines the revision comments for each chapter with the project identifier and draft version number into a request message, and sends it to the server through a secure communication channel.

[0285] After receiving the message, the server parses it and extracts the target chapter and specific text content corresponding to each revision suggestion.

[0286] The server uses natural language processing to segment and analyze the revision suggestions, identify newly added legal, technical, or market elements, and merge these elements into the original structured element set. At the same time, it records the source of the revision as "user feedback".

[0287] The input for this step is the revision message submitted by the user on the terminal. The server processes the data through parsing and feature extraction, and outputs an updated set of structured features that incorporates the user's revision information.

[0288] Step 12: The server regenerates the prompts based on the original draft and the updated set of structured elements, and calls a generative artificial intelligence model to generate the final version of the project proposal document.

[0289] The server reads the original project proposal document draft text from the database, integrates it with user revisions and updated structured elements, and constructs new prompt statements that explicitly include instructions such as "original draft content," "user feedback content," and "keep the original structure and add relevant parts."

[0290] The server re-encodes the prompt statement into an input sequence, feeds it into a generative artificial intelligence model to perform forward inference, and generates the final version of the revised project proposal document.

[0291] The server performs integrity and consistency checks on the final version of the document, confirming that all specified chapters have been appropriately modified or expanded, and stores the final version in the database. At the same time, it sends the final version to the terminal for user confirmation and use through the communication interface.

[0292] The input for this step is the original draft document, user revision comments, and the updated set of structured elements. The server performs data operations such as reconstructing prompts and further reasoning, and outputs the final version of the project proposal document text and its version record in the storage system.

[0293] 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.

[0294] 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."

[0295] Existing case management and risk assessment technologies typically employ rule engines or fixed templates to statically match and score business data. These technologies suffer from the following problems: First, when faced with complex and ever-changing legal regulations, information security requirements, and market feedback, the system often can only make judgments based on a few predefined scenarios, making it difficult to comprehensively understand the semantics of cases described in natural language, resulting in incomplete and imprecise risk identification. Second, most existing systems analyze data from a single dimension, such as only performing keyword matching on compliance clauses or checklist checks on security configurations. They lack the ability to uniformly model and jointly assess legal risk indications, information security risk indications, and customer reaction risk indications within the same computational process, leading to fragmented risk conclusions and hindering downstream automated processing. Third, even when analytical results are generated, traditional systems typically output reports in fixed formats, lacking the ability to automatically convert structured risk indications into multi-level natural language reports tailored to different processing entities (e.g., different business departments or functional departments), resulting in significant manual interpretation and reorganization workloads and high latency. Fourth, at the computer system architecture level, existing solutions lack a prompt generation mechanism and data flow control mechanism for generative artificial intelligence models. This prevents the full utilization of the semantic reasoning and document generation capabilities of generative AI models, hindering the completion of the end-to-end automated processing flow of "case information acquisition → risk calculation → multi-dimensional benchmark comparison → report allocation" within the same system. Therefore, how to leverage generative AI models and prompt generation technology at the computer technology level to perform unified semantic analysis and risk quantification of multi-source business data and benchmark information, and on this basis, automatically generate multi-format report information that can be allocated according to departmental categories and priorities, thereby improving the processing efficiency and intelligence level of information processing systems in terms of compliance, security, and customer response, has become an urgent technical challenge.

[0296] 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.

[0297] In this invention, the server includes: an acquisition means for obtaining case-related information from external sources; a prompting statement generation means for automatically generating prompting statements for a generative artificial intelligence model based on the acquired case information; a parsing means for calling the generative artificial intelligence model to parse the text of the case information to calculate various risk indication information, including legal risk indication information, information security risk indication information, and customer response risk indication information; an evaluation means for reading benchmark information corresponding to the business domain from a storage device based on the risk indication information and quantitatively evaluating the correspondence between the case information and the benchmark information; a report information generation means for generating natural language report information containing compliance countermeasure information, information security countermeasure information, and customer response policy information based on the correspondence and risk indication information; a classification means for classifying the report information according to departmental categories and priorities and constructing the classification results into additional prompting statements to be input back into the generative artificial intelligence model; and an allocation means for generating allocation information for allocating the report information to multiple processing entities based on the classification results and risk indication information and outputting the allocation information. This allows for the creation of an end-to-end automated processing pipeline within the computer, encompassing case information collection, semantic parsing using a generative AI model driven by prompt statements, multi-dimensional risk quantification, automatic comparison with regulations and security benchmarks, generation of multi-format natural language reports for different processing entities, and report distribution. This enhances the automation and processing performance of information processing systems in compliance review, information security assessment, and customer response prediction without relying on manual rule writing or document organization. It enables high-precision risk identification and efficient response to complex business scenarios, improving the overall processing capabilities of computer technology in the field of intelligent processing of multi-source text data.

[0298] A "system" refers to an overall technical solution consisting of one or more computing devices and the programs, storage devices, and interface devices running on them, used to process input information and output processing results.

[0299] "Means of acquisition" refers to the hardware, software, or combination thereof used in a system to receive or read information from external devices, networks, or storage media, including communication interfaces, input interfaces, and program functions that implement data reception control.

[0300] "Case information" refers to the general term for descriptive and structured information related to a specific business activity, project, or matter, including product or service descriptions, target market information, relevant regulatory information, summaries of historical customer feedback, and other relevant data used for risk assessment.

[0301] "Generative artificial intelligence models" refer to artificial intelligence models that are based on machine learning algorithms, especially deep learning models, and can automatically generate new text, structured results, or other output content based on input text or other data by learning from a large amount of sample data.

[0302] "Prompt statements" refer to textual or structured instructions used as input to generative artificial intelligence models to guide them in parsing, reasoning about, or generating outputs from input data according to predetermined tasks, formats, or constraints.

[0303] "Prompt statement generation method" refers to the program function or its implementation module in the system that automatically constructs or edits prompt statements based on case information and predetermined templates, so that the case information and task requirements can be input into the generative artificial intelligence model.

[0304] "Analysis methods" refers to the program functions or implementation modules that call generative artificial intelligence models to perform semantic analysis, feature extraction, and reasoning calculations on case information in order to obtain multiple risk-related indicators or attributes involved in the case.

[0305] "Legal risk indication information" refers to indicator information obtained by analytical or evaluation methods that characterizes the degree to which a case may violate or is close to violating relevant laws, regulations, or regulatory requirements, including the degree of relevance to legal provisions, risk level, or risk type.

[0306] "Information security risk indication information" refers to indicator information obtained through analysis or evaluation methods that characterizes the potential weaknesses and severity of information security aspects such as data protection, access control, encryption, and auditing in a case.

[0307] "Customer reaction risk indication information" refers to indicator information obtained by analysis or evaluation methods that characterizes the risk of negative reactions or complaints from customers or users regarding products or services related to a case in terms of privacy, experience, performance, etc.

[0308] "Benchmark information" refers to information pre-stored in a storage device as a reference for judging or evaluating case information, including legal norms, security control standards, business rules, historical project data, and other predefined comparison standards.

[0309] "Evaluation methods" refers to the program functions or implementation modules that calculate the correspondence or similarity between case information and benchmark information based on risk indication information obtained from analytical methods and benchmark information obtained from storage devices, and quantify or grade the degree of risk accordingly.

[0310] "Correspondence" refers to the representation of the relevance, matching degree, or similarity between case information and benchmark information at the semantic, logical, or rule level, and can be expressed in the form of scores, weights, levels, or association lists.

[0311] "Report information generation means" refers to the program function or implementation module that automatically constructs natural language report texts for human reading based on risk indication information and corresponding relationships, and organizes relevant risks and countermeasures in an understandable structure.

[0312] "Report information" refers to documents or text content generated by report information generation methods and described in natural language. It includes at least a description of legal risks, information security risks, and customer reaction risks, as well as corresponding compliance countermeasures, information security countermeasures, and customer response policies.

[0313] "Compliance action information" refers to the specific measures or recommendations in a report that indicate how to meet or improve compliance with applicable laws, regulations, regulatory requirements, and internal rules.

[0314] "Information security countermeasures information" refers to the content in the report information that indicates the technical measures, management measures, or procedural measures that should be implemented to reduce or avoid information security risks, including but not limited to recommendations on encryption, access control, auditing, and monitoring.

[0315] "Customer Response Policy Information" refers to the information in the report that indicates the response strategies and improvement directions to be taken in product design, service provision, communication strategies, etc., in response to anticipated customer reactions or potential customer complaints.

[0316] "Classification methods" refers to the program functions or implementation modules that categorize and organize the content of report information according to predetermined departmental categories, business types and priorities, and generate corresponding classification results or structured tags.

