Information processing system

Abstract

The application provides an information processing system. An information processing system, characterized by comprising: at least one processor; wherein the processor is configured to: provide an interface for receiving information from a user to the user; parse the received information by using a generative artificial intelligence model, generate information with higher relevance to the user based on a prompt word for indicating that the generative artificial intelligence model predicts user interest and emotion and generates high-relevance information; and communicate to notify the generated information to the user in real time.

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] Existing information delivery systems typically rely on simple keyword matching or pre-defined recommendation rules to push content to users during interactions. This makes it difficult to accurately and timely grasp users' true interests and emotional states in specific contexts, resulting in information that is not highly relevant or personalized. Users then need to spend considerable time and effort sifting through vast amounts of information to find useful content. Furthermore, existing systems often lack mechanisms for comprehensively utilizing user behavior history and emotional states, failing to fully leverage the capabilities of generative artificial intelligence models to achieve real-time information generation and feedback for individual users. Therefore, it is necessary to provide a system that can automatically receive user information, comprehensively analyze user interests and emotional states, generate highly relevant and personalized information through generative artificial intelligence models, and provide real-time feedback to users, thereby improving the accuracy, timeliness, and user experience of information services. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides an information processing system comprising at least one processor. The processor is configured to provide an interface to a user for receiving information from the user, through which it collects information in various forms, such as text, voice, and click operations input by the user. The processor is further configured to parse the received information using a generative artificial intelligence model, and based on prompts instructing the generative artificial intelligence model to predict user interests and emotions and generate highly relevant information, infer the user's current needs, thereby generating information highly relevant to the user. The processor is also configured to communicate to notify the user of the generated information in real time, thereby dynamically providing personalized content during user interaction.

[0005] Furthermore, the processor is configured to input the user's behavioral history and search history as prompts into the generative AI model, instructing it to generate personalized information that matches the user's long-term preferences and current query intent. This comprehensively utilizes the user's accumulated behavioral data to improve the accuracy and relevance of information generation. Further, the processor is configured to analyze the user's emotional state and provide the obtained analysis results as prompts to the generative AI model, instructing it to generate information content adapted to the user's current emotional state. This simultaneously optimizes recommendation results across both interest and emotional dimensions, achieving fine-grained personalized information delivery to the user.

[0006] "System" refers to an overall device or platform that includes at least one processor and its related hardware and / or software modules, used to perform functions such as information reception, parsing, generation and communication, and can be deployed on local devices, server clusters or cloud computing environments.

[0007] A processor is a hardware unit or logic unit that can execute program instructions to complete data processing and control flow. It can be a single central processing unit (CPU), graphics processing unit (GPU), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or any combination thereof.

[0008] An "interface" refers to a hardware or software channel provided by a processor for interacting with a user, including but not limited to graphical user interfaces (GUIs), command-line interfaces, application programming interfaces (APIs), web forms, dialog boxes, buttons, input boxes, etc., used to receive user input and / or output feedback information to the user.

[0009] "User information" refers to various types of data provided or triggered by the user through the interface, including but not limited to text input, voice input, images, click behavior, browsing path, selection operation, feedback results, etc., which are used to reflect the user's current needs, interests or status.

[0010] "Generative artificial intelligence models" refer to artificial intelligence models trained through machine learning or deep learning methods that can automatically generate text, speech, images, or other forms of content based on input data, including but not limited to large language models, text generation models, and multimodal generation models.

[0011] "Cue words" refer to text or structured input content used to provide task instructions and contextual information to generative artificial intelligence models. These include descriptions of user information, constraints on the model's generation goals, and requirements for output style and format, in order to guide the generative artificial intelligence model to output the desired results.

[0012] "Interest" refers to the degree of attention and preference a user shows toward a particular topic, category, product, or service within a specific time period, inferred by generative artificial intelligence models or related algorithms based on the user's input information, behavioral history, and context.

[0013] "Emotion" refers to characteristic information related to a user's psychological state, including but not limited to feelings of pleasure, sadness, tension, relaxation, satisfaction, and dissatisfaction. It is obtained by analyzing user input, tone, and behavioral patterns and is used to reflect the user's current emotional tendency.