[0317] "Department Category" refers to the category information obtained based on the division of functions or business within an organization, including but not limited to legal-related functions, information security-related functions, marketing-related functions, and other business functions.

[0318] "Priority" refers to the hierarchical information used to prioritize or allocate resources among multiple pending tasks, and is used to characterize the urgency or importance of a particular report or task.

[0319] "Additional prompt statements" refer to prompt statements that are further constructed based on the initial prompt statements, according to the classification results and risk indication information. These prompt statements are used to re-input the generative artificial intelligence model to generate more detailed or targeted control information for the output content of a specific processing subject.

[0320] "Assignment method" refers to the program function or implementation module that generates assignment information based on classification results and risk indication information, and automatically assigns report information or related tasks to multiple processing entities (including different departments, positions or system modules).

[0321] "Assignment information" refers to structured data or control instructions generated by assignment methods, which indicate which processing entity should process a specific report content or processing task under what priority.

[0322] "Processing entity" refers to the execution unit that specifically processes the reported information or tasks during system operation. It can be a human operator, a functional department within an organization, or other information processing modules or business systems.

[0323] "Storage device" refers to a hardware device used to store case information, benchmark information, analysis results, report information and allocation information in a read-write manner, including semiconductor memory, magnetic memory, optical storage media and the database system or file system formed therefrom.

[0324] The embodiments of this invention will be described in conjunction with the specific composition of the hardware structure, software modules, data structure, generative artificial intelligence model, and specific examples of prompt statements. In the following paragraphs, the terms server, terminal, and user are used as subjects to support the specific implementation of the system defined in the claims in a real computing environment.

[0325] I. Overall System Structure The server, as the core computing node of the system of this invention, is installed on a computing device with a multi-core central processing unit, main memory, non-volatile memory, and a network interface. The server can adopt a general-purpose server hardware architecture, such as a multi-core processor (e.g., an x86 architecture processor or an ARM architecture processor), large-capacity random access memory, solid-state drives, or disk arrays. The server runs middleware and application software on top of the operating system, including: web server software, application service frameworks, database management systems, and generative artificial intelligence model inference services.

[0326] A terminal, acting as a front-end device for user-server interaction, can be a desktop, mobile, or embedded information processing device. The terminal runs graphical user interface applications, such as web applications based on a general-purpose browser or specialized applications. The terminal establishes an encrypted communication channel with the server via a communication network.

[0327] Users can input case information, view server-generated reports, and make subsequent processing decisions based on the report results through the terminal's user interface.

[0328] II. Server-side hardware and software composition The server can run a general-purpose operating system, and the following software components run on that operating system: The server runs web server software to receive Hypertext Transfer Protocol (HTTP) requests from terminals. The server then forwards the decrypted requests to the backend application framework.

[0329] The server runs an application framework that implements the specific program logic for acquisition methods, prompt statement generation methods, parsing methods, evaluation methods, report information generation methods, classification methods, and allocation methods. The server can be divided into structured modules, with each "method" corresponding to one or more application modules.

[0330] The server runs database management software to store case information, baseline information, risk indication information, report information, and allocation information. Multiple tables or collections can be created in the database to manage different categories of data.

[0331] The server runs a generative artificial intelligence model inference service, used for semantic parsing, risk factor extraction, and natural language report generation of input prompts and case texts. The server can employ a deep neural network model based on a transformer architecture as the generative artificial intelligence model.

[0332] III. Structure and Learning Methods of Generative Artificial Intelligence Models The generative artificial intelligence model used by the server in this embodiment can be a language model based on a multi-layer transformer encoder-decoder structure or an autoregressive language model using only a decoder structure. Internally, the server employs a multi-head self-attention mechanism, a feedforward fully connected network layer, residual connections, and a layer normalization structure.

[0333] When training this generative AI model, the server uses a large-scale natural language corpus as pre-training data and can further train using domain corpora related to legal texts, information security regulations, and customer feedback data. The server employs a cross-entropy loss function as the error function during model training to measure the difference between the model-generated labeled sequence and the target labeled sequence. The server updates the model parameters through backpropagation and gradient descent optimization algorithms, progressively minimizing the error function value.

[0334] In the further instruction fine-tuning phase, the server uses pairs of "prompt statements - target output text" as training samples, where the target output text includes structured risk element descriptions, regulatory clause summary explanations, security control recommendations, and customer reaction predictions. During this phase, the server, through supervised fine-tuning, constrains the model behavior within the task space relevant to the system of this invention.

[0335] When needed, the server can also employ reinforcement learning methods to evaluate the quality of the model's output reports and optimize parameters in reverse. For example, the server can construct a rule-based scorer to rate whether the generated reports cover all necessary risk dimensions and conform to a predetermined structure, and adjust model parameters using policy gradient algorithms to increase the probability of high-scoring outputs.

[0336] IV. Data Structures and Internal Representation The server creates a data table for case information in the database. The fields in the data table include case identifier, user identifier, text description, target market, applicable laws and regulations, historical feedback summary, and submission time. After receiving case information uploaded by the terminal, the server converts the data into a unified internal structure for storage.

[0337] The server establishes multiple data tables or index structures for the baseline information, including tables for legal regulations, security control standards, and customer historical feedback. The server stores entry identifiers, original entry text, entry summaries, and related tag information in these tables.

[0338] During the evaluation process, the server annotates the case text and benchmark information entries with corresponding vector representations. The server uses the encoding portion of a pre-trained language model or a specialized embedding model to convert the text sequence into fixed-dimensional real-valued vectors. The server uses a vector operation library to calculate the similarity between the case text vector and the entry text vector. The server records the similarity in numerical form and uses it as a quantitative indicator of the correspondence.

[0339] When storing risk indication information, the server can create numerical and categorical fields for each risk type. For example, for legal risk indication information, the server records the risk level, the name of the relevant law, and a list of identifiers for potentially violated clauses. For information security risk indication information, the server records the category of potential vulnerabilities and the types of affected components. For customer feedback risk indication information, the server records the proportion of possible negative feedback and typical complaint topics.

[0340] The server stores report information in text format and can attach structured metadata, such as report type, target department category, and priority level, to facilitate subsequent classification and allocation.

[0341] V. Generation of Prompt Statements and Specific Examples In the prompt statement generation process, the server constructs prompt statements suitable for different stages based on case information and system predefined templates. The prompt statements used by the server during the semantic parsing stage may include the following examples: You are an expert in privacy law, cybersecurity, and market analysis.

[0342] Please read the following case information and extract the structured elements.

[0343] Case Description (Insert case text here) Target Market (Insert target market information here) Relevant laws or customer concerns (Insert relevant legal information here) Customer Feedback Summary (Insert feedback summary here) Please provide a structured description, including: 1. Industry category; 2. Data types involved (e.g., basic personal information, sensitive personal information, behavior logs, etc.); 3. Purpose of data processing (e.g., marketing, risk control, product optimization, etc.); 4. Key legal concerns (list the names of potentially relevant regulations or key terms of clauses); 5. Key security concerns (list the technical aspects that may be prone to problems); 6. Anticipate customer sensitivities (list the three aspects that customers are most likely to be concerned about).

[0344] During the comprehensive risk assessment and report generation phase, the server can use the following example prompt: You are a senior compliance and information security consultant. Based on the following structured analysis results, please generate a Chinese report for senior business executives.

[0345] Basic Case Information (Insert case summary here) Legal risk analysis (structured data) (Insert legal analysis summary here) Security risk analysis (structured data) (Insert security analysis summary here) Customer response prediction (structured data) (Insert customer analysis summary here) Please generate a report with the following requirements: 1. Use formal, clear written language; 2. Divided into the following chapters: I. Case Overview II. Legal Risk Analysis and Compliance Recommendations III. Safety Risk Analysis and Technical Recommendations IV. Customer Response Forecast and Market Recommendations V. Comprehensive Risk Assessment and Priority Action List 3. After each suggestion point, mark the priority with "Priority: High / Medium / Low".

[0346] The server inputs risk indication information and corresponding relationships into the generative artificial intelligence model in a fixed structure through the above prompt statements, thereby constraining the model output to maintain a consistent structure and dimension. This structured guidance is the key to the combination of prompt statement generation technology and model capabilities in this invention.

[0347] VI. Technical Features and Effects of Internal Server Processing Instead of simply calling a generative AI model to generate free text, the server uses prompts with explicit field requirements to instruct the model to output parsable structured information. The server can then directly map the model's output to its internal data structure, without relying on human interpretation. This design allows the output of the generative AI model to become direct data input for subsequent algorithms, reducing the significant reliance on human interpretation in traditional systems and lowering processing latency.

[0348] In its evaluation methods, the server uses a unified vectorized representation and similarity calculation mechanism to map case texts, legal provisions, safety regulations, and historical feedback into the same vector space. This unified vector space structure allows the server to replace multiple scattered keyword matching modules with a single well-optimized vector retrieval algorithm, thus gaining an advantage in computational complexity. Especially when the baseline information is large, the server can utilize the index structure to reduce retrieval latency and achieve higher processing throughput.

[0349] The server employs a closed-loop design between report generation and classification methods. First, it generates a complete, human-readable report based on risk indication information. Then, it automatically categorizes the report content according to departmental type and priority, generating additional prompts that allow the generative AI model to re-output simplified or reorganized report segments from a specific departmental perspective. For example, the server can generate a compressed report containing only security risks and technical recommendations for information security-related functions. This secondary generation relies on the structured tags obtained from the first generation, resulting in multi-layered, multi-perspective output. Because this process is automated internally by the server, and data at each stage is represented through a unified structure, data transfer and format conversion errors at the human-computer interaction boundary are reduced, improving overall processing accuracy and consistency.

[0350] During model training and usage, the server improves the quality of generated results by optimizing the error function. During pre-training and fine-tuning, the server adapts the model's parameter space to a data distribution with mixed legal, security, and customer feedback contexts, enabling the model to process multi-dimensional information in a single inference call. Compared to traditional solutions that connect multiple independent rule engines, this invention's unified model structure and unified data flow reduce intermediate format conversion and cross-module communication overhead, thus achieving more efficient utilization of computing resources and a shorter total processing time under the same hardware resource conditions.

[0351] VII. Specific Roles and Technical Coordination of Terminals and Users In this embodiment, the terminal not only acts as an input / output interface but also undertakes some data preprocessing and communication optimization tasks. The terminal performs format and constraint checks on user-input data locally, filtering out obviously erroneous or incomplete input before the data is sent to the server. This reduces the number of invalid requests, lowers server load, and prevents generative artificial intelligence models from being triggered by invalid input, indirectly improving the average response efficiency of the entire system.

[0352] Users can visually view server-generated report information through a terminal interface, including a risk score overview, a list of departmental recommendations, and a list of priority actions. Users then create new case information based on corrective measures taken after implementing the report (e.g., adding security controls or supplementing compliance modules) and re-enter it into the server. The server uses the same data structure and model reasoning path to compare the changes in risk indications between the old and new cases, providing users with multi-version comparison results. Through this iterative process, the server can perform differential analysis across multiple versions, improving the accuracy of risk reduction identification.

[0353] VIII. Explanation of Technical Effects and Causal Relationship By mapping case information, legal benchmark information, security benchmark information, and customer feedback information into a unified vector space, and utilizing generative artificial intelligence models for multidimensional semantic parsing and report generation, the system can achieve the following technical effects under the same hardware platform and network conditions: The server reduces the processing time from data reception to report output. A unified data representation and model reduce intermediate steps and avoid redundant parsing across multiple engines and formats. The server can simultaneously perform preliminary identification of multiple risks in a single model call, reducing the number of model calls and thus lowering overall inference time.

[0354] The server improves the accuracy and consistency of risk identification. Through pre-training and instruction fine-tuning, the model learns patterns in fine-grained legal entries, security controls, and customer feedback. Under the constraint of unified prompts, the model output is more likely to cover necessary risk dimensions, reducing the omission of key information and making the overall system output more stable.

[0355] The server improves the efficiency of data management and report distribution. By automatically generating additional prompts through classification methods, the server can generate targeted report content for different processing entities and automatically allocate it, reducing manual editing. The server adds priority information to the allocation information, enabling each processing entity to schedule resources based on the sorting information provided by the server, indirectly improving the processing efficiency at the entire organizational level.

[0356] In its overall architecture design, the server tightly integrates generative artificial intelligence models with prompt generation technology, automating risk analysis and implementing a novel data flow control and computational process optimization scheme within the computer. This scheme is not a simple replication of manual processes, but rather leverages the semantic representation capabilities and structured prompt constraint mechanisms of deep neural networks to achieve cross-domain comprehensive risk assessment and multi-perspective report generation—tasks difficult for traditional rule-based systems. Therefore, this invention represents an improvement over existing technologies in multiple technical dimensions, including computational speed, result accuracy, data management, and communication efficiency.

[0357] use Figure 13 The processing flow is explained.

[0358] Step 1: Users input case information on the terminal. Users input case-related information through the terminal interface as the initial input for subsequent processing. Inputs include: a textual description of the case, the target market, the names or keywords of relevant regulations, and summaries of historical customer feedback.

[0359] After receiving input, the terminal performs local validation on the input data (e.g., checking whether required fields are empty, whether the text length exceeds the limit, and whether the field format is correct).

[0360] The terminal will integrate multiple user inputs into an internal data object, such as a record containing multiple fields, through a graphical interface component.

[0361] The input for this step is the text and options that the user enters item by item in the interface, and the output is structured case data generated internally by the terminal (such as a data object containing several key-value pairs).

[0362] Step 2: The terminal preprocesses the data and sends it to the server. The terminal takes the structured case data obtained in step 1 as input and preprocesses it, including removing leading and trailing spaces, standardizing the encoding format, and performing simple normalization on certain fields (e.g., converting country / region names into unified codes).

[0363] The terminal then uses the communication module to serialize the processed data into a data packet, for example, constructing a message with field names and field values.

[0364] The terminal invokes an encrypted communication protocol to encrypt data packets and send them to the server over the network.

[0365] The input for this step is structured case data. The terminal performs data processing operations such as format standardization and encryption locally, and the output is an encrypted request message sent to the server via a network interface.

[0366] Step 3: The server receives and stores the original case data. The server receives encrypted request messages from the terminal via a network interface, using them as input. The server then uses a communication protocol stack to decrypt and unpack the messages, recovering the structured case data.

[0367] The server performs syntax-level parsing of the data, checking if all fields are complete and if the data type meets expectations. If the data is valid, the server calls the data access module to write the case record into the case table in the database and assign it a unique case identifier.

[0368] After the server completes the write operation, it also records the receiving event in the log system, including the timestamp, user ID, and case ID.

[0369] The input for this step is the encrypted message received from the terminal. The server converts it into an internal structured record through data processing operations such as decryption and parsing. The output is a new case record in the database and the corresponding case identifier.

[0370] Step 4: The server generates prompts for semantic parsing. The server takes case records stored in the database as input and reads the case text description, target market, relevant legal fields, and customer feedback summary fields.

[0371] The server invokes the prompt generation module to embed these fields into a predefined prompt template, concatenating them into a piece of natural language text. This prompt text clearly lists the structured elements that the generative AI model needs to output, such as industry category, data type, and legal concerns.

[0372] During the generation process, the server performs string replacement and text concatenation operations to ensure that each field is inserted into the specified position in the template and to ensure that the overall sentence is fluent.

[0373] The input for this step is multiple fields from the case record. The server processes the data through template filling and string manipulation, and the output is a complete prompt text, which serves as one of the inputs to the generative artificial intelligence model.

[0374] Step 5: The server invokes a generative artificial intelligence model for preliminary semantic analysis. The server takes the prompt generated in step 4 as input, and also sends the case text and other content into the reasoning interface of the generative artificial intelligence model.

[0375] The server segments the input text into sub-words or tokens using an embedding module and converts them into vector representations that the model can process. Subsequently, the server performs multi-layer self-attention operations and feedforward network operations on a processor or graphics processing unit to obtain a hidden representation of the input text.

[0376] The server uses the model's output layer to generate structured explanatory text in sequence, such as itemized text containing industry categories, data types, data processing purposes, legal concerns, security concerns, and expected customer sensitivities.

[0377] The server performs syntactic analysis on the generated text, mapping each element to internal structure fields to form an intermediate analysis result object.

[0378] The input for this step is the prompt statement along with case-related text. The server obtains structured information text from the model through data operations such as vectorization, attention calculation, and sequence generation. The output is an intermediate structured analysis result from within the server.

[0379] Step 6: Correspondence between server calculations and legal benchmark information The server takes the legal concerns field from the intermediate structured analysis results as input and reads legal benchmark information entries related to these concerns from the database.

[0380] The server performs embedding calculations on the case description text and the summaries of each legal provision, mapping them into numerical vectors respectively.

[0381] The server uses a similarity calculation algorithm (such as cosine similarity) to perform dot product and normalization operations on each pair of vectors to obtain the relevance score between each legal provision and the case.

[0382] The server sorts the cases based on their scores, selects several legal provisions with high relevance, and records these provisions and their scores as a correspondence between the cases and legal benchmarks.

[0383] The input for this step is the intermediate analysis results and legal benchmark information. The server processes the data through operations such as vector embedding and similarity calculation, and the output is a list of legal entries with relevance scores.