[0014] "Highly relevant information" refers to text, recommendation results, or other information content generated by generative artificial intelligence models based on predictions of user interests and emotional states. This information is highly matched to user needs in terms of topic content and highly relevant to the user's current context in terms of timing.

[0015] "Behavioral history" refers to a collection of user-related interaction records within a certain time frame, including but not limited to browsing history, click history, dwell time, purchase history, feedback behavior, etc., which are used to characterize users' long-term or medium-term behavioral patterns and preference characteristics.

[0016] "Search history" refers to a collection of records of search requests initiated by a user in one or more applications or platforms, including search keywords, search time, search result selection, etc., which are used to reflect changes in the user's query intent and information needs over a period of time.

[0017] "Personalized information" refers to content or recommendation results generated specifically for a user by a generative artificial intelligence model based on the user's multi-dimensional characteristics such as interests, emotions, behavioral history, and search history, and matched with the user's needs and preferences. The personalized information corresponding to different users can be significantly different.

[0018] "Emotional state analysis" refers to the process by which a processor analyzes user input information and / or behavioral data to identify and infer the user's current emotional type and intensity. The analysis results can be used for further content generation or recommendation adjustments.

[0019] "Communication" refers to the process by which a system transmits data between a processor and a user terminal, or between different servers, through wired or wireless communication networks. This includes sending requests, receiving responses, pushing messages, and real-time notifications, and is used to transmit and provide feedback on the generated information.

[0020] "Real-time notification" refers to the behavior of automatically pushing new information to the user's terminal within a predetermined time period after the generative artificial intelligence model generates new information during user interaction, so that the user can receive timely updated feedback or recommended content in the current session or operation scenario. Attached Figure Description

[0021] Figure 1 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the first embodiment.

[0022] 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.

[0023] Figure 3 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the second embodiment.

[0024] 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.

[0025] Figure 5 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the third embodiment.

[0026] 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.

[0027] Figure 7 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the fourth embodiment.

[0028] 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.

[0029] Figure 9 This represents an emotion map that maps multiple emotions.

[0030] Figure 10 This represents an emotion map that maps multiple emotions.

[0031] Figure 11 This is a sequence diagram illustrating the processing flow of the data processing system of the first embodiment.

[0032] Figure 12 This is a sequence diagram illustrating the processing flow of the data processing system in Application Example 1.

[0033] Figure 13 This is a sequence diagram illustrating the processing flow of the data processing system of the second embodiment.

[0034] Figure 14 This is a sequence diagram illustrating the processing flow of the data processing system in Application Example 2. Detailed Implementation

[0035] Hereinafter, an example of an implementation of the system according to the present disclosure will be described with reference to the accompanying drawings.

[0036] First, let me explain the terminology used in the following instructions.

[0037] 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.

[0038] 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.

[0039] 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.

[0040] 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.

[0041] 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.

[0042] First Implementation Method Figure 1 An example of the configuration of the data processing system 10 according to the first embodiment is shown.

[0043] 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.

[0044] 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).

[0045] 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.

[0046] 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.

[0047] 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.

[0048] 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.

[0049] Figure 2 The diagram shows an example of the main functions of the data processing device 12 and the smart device 14.

[0050] 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.

[0051] 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).

[0052] 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.

[0053] 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.

[0054] 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."

[0055] In modern information processing environments, users generate a large amount of behavioral and retrieval history when accessing network resources, enterprise internal knowledge bases, and various application systems through terminals. Existing personalized information delivery technologies typically suffer from the following computer technology shortcomings: First, servers often directly input user behavior data into the search engine as search criteria, performing only simple weighting during the result ranking stage. This lack of high-level semantic modeling of user interests results in insufficient relevance between recommended results and the user's current context. Second, when calling generative artificial intelligence models, existing systems often use manually written fixed prompts, failing to dynamically generate prompts based on real-time user behavior, interest characteristics, and internal organizational information structures. This results in the underutilization of the reasoning capabilities of generative artificial intelligence models, making it difficult for generated content to adapt to the user's current browsing scenario in a timely manner. Third, the collaboration between servers and information management devices is limited to a pipeline-style "retrieve first, then display" process, lacking a mechanism to use the output results of generative artificial intelligence models to back-optimize search criteria and information integration strategies, leading to low internal information utilization and coarse-grained information integration. Fourth, user feedback on personalized information is usually only used for simple statistical analysis, failing to provide real-time feedback as a closed-loop signal to the interest modeling and prompt construction stages, making it difficult for the system to adaptively adjust the information generation process at the computer level.

[0056] Therefore, in the field of computer technology, there is an urgent need for a system solution that can perform structured processing of user behavior history and browsing information resources on the server side, transform them into prompts for generative artificial intelligence models, and then perform deep semantic integration with information held by the organization. At the same time, it can utilize user feedback to form a closed-loop optimization system solution to improve the processing efficiency and quality of information processing devices in personalized information generation, information retrieval control, and multi-source data integration, thereby substantially improving the computer-based information recommendation and search experience.

[0057] 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.

[0058] In this invention, the server includes means for acquiring usage information including user behavior history and retrieval history, and currently browsed information resources from a user-operated terminal; means for generating characteristic information representing user interests based on the usage information and the currently browsed information resources, and describing the characteristic information in natural language as prompt statements for use by a generative artificial intelligence model, thereby generating input instruction information; means for acquiring information held by an organization from an information management device and extracting the information held by the organization through retrieval processing; means for inputting the prompt statements and the information held by the organization into the generative artificial intelligence model, causing the generative artificial intelligence model to generate personalized information that integrates the currently browsed information resources with the information held by the organization based on user interests; means for sending the generated personalized information to the terminal and causing the terminal to associate the personalized information with the currently browsed information resources for display; and means for acquiring user evaluation information about the personalized information from the terminal and updating the generation conditions of the characteristic information and the prompt statements based on the evaluation information. This allows for the automatic generation of adaptive prompts for generative artificial intelligence models by extracting semantic features and structuring user behavior data and the current context on the server side. This drives the generative artificial intelligence model to work collaboratively with information management devices, achieving deep integration of organizational information and external browsing information. Furthermore, it utilizes user feedback to form a closed-loop optimization, thereby improving the processing performance and intelligence level of the computer system in personalized information generation, retrieval control, and data integration.

[0059] A "system" refers to a collection of information processing devices consisting of multiple functional units, used to perform information acquisition, analysis, generation, and provision processing on a computer.

[0060] "Terminal" refers to an electronic device operated by a user for browsing information resources, sending usage status information, and receiving and displaying personalized information, including but not limited to portable terminal devices, fixed terminal devices, or computing devices capable of running applications.

[0061] "User behavior history" refers to time-series data generated and recorded by the system during user operations, including page visits, link clicks, dwell time, scrolling behavior, and other behavioral events. It is used to reflect the user's past operating patterns and interests.

[0062] "Search history" refers to the collection of information such as search keywords entered by users on the terminal, submitted query requests, and access records corresponding to these queries, which is used to reflect users' needs and topics of interest in information retrieval.

[0063] "Usage status information" refers to a comprehensive data set that includes user behavior history and retrieval history, used to represent the user's usage status of various information resources on the terminal.

[0064] "Information resources being browsed" refers to the digital content and related data that a user accesses and displays through a terminal at a certain moment, including page address, title, body content, metadata, etc.

[0065] "Characteristic information" refers to structured or semi-structured information generated based on usage information and the information resources being browsed, used to represent attributes such as user interests, topics of interest, and their importance.

[0066] "Prompt statements" refer to text instructions generated by the server in natural language and provided to the generative artificial intelligence model to instruct the model to perform specific information generation, analysis, or integration tasks.

[0067] "Generative artificial intelligence models" refer to artificial intelligence models that are trained through machine learning and are able to generate new text content, summaries, recommendation results or other information outputs based on input text or other data, including but not limited to natural language generation models.

[0068] "Input instructions" refers to the set of input data, including prompts, used to control the processing method and output content of generative artificial intelligence models.

[0069] "Information management device" refers to a data management system or server used to store, manage, and provide information held by an organization, including information management systems implemented using databases, indexes, or file systems.

[0070] "Information held by an organization" refers to structured or unstructured data stored by an organization in its information management system, including internal information resources such as documents, records, reports, and knowledge entries.