[0384] Step 7: Correspondence between server computing and security baseline information The server takes the security concern field and case technical description from the intermediate structured analysis results as input, and loads a list of security control standards from the database, such as encryption requirements, access control requirements, and log management requirements.

[0385] The server generates simplified text descriptions for security matching, organizing elements such as data processing methods, storage locations, and access patterns involved in the case into short sentences.

[0386] The server performs embedding operations on these short phrases and security control entry texts, and uses the same similarity metric to calculate the similarity between the case security scenario and each control item.

[0387] The server identifies security controls that are under-covered or not yet met by the cases, assigns a risk level to each control, and formalizes these results into security risk indication information.

[0388] The input for this step is security-related case information and a security standard table. The server performs data calculations such as text embedding and similarity assessment, and the output is a set of security risk indication items and their corresponding risk levels.

[0389] Step 8: The server generates comprehensive risk indication information. The server takes legal correspondence, security correspondence, and customer reaction-related fields (such as expected customer sensitivities) as input to construct a unified set of risk indication information.

[0390] The server calculates a sub-score for each type of risk, such as classifying risks into different levels based on correlation scores and preset thresholds. The server can combine multiple sub-indicators into a unified risk score through weighted summation or rule-based scoring methods.

[0391] The server records these scores and ratings into the data structure corresponding to the case, providing a quantitative basis for subsequent report generation and task allocation.

[0392] The input for this step is the correspondence between multiple dimensions and intermediate analysis results. The server processes the data through weighted combination, threshold comparison and classification, and the output is a comprehensive risk structure that includes legal risk indication information, information security risk indication information and customer reaction risk indication information.

[0393] Step 9: The server generates a report-level prompt and calls a generative artificial intelligence model to generate a report. The server takes comprehensive risk indication information and corresponding relationship data as input, and reads basic case information and historical analysis results.

[0394] The server builds prompt text for report generation, embedding content such as case summary, legal analysis overview, security analysis overview, and customer response prediction into a pre-designed report template.

[0395] The server inputs this long text prompt into the generative artificial intelligence model, where it also performs tokenization, vectorization, and multi-layer neural network operations.

[0396] The server retrieves the long text report generated by the model. The report is output according to the chapter structure required by the prompt statement. Each chapter contains explanatory text and specific suggestions, and priority information is marked at the end of the suggestions.

[0397] The server performs basic checks on the generated report, such as checking whether chapter titles are complete and whether key sections are empty.

[0398] The input for this step is the combined structured data of cases and risks. The server performs data operations such as template filling and generative artificial intelligence model reasoning, and the output is a complete report text written in natural language.

[0399] Step 10: The server categorizes the report content and generates additional prompts. The server takes the report text generated in step 9 as input, parses the report content in segments, and maps different parts to different department categories and task types.

[0400] The server sorts the recommended items according to their priority based on the priority field tagged in the report, generating a structured classification result grouped by department and priority.

[0401] The server then uses these classification results to create additional prompts for different processing entities, such as simplified report prompts for legal functions or rectification checklist prompts for security functions.

[0402] The server stores the classification results and additional prompts in the database and provides input for subsequent secondary model calls or external interface calls.

[0403] The input for this step is the complete report text. The server performs data processing operations such as text parsing, classification mapping, and structured assembly. The output is the results categorized by department and priority, along with corresponding additional prompt text.

[0404] Step 11: The server generates allocation information and prepares to output it to the terminal. The server takes the classification results and risk level as input and generates allocation information records for each department category and processing entity.

[0405] The server labels each allocation record with fields such as case identifier, target department, task description, priority, and suggested execution period.

[0406] The server can convert some allocation records into a data format that can be recognized by external systems (such as task management systems or work order systems) based on system configuration, thereby reducing manual transcription work.

[0407] The server ultimately constructs a response message, packaging the complete report text, allocation information list, and necessary metadata into a transmittable data structure, ready to be sent to the terminal.

[0408] The input for this step is the categorized report structure and priority information. The server performs data processing operations such as field combination and record construction, and the output is response data containing report and assignment information.

[0409] Step 12: The terminal receives and displays the report and allocation results returned by the server. The terminal receives the response message sent by the server from the network interface, parses it as input, and recovers the report text and the list of allocation information.

[0410] The terminal displays the report text as a multi-chapter document through the interface rendering engine and highlights high-priority items with visual markers.

[0411] The terminal lists the allocation information in a separate interface area, sorted by department category and priority, so that users can easily see the matters that different processing entities are responsible for.

[0412] The terminal can provide users with quick access points, such as buttons or links, based on the assigned information, to help users forward a task to the corresponding business system.

[0413] The input for this step is the response data returned by the server. The terminal processes the data through parsing, sorting, and rendering, and outputs a report page and assignment list that are visually displayed on the user's screen.

[0414] Step 13: Users revise and resubmit based on the report and allocation information. Users read reports and assign information on the terminal interface, compare potential problems and suggestions with the actual business situation, and use them as input for subsequent processing.

[0415] Users can adjust product design, system configuration, or process settings based on suggestions generated by the server, and enter the modified case information or new information in the terminal, such as "database encryption has been enabled" or "privacy policy terms have been improved".

[0416] When a user submits new case data through the terminal, it triggers a new round of processing on the server, enabling the system to compare and analyze the risk indication information before and after the modification.

[0417] The input for this step is the business adjustment results made by the user based on the report and the updated case data. The user and the terminal jointly provide the server with a new round of data sources by resending structured input data. The output is a new case submission request, which enters the next complete processing cycle of the system.

[0418] 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".

[0419] Existing information processing systems typically rely solely on keyword matching or fixed rule-based judgment of user-submitted text content to allocate information to corresponding business departments, or simply prioritize processing based on a basic priority field. These systems suffer from the following technical problems: First, they lack the capacity to fully understand complex natural language and have limited ability to extract multi-dimensional semantic information related to legal risks, security risks, and user reactions, resulting in low accuracy in information classification and risk identification. Second, they generally lack automatic analysis and correlation mechanisms for user emotional states, failing to dynamically adjust information processing priorities and response strategies based on real-time user emotions, leading to inefficient resource allocation and poor user experience. Third, existing systems often rely on manual writing or template replacement for generating text for internal processing or external feedback, lacking a context-based dynamic generation mechanism. This makes it difficult to automatically generate targeted, highly executable risk reduction and value-maximizing solutions while ensuring compliance and security. Therefore, how to introduce an information processing mechanism within the computer architecture that can leverage generative artificial intelligence models and sentiment analysis techniques to work together, enabling the processor to automatically generate structured prompts and uniformly drive complex processing flows such as compliance assessment, security assessment, sentiment perception, departmental triage, and priority control, thereby improving information understanding, classification accuracy, risk control capabilities, and response generation quality at the system level, has become a pressing technical challenge in this field.

[0420] 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.

[0421] In this invention, the server includes: a preprocessing unit for receiving transaction information, case information, or consultation information submitted by an input device via a communication processing device, and performing natural language preprocessing on the received information to generate structured information; an analysis unit for automatically constructing prompt statements for input to a generative artificial intelligence model based on the structured information and regulatory information, security requirement information, and historical response information stored in a storage device, and performing multi-dimensional analysis of the structured information through the generative artificial intelligence model to obtain analysis results regarding legal concerns, security impacts, and user responses; and a prompt statement for generating classification information including department type and processing type based on the analysis results. The system includes: a classification unit that automatically maps the information to classification instructions for business processing departments; an emotion estimation unit that applies emotion analysis technology to voice, image, or text information from input devices or emotion recognition devices to estimate the user's emotional state; a priority control unit that generates prompt statements containing management information such as processing priorities or response policies based on the estimated emotional state and the analysis results, and dynamically corrects the classification instructions according to the processing priorities; and an output unit that generates improvement suggestion information containing risk reduction schemes and value maximization schemes based on the analysis results and the management information, and generates output data for the display device. This allows for deep semantic understanding and structured processing of user information within the server using a generative artificial intelligence model. It automatically generates consistent prompts and instructions for departmental workflow and priority control, achieving an integrated computational process from input reception, risk assessment, and emotion perception to response generation. This significantly improves information classification accuracy and processing efficiency, enhances the system's ability to identify and manage legal and security risks, and adaptively generates more appropriate explanatory and improvement suggestions based on the user's emotional state. Overall, it improves the information processing performance and user interaction experience of the computer system in complex business environments.

[0422] "Information processing device" refers to an electronic device used to receive, process and output digital information, including servers, terminal equipment and the operating systems and applications running on them.

[0423] A processor is a hardware or virtual computing unit in an information processing device that executes program instructions to perform calculations on data and control the flow of information. It includes a central processing unit and a programmable logic unit.

[0424] "Input device" refers to an interactive component used to obtain transaction information, case information or consultation information from a user, including keyboards, touch screens, voice input interfaces and web forms.