[0071] "Retrieval processing" refers to the computational processing of querying, matching, sorting, and other operations performed on information held by an organization stored in an information management device in order to extract information related to given conditions.

[0072] "Personalized information" refers to comprehensive information or recommendation results that are customized for specific users, based on the user's characteristic information and the information resources the user is currently browsing, combined with information held by the organization.

[0073] "User feedback information" refers to the feedback data that users provide to personalized information through their terminals, including but not limited to usefulness evaluations, "not interested" tags, click behavior, and other feedback information that reflects the user's subjective judgment.

[0074] "Generation conditions" refer to the rules, parameters, or weight configurations used when generating characteristic information and prompts, including topic selection strategies, importance thresholds, and expression constraints, which are used to control the generation process and the form of output content.

[0075] The embodiments of this invention will describe the specific structure and technical effects of the system in conjunction with the hardware structure, software modules, data structure, and internal processing of the generative artificial intelligence model. In the following description, the subject is limited to server, terminal, or user.

[0076] I. Overall System Composition A server includes a processor, memory, network interface, and non-volatile storage. The processor can be a multi-core central processing unit, the memory is dynamic random access memory, and the non-volatile storage device can be a solid-state drive (SSD). The server runs application server software, database management system software, vector retrieval engine, and generative artificial intelligence model inference framework on an operating system (such as a general-purpose server operating system). The server can use a relational database management system (such as general-purpose relational database software) to store user event data, user profile data, and internal document data.

[0077] A terminal includes a computing device operated by a user, such as a smartphone, tablet, or desktop computer. The terminal runs a browser or native application to render web pages, collect user behavior data, and display personalized information returned by the server. The terminal can execute front-end scripts (such as event tracking code implemented in a scripting language) to collect user behavior data.

[0078] Users access network resources and internal organizational information resources through terminals, browsing, searching, and clicking. Users do not need to understand the internal server processing flow to obtain personalized information through natural interaction.

[0079] II. Server-side functional modules and data structures The server comprises multiple logical modules, which can be implemented in memory by programs executed by the processor: 1. Usage Status Information Acquisition Module The server receives usage information from the terminal. This information includes user behavior history and search history. User behavior history includes the Uniform Resource Locator (URL) of the visited pages, page titles, dwell time, scroll depth, and clicked targets. Search history includes search keywords, search initiation time, and click history of search results.

[0080] The server stores usage information in an event table in the database. This event table may have fields such as: user identifier, session identifier, event type, page URI (Uniform Resource Locator), title text, search keywords, content summary text, and timestamp. The server improves query performance through index structures (such as implementations based on B+ trees or inverted indexes).

[0081] 2. Current browsing information resource acquisition module The server receives or requests information resources that a user is browsing from a terminal. Based on the Page Uniform Resource Locator (URL) sent by the terminal, the server can retrieve page content from the web content server and extract elements such as the title, body, and metadata. The server can use a text parsing library to convert hypertext content into plain text and extract predefined areas, such as the main content area.

[0082] The server temporarily stores the information resources being browsed in the form of structured data in memory or cache for subsequent interest modeling and prompt generation.

[0083] 3. Characteristic Information Generation Module The server generates characteristic information based on usage information and the information resources being browsed. The server uses natural language processing software components (such as word segmenters, part-of-speech taggers, and sentence vector encoders) to perform word segmentation, stop word filtering, and vectorization on the title text, search keywords, content summary text, and current page summary in the user's behavior history.

[0084] The server can encode each text sample into a high-dimensional vector using a pre-trained language representation model (e.g., a Transformer-based encoding model). The Transformer structure includes multi-head self-attention layers, feedforward layers, layer normalization, and residual connections. During inference, the server converts the text sequence into word embeddings, inputs them to multiple stacked attention layers, and outputs sentence-level vector representations. The server can then cluster these vectors (e.g., using the K-means algorithm) to identify multiple topics, each corresponding to a set of highly similar text samples.

[0085] The server calculates topic importance based on the number of samples in each cluster, recent access frequency, and sample time decay factor. The server can assign weights to different time periods, for example, giving higher weight to recent behavior. The server then generates feature information, including several topic tags and their weight values. The server saves this feature information in a structured format to a user profile table, providing high-level semantic features for generating subsequent prompts.

[0086] 4. Prompt Statement Generation Module The server generates prompts based on feature information and currently viewed information resources for use by generative AI models. The server converts topic tags and weights into natural language descriptions so the model can understand the user's interest structure. For example, when feature information includes topics such as "smartphone reviews," "phone photography," and "battery life," the server generates text descriptions similar to the following: Users have recently been browsing content about smartphone reviews, mobile phone photography, and battery life, with the most attention paid to mobile phone photography, especially night scene imaging and portrait mode.

[0087] The server also converts the current page's title and summary into natural language context, for example: The page the user is currently viewing is titled "Latest Smartphone Reviews in 2026," and the content mainly discusses the photography performance and night scene imaging quality of several flagship phones.

[0088] The server combines the above description with the role settings and task descriptions of the generative artificial intelligence model to form a complete prompt statement. For example, the server generates the following prompt statement: You are a generative artificial intelligence model that generates personalized information recommendations for users.

[0089] The following is a description of the user's interests and information about the page they are currently browsing. Please generate an internal prompt based on this information for further information retrieval and content integration.

[0090] User Interest Description Users have recently been frequently browsing content related to smartphone reviews, mobile phone photography, and battery life, with mobile phone photography being the most of their focus.

[0091] Current page information Title: Latest Smartphone Reviews for 2026 Abstract: This article compares the photography and battery life of several flagship mobile phones, focusing on the analysis of night scene imaging and portrait mode effects.

[0092] Please output an internal prompt statement in Simplified Chinese: From which aspects (e.g., camera performance, battery life, price, brand reputation) should external network information and internal company document information be integrated for this user?

[0093] The server thus generates prompts to control generative AI models, enabling dynamic and adaptive model input construction rather than fixed templates.

[0094] 5. Information Acquisition Module of Information Management Device The server retrieves information held by the organization from an information management device. This information management device may include a relational database system, a full-text search engine, and a document storage system. The server constructs search criteria based on keywords extracted from the prompt and the search intent, and sends a query request to the information management device.

[0095] The server can generate search criteria with different field weights based on topic tags and weights, such as increasing the weight of content related to terms like "mobile phone photography" and "night scene imaging." Results obtained by the server from the information management device include internal evaluation reports, product specifications, and user feedback records. The server performs summary truncation and length control on the results for subsequent input into the generative artificial intelligence model.

[0096] 6. Generative Artificial Intelligence Model Invocation and Integration Module The server invokes a generative AI model, inputting prompts, feature information, and organizational information. This generative AI model can be a Transformer-based model, whose decoder generates output text word-by-word through an autoregressive approach. Internally, the model incorporates multiple layers of self-attention and cross-attention mechanisms, utilizing a joint representation of the prompts and internal document summaries for semantic reasoning.

[0097] When calling the model, the server sets parameters such as temperature, maximum generation length, and penalty parameters to control generation stability and diversity. The server can use techniques such as gradient pre-pruning to stabilize the model learning process during training. During training, the server uses pre-constructed training corpora, defines a cross-entropy loss function, and employs optimization algorithms to iteratively update the model parameters. Through large-scale training, the server enables the model to learn the ability to generate highly relevant outputs by integrating user interests, current context, and internal knowledge.

[0098] The server sends the following types of prompts to the generative AI model to generate personalized information for the final user. For example: You are a content generation assistant (generative artificial intelligence model) for end users.

[0099] Below is user interest information, a summary of the current page, and internal company reviews of several mobile phones. Please generate a comprehensive recommendation and three recommended models for the user.

[0100] User Interests - Mobile phone photography (pays close attention, especially to night scenes and portraits) - Smartphone Reviews Current page summary ... Internal evaluation summary 1. Phone A: Excellent night scene photography performance, rich detail in portrait mode... 2. Phone B: Balanced overall performance, good battery life, stable camera performance... 3. Phone C: Primarily features a slim and lightweight design; its camera performance is above average for its price range... Require: 1. Use Simplified Chinese.

[0101] 2. First, summarize in about 200 words the key camera features that users should focus on when choosing a mobile phone.