[0425] "Communication processing device" refers to the hardware and software components used to send and receive data between the information processing device and external devices, including network interfaces, telecommunications protocol stacks, and communication middleware.

[0426] Natural Language Processing (NLP) refers to the technology of using computer programs to perform word segmentation, syntactic analysis, semantic understanding, and structured transformation of human language text.

[0427] "Structured information" refers to a representation of unstructured text that has been processed and has predefined fields or data formats, which is used for subsequent automatic analysis and calculation.

[0428] "Storage device" refers to data storage resources used to store regulatory information, safety requirement information, historical response information, and intermediate processing results, including disk storage, semiconductor storage, and database systems.

[0429] "Regulatory information" refers to textual or rule data related to laws, administrative regulations, industry standards, and internal compliance rules, used to determine legal concerns.

[0430] "Security requirements information" refers to the technical or management requirements that are set in advance to ensure the security of information systems and data, including data such as encryption standards, access control policies, and audit requirements.

[0431] "Historical response information" refers to user evaluations, behavioral feedback, and processing results recorded for past transactions or cases, which are used to analyze user response trends.

[0432] "Generative AI models" refer to AI models that, through training, can automatically generate output content such as text based on input prompts, including language models based on deep learning.

[0433] "Prompt statements" refer to textual instructions or contextual descriptions that serve as input to generative artificial intelligence models, and are natural language content used to guide the model in performing specific analysis or generation tasks.

[0434] "Analysis results" refer to evaluation data on legal concerns, security impacts, and user reactions, output by generative artificial intelligence models or other analytical programs after processing structured information.

[0435] "Legal concerns" refer to situations or factors related to the content of information that may lead to the risk of violating laws, regulations, or mandatory norms.

[0436] "Security impact" refers to the risks or adverse consequences that information processing activities may cause to information assets, system operations, and network environments in terms of confidentiality, integrity, and availability.

[0437] "User feedback" refers to the results of user feedback on transactions, cases, or services in terms of attitude, satisfaction, and behavior, including evaluations, complaints, and usage behavior patterns.

[0438] "Classification information" refers to tagged data used to identify the type of department and processing of information, and is used to guide the automatic allocation of information.

[0439] "Business processing departments" refer to functional units within an organization that are responsible for processing specific types of information or tasks, including legal departments, security departments, customer service departments, and operations departments.

[0440] "Classification instruction information" refers to control data or instructions generated based on classification information, used to control the routing or allocation of information to the corresponding business processing departments.

[0441] "Emotion recognition device" refers to the hardware and software components used to collect and identify the emotional state of a user's voice, images, or text, including microphones, cameras, and emotion analysis programs.

[0442] "Emotion analysis technology" refers to the technology that uses algorithms to analyze voice features, facial expression features, or text features to infer the category and intensity of a user's emotions.

[0443] "User emotional state" refers to the type and intensity of emotions a user exhibits at a specific point in time, including emotional labels such as anxiety, anger, and satisfaction.

[0444] The “emotion inference unit” refers to a functional module that applies emotion analysis technology to input voice, image, or text information to output the user’s emotional state.

[0445] "Management information" refers to a set of control parameters generated based on the analysis results and the user's emotional state, used to control the priority of information processing, response strategies, and output style.

[0446] "Processing priority" refers to a hierarchical indicator used to represent the relative urgency of different information in terms of processing order or resource allocation.

[0447] "Response policy" refers to the strategy for setting the tone, level of detail, and content focus when generating explanatory and recommended information for users or internal personnel.

[0448] The "Priority Control Unit" is a functional module that generates management information based on the user's emotional state and analysis results, and corrects the classification instructions accordingly to adjust the processing order and resource allocation.

[0449] "Improvement suggestion information" refers to the suggestions generated based on the analysis results and management information, which are used to reduce risks and increase business value, including risk reduction plans and value maximization plans.

[0450] A “risk reduction plan” refers to specific measures or strategies proposed to reduce or control risks in response to identified legal concerns or security impacts.

[0451] "Value maximization solutions" refer to optimization measures used to improve business performance, user satisfaction, or resource utilization efficiency while meeting compliance and security requirements.

[0452] "Output data" refers to formatted data generated to present improvement suggestions, analysis results, and related prompts on display devices or other output terminals.

[0453] "Display device" refers to a device used to display output data to users or operators, including displays, terminal interfaces or visualization panels.

[0454] In various embodiments of this invention, the server, terminal, and user work collaboratively, combining generative artificial intelligence models with sentiment analysis technology to perform multi-dimensional technical processing on user-input information. The following describes the form of the invention in conjunction with its hardware structure, software modules, data structure, and algorithm flow.

[0455] I. Overall Hardware and Software Composition of the System In this invention, the server operates as the core information processing device. The server includes a multi-core central processing unit, semiconductor memory, persistent storage, and a network interface. The server runs application services, a database system, and generative artificial intelligence model inference services on an operating system. The server can employ a general-purpose high-performance computing platform and can be connected to a graphics processing unit or a dedicated accelerator to accelerate the inference and training of deep neural networks.

[0456] In this invention, the terminal serves as a user-side information collection and result display device. The terminal can be a smartphone, tablet, or general-purpose computing device. It includes a display unit, an input unit, a microphone, a camera, and a local processing unit. The terminal runs a client application or web browser on an operating system to present the user interface, collect user input and emotion-related data, and interact with the server via a communication network.

[0457] Users input transaction information, case information, or consultation information through the terminal, and can provide emotion-related information to the system via voice, text, or images. Users can view the risk assessment results and improvement suggestions returned by the server on the terminal, and choose whether to adopt the relevant suggestions.

[0458] At the software level, a server comprises multiple functional modules. These modules can run on the same physical processor as program modules or be deployed as distributed microservices. In different implementations, a server includes at least the following modules: a preprocessing module, a generative artificial intelligence model interface module, a sentiment analysis module, a classification and priority control module, an output generation module, and a data storage and indexing module.

[0459] II. Specific Forms of Program Modules and Data Processing 1. Server-side preprocessing and structured information generation After receiving user input from the terminal, the server sequentially performs data cleaning and structuring using a preprocessing module. The server employs natural language processing techniques, including word segmentation, part-of-speech tagging, syntactic analysis, and named entity recognition. In one implementation, the server uses a deep bidirectional encoder-based model to encode the text and extract semantic feature vectors. The server then converts the raw text into a data structure containing field identifiers and attribute values. The fields include at least: information category, subject, amount, time, location information, and key actions.

[0460] When processing audio data, the server uses speech recognition software to convert the speech signal into text, and then passes the text to the same natural language preprocessing pipeline for further processing. When processing image data, the server can extract facial expression categories through an expression analysis network, which is then used in conjunction with the sentiment analysis module.

[0461] Through the aforementioned structured processing, the server transforms natural language information, which is inherently difficult to compute directly, into a multi-dimensional feature vector and field set located in a unified feature space. This allows subsequent generative AI model calls and rule-based reasoning to be performed on a unified data structure. This structured approach reduces repetitive text parsing calculations in subsequent steps, thereby lowering overall computational complexity and latency.

[0462] 2. Generative Artificial Intelligence Model Interface and Internal Structure In this invention, the server uses a generative artificial intelligence model to comprehensively analyze and generate structured and contextual information. In a preferred embodiment, the server employs a sequence-to-sequence network structure based on a multi-layer self-attention mechanism. This network consists of an input embedding layer, a multi-layer self-attention encoder, a multi-layer self-attention decoder, and an output mapping layer. During the model training phase, the server uses a large-scale corpus to optimize the model's parameters; during the online operation phase, the server only performs forward inference, taking the prompt statement as input to obtain natural language output.

[0463] During the model training phase, the server employs supervised learning methods. It takes labeled training samples as input and uses the difference between the actual output and the model output as the loss function. The server can use cross-entropy loss as the primary loss function, while adding a regularization term to prevent overfitting. The server uses a gradient descent-based optimization algorithm to iteratively update the model parameters in the parameter space.

[0464] During runtime, the server receives prompts through the model interface module, converts them into word sequences, maps them to embedding vectors, and feeds them into the encoder. The encoder performs self-attention calculations on each word in the prompts to obtain context-sensitive semantic representations. Based on the encoder output, the decoder predicts the next output token at each time step, selectively focusing on the input content through a multi-head attention mechanism, enabling the model to output refined results based on different factors such as legal concerns, security implications, and user reactions.

[0465] By using this context-based self-attention generative AI model structure, the server can integrate multiple information sources in a single inference, improving the accuracy of comprehensive judgment, reducing the cumulative error caused by multiple rounds of rule processing, and also increasing the inference speed.

[0466] 3. Construction of prompt statements and multi-task joint processing In this invention, the server drives a generative artificial intelligence model through multi-task operation by constructing prompts for different purposes. In one embodiment, the server merges legal assessment, security assessment, and user reaction analysis into a unified generation task. When constructing prompts, the server adds task instructions to the text, enabling the model to output different types of results simultaneously.

[0467] When conducting a legal risk assessment, the server can construct the following prompt statement: "You are a senior legal compliance consultant. Please analyze the following electronic transaction information and relevant legal provisions:" - Transaction information: ... - Relevant legal provisions: ... Please assess whether the transaction carries any legal risks (high / medium / low), list the key reasons, and provide one or two suggestions to mitigate the risks. When predicting customer satisfaction, the server can construct the following prompt statement: "The following is a summary of several past customer reviews for this product: ... Please predict the new customer's satisfaction level with this transaction (high / medium / low) and explain the main reasons." When generating suggestions for improving a case, the server can construct the following prompt statement: "Below is a summary of the key terms of a contract and the identified risk points. Please provide a suggested revision text for each risk clause without changing the essence of the business, along with a brief reason, for relevant personnel to refer to." When generating emotion perception and description, the server can construct the following prompt statements: "The user's current emotion is 'anxious,' and below is the user's business description: '...'. In your response, please prioritize alleviating the user's anxiety by providing a clear, easy-to-understand, and low-risk solution, avoiding overly technical or stressful language." By embedding task descriptions and contextual information in prompt statements, the server utilizes the same generative artificial intelligence model to achieve multi-task joint reasoning, reducing the need to build separate models for different functions, thereby saving storage space, improving operating efficiency, and reducing communication overhead caused by cross-module data conversion.

[0468] 4. Refinement of Sentiment Analysis and Prioritization Techniques In this invention, the server quantifies and estimates the user's emotional state through a sentiment analysis module. In one embodiment, the server uses a multimodal sentiment recognition network. This network includes a speech feature subnetwork and a text feature subnetwork. The server extracts feature vectors such as pitch, loudness, and speech rate from the speech data, and extracts features such as sentiment words, negation words, and intensity words from the text. The two are concatenated and fed into a multi-layer feedforward network or convolutional network for sentiment classification. The server ultimately obtains sentiment labels and corresponding probability distributions.

[0469] The server combines sentiment tags with the output of a generative AI model to adjust the processing priority and interpretation style of information. In its priority control module, the server sets weight matrices for different sentiment states and question types. The server calculates a comprehensive score based on the matrix and discretizes this score into different priority levels. The server then reorders the queue of processes according to priority, ensuring that information with strong emotions and involving security issues is processed faster, achieving technical-level scheduling optimization.

[0470] Through this emotion-driven scheduling mechanism, the server not only changes the traditional "first-come, first-served" or fixed-rule sorting method, but also directly maps the user's emotional state to the priority parameters inside the computer system. This enables dynamic and automatic adjustment of resource allocation, thereby processing more urgent and high-risk information per unit of time, reducing the probability of potential accidents, and improving the overall system response quality.

[0471] III. Classification, Routing, and Specific Forms of Output Generation In the classification module, the server maps the department descriptions and processing categories output by the generative AI model to internal encoding. The server can maintain a mapping table to match the natural language phrases generated by the model with predefined department identifiers. The server uses vector similarity calculations during matching, rather than being limited to exact string equality. This vector matching method tolerates diversity in descriptions, improving the robustness and accuracy of classification.

[0472] In the output generation module, the server combines the natural language results generated by the model with structured control information. The server can predefine labels for each type of suggestion, allowing the terminal to display different types of content using different visualization methods, thereby improving the comprehensibility of the human-computer interface. During output, the server simultaneously records intermediate features and evaluation results, facilitating subsequent statistical analysis and model improvement.

[0473] IV. Technical Effects and Causal Relationships Through the above series of technical solutions, the server achieves the following verifiable technical effects within the computer: The server improves the inference accuracy of generative AI models when facing various business scenarios by using a unified structured preprocessing and multi-task prompt statement construction mechanism, avoiding redundant calculations caused by multiple model calls, thereby reducing inference latency.

[0474] The server uses an emotion analysis module and a priority control module to transform emotional information, which is originally perceptible only to human operators, into calculable priority parameters, thus achieving automated task scheduling. Compared with traditional queue systems that sort only by timestamps, this mechanism significantly improves response time to urgent issues.

[0475] By explicitly injecting task instructions, emotional states, and structured features into the prompt statements, the server enables the generative AI model to form multidimensional contextual relationships during the encoding phase. Since the model has already learned the statistical relationships between these fields through a large number of samples during the training phase, it can output risk assessments, explanations, and improvement suggestions in a single forward computation during inference, reducing the need for multiple database queries and phased rule processing.

[0476] The server solves the parsing difficulties caused by the diversity of natural language department names through a vector space matching classification mechanism, making the information routing process more robust; under the emotion-driven scheduling strategy, the server explicitly incorporates user experience indicators into the internal scheduling logic of the system, realizing the unified optimization of performance indicators and user perception indicators.

[0477] V. Different Implementation Forms and Alternative Methods In another implementation, the server can deploy generative AI models as local inference services, directly loading deep model parameters using local acceleration hardware, reducing cross-network call latency, and is suitable for latency-sensitive scenarios.

[0478] In another implementation, the server can divide legal concern assessment and security impact assessment into two different sub-models. In the first sub-model, the server focuses on learning the relationship between legal texts and contract texts; in the second sub-model, it focuses on learning the relationship between security event logs and configuration items. Through a unified prompt interface and a unified output fusion module, the server integrates the results of multiple models into a single unified report, improving the comprehensiveness of the assessment.

[0479] In different implementation forms, the terminal can be responsible only for text input and result display, leaving all speech recognition and emotion recognition to the server. Alternatively, it can perform preliminary speech-to-text and facial expression recognition locally, and then upload the data to the server after data dimensionality reduction, thereby reducing network bandwidth usage and improving communication efficiency.

[0480] In typical use cases, users can enter the following natural language request on the terminal: "Please analyze whether this cross-border order violates any relevant laws and predict whether the customer will be satisfied with this purchase." Upon receiving the request, the server integrates the relevant transaction data and historical evaluation data into structured information and constructs a prompt statement: "Please analyze whether the following electronic transactions violate the data protection laws of any country, and predict the customer's satisfaction level with the transaction outcome based on past customer feedback: 'Transaction details: ……' 'Historical feedback: ……'." The server generates evaluation results and improvement suggestions through a generative artificial intelligence model. The terminal presents the results to the user, who can then modify the transaction terms or take additional security measures.

[0481] Through the aforementioned implementation, the server, terminal, and user form a closed-loop information processing system driven by a generative artificial intelligence model and prompts. This closed loop changes the traditional processing method that relies solely on rules and human judgment. It achieves multi-signal fusion, multi-task joint reasoning, and emotion-driven scheduling within the computer, thereby achieving comprehensive technological improvements in processing accuracy, response speed, communication load, and resource utilization efficiency, rather than simply automating manual business processes.

[0482] use Figure 14 The processing flow is explained.

[0483] Step 1: Users input information via the terminal. Users input transaction information, case information, or inquiry information into the terminal interface, and can input text via keyboard, voice via microphone, or image via camera. Input includes: text content (e.g., order details, contract summary, problem description), optional voice data, and image data. The terminal performs voice acquisition, encodes the audio stream into digital audio data, and performs basic compression processing on the image data. Based on the input content, the terminal generates a request data packet containing text fields, a timestamp, user identifier, and device identifier. The terminal sends this request data packet to the server via an encrypted communication channel. The output is the raw user request data sent to the server.

[0484] Step 2: The server receives and preprocesses user information. The server receives user request data from the terminal via the network interface. The input is a request data packet containing text, audio, images, and metadata. The server first performs integrity and format checks on the data packet, then extracts the text and performs natural language preprocessing, including character normalization, sentence segmentation, word segmentation, stop word removal, and part-of-speech tagging. For requests containing audio data, the server calls a speech recognition program to convert the audio into text and appends this text to the original text field. For requests containing image data, the server can call an image analysis program to extract facial regions and facial expression feature vectors as input for subsequent sentiment analysis. Based on the preprocessed text, the server uses a named entity recognition algorithm to extract entities and organizes the results into structured information objects. The fields include at least: information category, subject, amount, time, location, and key actions. The output is a structured information object and optional audio text and facial expression feature vectors.