[0102] 3. Then recommend 3 more models in a separate list format, with each list not exceeding 150 characters.

[0103] The server receives the text output by the generative artificial intelligence model, parses it into personalized information, including overall suggestions and structured recommendation items.

[0104] 7. Result Transmission and Display Control Module The server packages personalized information into response data and sends it to the terminal via the network interface. The response data includes a context page identifier, text suggestions, and a list of recommended items. The server can compress the data to reduce bandwidth consumption.

[0105] After receiving personalized information, the terminal creates an interface area in the browser, such as a sidebar or floating card. The terminal displays text suggestions in the summary area, shows the recommendation list as items, and associates it with the current page. The terminal can dynamically adjust the display position and style based on screen size and interaction design.

[0106] 8. User review acquisition and generation condition update module The terminal provides feedback controls such as "useful," "useless," and "not interested" next to personalized information. Users rate the information by touching or clicking. The terminal then sends the rating information to the server.

[0107] After receiving user feedback, the server writes it into a feedback table. Based on this feedback, the server updates feature information and the conditions for generating prompts. For example, if the server repeatedly detects that a user has marked a recommendation primarily focused on "battery life" as "not interested," it reduces the weight of that topic and decreases its emphasis in subsequent prompts.

[0108] Through this closed-loop adjustment mechanism, the server incorporates user feedback during the interest modeling and prompt generation stages, dynamically altering data feature weights and language description emphasis to improve the relevance and accuracy of the generative AI model's output. This update is not a simple manual rule adjustment, but rather an automated parameter update based on feedback statistics and optimization strategies, creating a technological difference from traditional static recommendation systems.

[0109] III. Internal Processing and Technical Effects of Generative Artificial Intelligence Models The server employs a multi-layered Transformer decoding structure to implement generative artificial intelligence model functionality. Each layer includes a multi-head self-attention sub-layer, a cross-attention sub-layer, and a feedforward sub-layer. The self-attention sub-layer models the contextual arrangement of words within the prompt statement, while the cross-attention sub-layer aligns the information summary held by the organization, thereby realizing the association between user interests, current page information, and internal documents in a high-dimensional vector space.

[0110] During model training, the server selects appropriate learning rate scheduling strategies and regularization methods, such as using a warm-up and decay learning rate curve and parameter regularization, to suppress overfitting. The server can use data augmentation techniques to reorganize corpora of different user behavior patterns to enhance the model's generalization ability to diverse combinations of interests.

[0111] By employing the aforementioned structure and training methods, the server achieves improvements in computer technology compared to traditional systems that rely solely on keyword matching and ranking, in the following aspects: The server compresses a large amount of sparse behavioral event data into low-dimensional interest vectors through vectorization and clustering. This process reduces the dimensionality of the retrieval space, improves the speed of feature retrieval and similarity calculation, and thus reduces the overall response latency.

[0112] By dynamically generating prompts, the server guides the generative AI model to generate only semantic regions highly relevant to the user's interests and the current page. This reduces irrelevant information output, improves the accuracy of the generated results, reduces the number of retries, and indirectly reduces network communication and computing resource consumption.

[0113] By injecting topic weights and internal document summaries into the prompt statements, the server enables generative AI models to integrate multi-source information in a single inference process, reducing the need for multiple rounds of retrieval and generation, thereby improving the overall computational efficiency of the data pathway.

[0114] The server employs a feedback-driven generation condition update mechanism, enabling the feature information weights and prompt statement construction strategies to adapt to changes in user behavior over time. This feedback-based iterative parameter update is not limited to the display layer but directly affects the feature extraction and prompt statement generation algorithms. It represents an adaptive optimization of the computation process itself, helping to reduce error accumulation over long periods.

[0115] IV. Technical Effects of Terminal-Side Behavior Collection and Display When collecting user behavior data, the terminal employs a batch reporting mechanism and a local caching mechanism. By maintaining an event queue locally, the terminal merges and sends multiple behavior events, reducing the number of network requests and thus lowering communication load and server concurrency pressure. When the network is unstable, the terminal temporarily stores events locally to avoid data loss and improve the integrity of behavior data. Based on the complete behavior data, the server generates more accurate characteristic information, further improving the personalized generation effect.