[0485] Step 3: The server integrates external regulatory information and historical response information. The server takes a structured information object as input and retrieves relevant regulatory information, security requirements, and historical feedback information from storage. Based on the geographic and business type fields in the structured information, the server performs database queries to obtain corresponding legal provisions summaries, security baseline entries, and historical user evaluation records. The server performs simple aggregation operations on the historical evaluation records, such as calculating the proportion of positive evaluations, average scores, and common keywords. The server merges the regulatory text fragments, security requirement entries, and historical feedback statistics with the structured information object to form extended contextual data. The output is a comprehensive data object containing both business structured information and environmental contextual information.

[0486] Step 4: The server constructs prompts for generative artificial intelligence models. The server takes a comprehensive data object as input and assembles prompts from a generative artificial intelligence model according to predefined templates. The server selects different prompt templates based on the information category; for example, it uses a compliance and satisfaction joint analysis template for transaction requests and a risk assessment and modification suggestion template for case texts. The server transcribes structured fields into natural language descriptions, inserts regulatory texts and security requirements into the "Relevant Rules" paragraph, and inserts historical response summaries into the "Historical Feedback" paragraph. The server then generates a continuous natural language text as the prompt, for example: "Please analyze whether the following electronic transactions violate the data protection laws of any country, and predict the customer's satisfaction level with the transaction outcome based on past customer feedback: 'Transaction details: ……' 'Historical feedback: ……'." The server performs length checks and encoding on the generated prompt statements. The output is a formatted prompt statement text.

[0487] Step 5: The server invokes a generative artificial intelligence model for multi-dimensional analysis. The server passes the prompt statement as input to the generative AI model interface module. The server performs word segmentation or sub-word splitting on the prompt statement, mapping the text to a vector sequence, and then runs a multi-layer self-attention network on computationally accelerated hardware. During forward inference, the model calculates attention weights for different parts of the prompt statement (business description, regulatory text, historical feedback) and progressively generates response text during the decoding phase. The server receives parsed results in natural language form from the model output, which typically includes legal risk levels, security impact analysis, user reaction predictions, and corresponding reasons. The server parses the output text, extracting markers from key locations (e.g., "risk level: high / medium / low", "satisfaction level: high / medium / low") and mapping them to structured fields. The output is a structured dataset of legal concern assessment results, security impact assessment results, and user reaction prediction results.

[0488] Step 6: The server performs independent security rule checks and consolidation assessments. The server takes structured business information and security requirements as input and performs rule-based security checks independently of the generative AI model. Based on fields such as transaction amount, data sensitivity level, and network environment, the server matches security rules one by one, for example, whether multi-factor authentication is required or whether such data transmission is permitted under the current network environment. The server scores and accumulates the matched rules to form a security risk score, and compares and merges this score with the security impact analysis output by the generative AI model. The server uses a weighted fusion algorithm to generate a unified security risk level and trigger rule list. The output is the fused security risk assessment result object.

[0489] Step 7: The server performs sentiment analysis to infer the user's emotional state. The server takes user text, optional speech text, and facial expression feature vectors as input to infer the user's emotional state. For the text portion, the server uses a sentiment classification model to extract sentiment tendency and intensity; for the speech portion, audio features (fundamental frequency, energy, speech rate) are input into an acoustic sentiment classification sub-network; for the image portion, facial expression feature vectors are input into an expression classification sub-network. The server fuses the sentiment probability distributions output from each sub-network, using a weighted average or voting method to obtain the final sentiment label (e.g., "anxiety," "anger," "satisfaction") and corresponding confidence score. The output is a sentiment state object containing the sentiment category and confidence score.

[0490] Step 8: The server generates management information based on the emotional state and the analysis results. The server takes emotional state objects, legal concern assessment results, security risk assessment results, and user response prediction results as input to generate management information for controlling processing priorities and response strategies. The server first looks up predefined priority weights based on emotional category and confidence level, then combines these with legal and security risk levels to calculate a comprehensive priority score. The server maps this score to discrete priority levels (e.g., high, medium, low) and sets the response tone (soothing, objective, concise) and level of detail based on the emotional category. The server organizes these parameters into management information objects, including processing priority, suggested response tone, description length, and risk warning strength. The output is a management information object.

[0491] Step 9: The server generates classification indication information and adjusts the allocation strategy. The server takes the comprehensive analysis results and management information objects as input, maps the department descriptions in natural language to internal department codes, and generates classification indication information. Based on the department suggestions and processing type suggestions output by the generative artificial intelligence model, the server uses vector similarity or a keyword mapping table to determine the most matching business processing department code. The server incorporates processing priorities from the management information to refine the classification results: when user emotions are intense and the risk level is high, the server may force the information to be assigned to a specific high-priority processing queue. The server ultimately generates classification indication information containing the target department code, processing type code, and priority level. The output is classification indication data that can be directly used for task scheduling and routing.

[0492] Step 10: The server generates improvement suggestions and explanatory output. The server takes legal concern assessment results, security risk assessment results, user response prediction results, and management information as input to reconstruct prompts for the generative artificial intelligence model. The server explicitly requests the generation of suggestions that include risk reduction and value maximization solutions, and specifies the response tone and level of detail, for example: Based on the following analysis, please provide concise and reassuring suggestions for improvement to reduce risk and increase customer satisfaction: - Legal risk: Medium, Reason: ... - Security risk: Low, reason: ... - Expected customer satisfaction: Lower than average, reasons: ... Please provide no more than three specific, actionable measures. The server inputs the prompt statement into a generative artificial intelligence model to obtain improvement suggestions in natural language. The server then formats the output, such as organizing it by item number and adding label fields, to create an output data structure for display on the terminal. The output is a visual representation containing explanatory text and a list of improvement suggestions.

[0493] Step 11: The server sends the results to the terminal, which then displays them. The server takes classification instructions, management information objects, and output data structures as input, packages them into a response message, and sends it to the terminal. The response includes: a risk assessment summary, sentiment analysis, improvement suggestions, and optional internal routing information. Upon receiving the response, the terminal parses the output data into interface elements and displays them on the screen with titles, summaries, and expandable details. Based on the level of detail and tone parameters in the management information, the terminal selects different interface templates and color cues; for example, high-risk information uses a prominent warning color. The terminal then presents the final interface to the user, outputting visual risk and recommendation information.

[0494] Step 12: Users follow the prompts to perform the operation and provide feedback on the results. After reading the server-generated instructions and improvement suggestions on the terminal, the user enters subsequent operation commands, such as adopting a suggestion, modifying transaction conditions, adjusting security settings, or submitting further questions. The input includes the user's operation selection and possible new text descriptions. The terminal then resends this feedback information to the server. Upon receiving the feedback, the server updates its historical response database for subsequent model optimization and rule adjustments, thereby improving evaluation accuracy and suggestion quality over long-term operation. The output includes the updated historical records and an optional confirmation response.

[0495] 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.

[0496] 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.

[0497] 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.

[0498] 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.

[0499] 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.

[0500] Second Implementation Method Figure 3 An example of the configuration of the data processing system 210 according to the second embodiment is shown.

[0501] 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.

[0502] 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).

[0503] 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.

[0504] 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.

[0505] 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).

[0506] 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.

[0507] 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.

[0508] 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.

[0509] 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).

[0510] 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.

[0511] 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".

[0512] 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.

[0513] 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.

[0514] 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.

[0515] 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.

[0516] 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.

[0517] 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.

[0518] 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.

[0519] 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.

[0520] 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.

[0521] Third Implementation Method Figure 5 An example of the configuration of the data processing system 310 according to the third embodiment is shown.

[0522] 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.

[0523] 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).

[0524] 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.

[0525] 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.

[0526] 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).

[0527] 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.

[0528] 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.

[0529] 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.

[0530] 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.

[0531] 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.

[0532] 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".

[0533] 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.

[0534] 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.

[0535] 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.

[0536] 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.

[0537] 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.

[0538] 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.

[0539] 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.

[0540] 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 to analyze 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 to generate a menu using a generation AI. 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 to provide 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.

[0541] 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.

[0542] Fourth Implementation Method Figure 7 An example of the configuration of the data processing system 410 according to the fourth embodiment is shown.

[0543] 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.

[0544] 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).

[0545] 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.

[0546] 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.

[0547] 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).

[0548] 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.

[0549] 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.

[0550] 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.

[0551] 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.

[0552] 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.

[0553] 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.

[0554] 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".

[0555] 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.

[0556] 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.

[0557] 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.

[0558] 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.

[0559] 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.

[0560] 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.

[0561] 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.

[0562] 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.

[0563] 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.

[0564] 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...). 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.

[0565] 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.

[0566] 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.

[0567] 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).

[0568] 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.

[0569] 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."

[0570] 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.

[0571] 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).

[0572] 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.

[0573] 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.

[0574] 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.

[0575] 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.

[0576] 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.

[0577] 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.

[0578] 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.

[0579] 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.

[0580] 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.

[0581] 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.

[0582] In addition, the following notes are provided in response to the above explanation.