[0116] When displaying personalized information, the terminal presents recommended content side-by-side with the currently viewed page, allowing users to compare and read without navigating to a new page. This display method helps the server clearly define contextual relationships within the interface structure, enabling the server to leverage the fixed semantic anchor of "current page" in prompts to enhance the generative AI model's understanding of the scene.

[0117] V. Alternative Implementation Methods and Extensions The server can implement the same processing flow under different hardware configurations and model sizes. For example, the server can deploy a generative AI model with a smaller number of parameters in a local environment to reduce dependence on external model services; or it can use a larger-scale cloud model to improve the quality of generation. In different implementations, the server achieves the technical effect by generating prompts and integrating characteristic and organizational information.

[0118] The server can also store interest vectors in a vector database, quickly find historical patterns most similar to the current user's interests using an approximate nearest neighbor retrieval algorithm, and reuse the corresponding prompt fragments, thereby further shortening the prompt generation time. This rapid prompt construction method based on vector retrieval and template concatenation demonstrates a technical improvement in computing resource allocation and latency control.

[0119] The server can also employ a multi-model collaborative approach, where a lightweight classification model is first used to predict the main topic currently of interest to the user, and then a generative AI model generates more granular prompts based on the main topic. Through this hierarchical processing, the server breaks down complex tasks into multiple sub-tasks, reducing the number of large model calls and improving overall throughput.

[0120] Through the various implementation forms described above, the server, terminal, and user form a collaborative working system. The system is not merely an automation of human recommendation behavior, but rather achieves structural optimization of the internal data processing of the computer through refined feature modeling, dynamic prompt generation, deep semantic integration, and feedback-driven parameter updates. This results in measurable technical effects in terms of processing speed, output accuracy, and resource utilization efficiency.

[0121] use Figure 11 The processing procedure is explained.

[0122] Step 1: The terminal collects usage information from user operations and sends it to the server.

[0123] The terminal uses user page visits, search inputs, and clicks in the browser or application as input data, including page URLs, page titles, search keywords, click targets, and timestamps. When a page loads, the terminal executes a front-end script, listens for page load events, search box submission events, and link click events, and encapsulates each event into a structured record containing a user identifier, session identifier, and event type, caching it in a local event queue. When the number of events reaches a preset threshold or a time interval is reached, the terminal batch packages multiple records in the queue, generates a request message containing multiple event objects, and sends this request to the server's tracking interface via the network interface. The output is an encoded usage status information request transmitted over the network.

[0124] Step 2: The server receives and stores usage information, forming a behavior history dataset.

[0125] The server receives a request for usage information from a terminal at the network interface, containing multiple event records. At the parsing layer, the server decodes the message body, verifies field integrity and format validity, and discards or marks invalid records as abnormal. The server calls the data access layer to map valid events to fields in the event table, performs batch insert operations, and creates indexes for user and session identifiers. The server merges repeated accesses by the same user within a short period based on timestamps to reduce redundancy. After data cleaning and storage, the output is a dataset of user behavior history and retrieval history stored in the database.

[0126] Step 3: The server generates characteristic information based on behavioral history and the information resources being browsed.

[0127] The server reads the recent behavior history of a specified user from the database as input, and simultaneously receives or retrieves the title and content summary of the page the user is currently browsing. The server performs Chinese word segmentation and stop word filtering on the title, search keywords, and content summary, converting the text into a word sequence. The server uses a pre-trained text encoding model to map each text into a fixed-length vector and performs a clustering algorithm on all vectors to identify multiple topic clusters. For each topic cluster, the server calculates the number of samples, average time weight, and access frequency, and calculates the topic weight according to a predetermined formula. The server combines topic tags and weights into a feature information structure, marking the user's interest intensity for each topic. The output is a feature information data structure representing the user's interest topics and their weights.

[0128] Step 4: The server converts the feature information and the current page information into prompts that can be used by the generative artificial intelligence model.