[0583] Example 1 (Note 1) An information processing system, characterized in that it comprises: An apparatus for receiving, via a communication device, purpose information and summary information about a business matter from a terminal by an information processing device, and for generating business matter identification information and determining the category of the business matter based on the purpose information and summary information; An apparatus for retrieving historical business item information and rule information from a storage device based on the business item identification information and the business item category, and obtaining regulatory information and market information from external information sources, and integrating the historical business item information, rule information, regulatory information and market information into business item related information; An apparatus for using natural language processing technology to segment and extract features from text data contained in the business-related information by an information processing device, in order to generate structured information representing legal requirements, information protection requirements, and customer response requirements. An apparatus for performing risk assessment processing and advantage assessment processing on the structured information by an information processing device to generate assessment information that classifies risk factors and advantage factors related to business matters. A device for automatically generating prompt statements for input into a generative artificial intelligence model by acquiring a stored prompt statement template corresponding to the business item category from an information processing device, and filling the evaluation information and the structured information into the prompt statement template; An apparatus for inputting the prompt statement into a generative artificial intelligence model by an information processing device and obtaining natural language text containing the draft content of business matters from the generative artificial intelligence model; An apparatus for associating the drafted content obtained by an information processing device with the business item identification information and storing it in a storage device, and for outputting the drafted content to a terminal through a display device; A device for displaying the draft content on a terminal, receiving modification and append information input by the user, and sending the modification and append information to the information processing device; And an apparatus for integrating the modified information and the supplementary information with the draft content by an information processing device, updating it into the final draft information, and storing the update record in association with the business item identification information.

[0584] (Note 2) According to the information processing system described in Appendix 1, the information processing device is configured to: when automatically generating the prompt statement, select a prompt statement template from multiple prompt statement templates based on the business item category and the structured information, and control the structure and output format of the prompt statement so that the legal requirements, information protection requirements and customer response requirements are recorded in separate sections.

[0585] (Note 3) According to the information processing system described in Appendix 1, the information processing device is configured to: generate additional prompt statements for instructing the generative artificial intelligence model to strengthen the description of elements and simplify the description of elements based on the weights of risk elements and favorable elements contained in the evaluation information, and input the additional prompt statements and the prompt statements together into the generative artificial intelligence model.

[0586] Application Example 1 (Note 1) An information processing system, characterized in that it comprises: This is a means of receiving natural language information containing project objectives, project summary, and constraints from a terminal via a communication interface, and structuring the natural language information into project analysis data. This is a processing method used to perform word segmentation, sentence segmentation, text classification, and entity extraction on text data in the project analysis data using natural language processing programs, in order to generate a structured set of elements containing legal factors, technical factors, market factors, and user emotional states. This is used to extract legal factors as legal concerns from the structured set of elements, extract technical factors and security-related expressions as security impacts, and extract market factors and customer-related expressions as customer reactions and market environment handling methods. A processing method for automatically generating prompt statements based on the structured set of elements, for inputting into a generative artificial intelligence model to generate a draft project proposal document for a new business project related to electronic transaction services, and inputting the prompt statements into the generative artificial intelligence model to generate the draft project proposal document; This processing method is used to determine the user's emotional state based on the natural language information and user feedback information using an emotion analysis program, and to generate prompt statements based on the emotional state to adjust the weight and priority of the legal concerns, the security impact and the customer reactions in the project proposal document, and to revise the draft project proposal document using the generative artificial intelligence model. A processing method for generating a final version of the project proposal document by constructing prompt statements based on user corrections and supplementary information received from the terminal and the draft project proposal document, and by using the generative artificial intelligence model. Processing means for sending the final version of the project proposal document to a terminal and storing the final version of the project proposal document in a storage device.

[0587] (Note 2) The information processing system according to Appendix 1 is characterized in that, The processing means is configured to use the natural language processing program and the generative artificial intelligence model to generate prompt statements for automatically generating structured project proposal documents for each new business project related to electronic transaction services. The structured project proposal documents include at least a plurality of chapters from the following: project overview, legal and compliance analysis, technical solutions and security design, market environment and customer response analysis, and risk control and project advantage maximization strategies.

[0588] (Note 3) The information processing system according to Appendix 1 is characterized in that, The processing method is configured to generate prompt statements based on the user's emotional state and structured elements related to the market environment, for prioritizing detailed explanations of at least a portion of the legal concerns, security impacts, and customer reactions in the project proposal document, and to automatically adjust the content density and recording order of the project proposal document according to priority using the generative artificial intelligence model.

[0589] Example 2 (Note 1) An information processing system, characterized in that it comprises: A means of acquiring information from external sources; A means for generating prompt statements for inputting the acquired case information into a generative artificial intelligence model; The generative artificial intelligence model is used to parse the text of the case information to calculate the parsing means of various risk indication information, including legal risk indication information, information security risk indication information, and customer reaction risk indication information. An evaluation method for obtaining benchmark information corresponding to relevant business areas from a storage device based on the risk indication information obtained by the analysis method, and evaluating the correspondence between the case information and the benchmark information; A means for generating report information in natural language based on the correspondence obtained by the evaluation method and the risk indication information, wherein the report information includes compliance countermeasure information, information security countermeasure information and customer response policy information; This is a classification method used to categorize the content of the report information according to department category and priority, and to construct the classification results as additional prompts input to the generative artificial intelligence model. Based on the classification results and the risk indication information, the system generates allocation information for distributing the report information to multiple processing entities and outputs the allocation method of the allocation information.

[0590] (Note 2) According to the information processing system described in Appendix 1, the evaluation method is configured to generate prompt statements for inputting multiple categories of information, including case information, legal benchmark information, safety benchmark information, and customer feedback information, into the generative artificial intelligence model, and to use the prompt statements to enable the generative artificial intelligence model to calculate the similarity between the case information and the multiple categories of information, thereby evaluating the correspondence.

[0591] (Note 3) According to the information processing system described in Appendix 1, the report information generation means is configured to generate a prompt statement for converting the risk indication information and the information representing the correspondence into at least one of a report format, an electronic document format, or a screen display format, and to generate multiple types of report information with different content and structure in different formats by inputting the prompt statement into the generative artificial intelligence model.

[0592] Application Example 2 (Note 1) An information processing system, characterized in that it comprises: A processor for performing information processing in an information processing device, the processor being configured to: A communication processing device receives transaction information, case information, or consultation information input by the user from the input device, and performs natural language processing on the received information to generate structured information. Based on the structured information and the regulatory information, security requirement information and historical response information stored in the storage device, a parsing device generates prompt statements for input to the generative artificial intelligence model, and uses the generative artificial intelligence model to parse the structured information to obtain analysis results regarding legal concerns, security impacts and user responses. Based on the analysis results, a prompt statement containing classification information of department type and processing type is generated, and a classification device is generated to map the information to the business processing department according to the classification information. An emotion estimation device that applies emotion analysis technology to voice, image, or text information from input devices or emotion recognition devices to estimate the user's emotional state; Based on the presumed emotional state and the analysis results, a priority control device generates prompt statements containing management information on the processing priority or response policy for the information, and modifies the classification indication information according to the processing priority. Based on the analysis results and the management information, improvement suggestion information including risk reduction schemes and value maximization schemes is generated, and an output device is generated for outputting the improvement suggestion information to the display device.

[0593] (Note 2) The information processing system according to Appendix 1 is characterized in that, The processor is configured to generate a prompt statement containing relevant business processing departments and descriptive summary information for the business processing departments to understand, based on the structured information and the parsing results, and input the prompt statement into the generative artificial intelligence model to generate classification indication information for automatically assigning the information to the business processing departments.

[0594] (Note 3) The information processing system according to Appendix 1 is characterized in that, The processor is configured to, based on the presumed emotional state, add indication information to the prompt statement to specify the response tone, level of detail, and intensity of risk description, and input the prompt statement containing the indication information to the generative artificial intelligence model, so that the generative artificial intelligence model generates explanatory information and improvement suggestions that are adapted to the user's emotional state.

Claims

1. An information processing system, characterized in that, include: processor; The processor is configured to: receive information through an interface for receiving information from a user; analyze the received information using a generative artificial intelligence model to evaluate legal concerns, security impacts, and customer reactions related to the information; generate prompts based on the analysis results to sort the information to the appropriate department; and analyze the user's emotional state using sentiment analysis technology to generate prompts based on the emotional state to set the priority of the information.

2. The information processing system according to claim 1, characterized in that, The processor is further configured to: utilize a generative artificial intelligence model to generate prompt information based on the information content for automatically determining the department related to the information.

3. The information processing system according to claim 1, characterized in that, The processor is further configured to generate prompts based on the user's emotional state for prioritizing the sorting of the information.

Images