[0129] The server takes characteristic information data structures and the title and summary of the current page as input. It reads the highest-weighted topics from the characteristic information and converts them into natural language descriptions, such as stating that the user "is very concerned about mobile phone photography and night scene imaging." The server also organizes the current page's title and summary into scene description text. Based on this, the server combines role settings, task descriptions, and constraints to generate a complete prompt statement to guide the generative AI model in performing a specific task. The server converts structured data into continuous natural language text through string concatenation and template filling, forming input instructions for the model. The output is the prompt text for the generative AI model.

[0130] Step 5: The server extracts the search intent based on the prompts and retrieves the information held by the organization from the information management device.

[0131] The server takes the prompts generated in step 4 as input and uses a key phrase extraction algorithm or a lightweight language understanding model to identify the core search topics and constraints, such as "mobile phone photography," "night scene imaging," and "evaluation report." The server maps these key phrases to search criteria, constructing database query statements or full-text search requests. The server sends search requests to the information management device through a data interface to retrieve internal document records matching the search criteria. The server sorts the returned results by relevance, filters by time, and truncates the summaries, organizing the titles and summaries of multiple documents into a set of text fragments suitable for re-inputting into a generative artificial intelligence model. The output is a filtered and organized set of information summaries held by the organization.

[0132] Step 6: The server inputs the prompts along with information held by the organization into a generative artificial intelligence model to generate personalized information.

[0133] The server takes the prompt from step 4 and the internal information summary from step 5 as input, using the prompt as the main instruction and the internal information summary as contextual material, and organizes them together into the model input text. The server calls the generative AI model inference interface, feeding the input text into the model's encoding / decoding structure. In the multi-head attention layer, it calculates the semantic relationship between the prompt and the internal information, generating an output sequence. The server sets the maximum output length and control parameters in the inference request and receives the natural language results returned by the model. The server performs basic format checks and necessary sensitive word filtering on the results, compiling a comprehensive suggestion text and recommended items for display on the terminal. The output is personalized information content customized for the current user and the current page.

[0134] Step 7: The server encapsulates personalized information and sends it to the terminal for display in association with the current page.

[0135] The server takes the personalized information content generated in step 6 as input, structures the comprehensive suggestion text and recommended items into response data, including context identifiers corresponding to the current page. The server constructs a response object at the application layer, sets appropriate field names and data types, and returns it to the requesting terminal via the network interface. The server can compress the response or add caching flags before sending to reduce network transmission load. The output is the personalized information response message sent to the terminal.

[0136] Step 8: The terminal receives personalized information and displays it on the current page.

[0137] The terminal takes the personalized information response returned by the server as input and parses the comprehensive suggestion text and recommended item list in the front-end script. The terminal creates a new display area in the current page's document object model, such as a sidebar container or floating information box, inserts the comprehensive suggestion into the text display component, presents the recommended items in list format, and adds jump links or expand buttons. The terminal adjusts the position and style of the display area according to the current page layout and screen size to achieve associated display with the information resource being viewed. The output is a visual personalized information presented on the user interface.

[0138] Step 9: Users interact and provide feedback on personalized information; the terminal collects the feedback and sends it back to the server.

[0139] Users input personalized information displayed on the terminal interface and express their evaluations by clicking "Helpful," "Useless," or "Not Interested" buttons, or by clicking recommended items. The terminal listens for these interaction events and encapsulates the evaluation type, associated personalized information identifier, and timestamp into an evaluation record. After collecting the evaluation events, the terminal packages these records into a feedback request and sends it to the server. The output is a feedback request message containing the user's evaluation content.

[0140] Step 10: The server receives user evaluation information and updates feature information and prompt statement generation conditions.

[0141] The server takes the feedback request sent by the terminal as input, parses out the evaluation type, corresponding topic, and recommended item identifier, and stores them in the feedback data table. Based on the statistical analysis of multiple evaluation results, the server adjusts the topic weights in the feature information; for example, it lowers the weight of topics frequently marked as "not interested" and increases the weight of topics frequently marked as "useful." Simultaneously, the server modifies the parameters of the prompt generation module, causing subsequent prompts to weaken low-weight topics and strengthen high-weight topics in natural language descriptions, and adjusts the weight settings in the information retrieval conditions held by the organization. The server thus completes the feedback-based parameter update and rule correction, outputting updated feature information and new prompt generation conditions, providing an improved input basis for the next round of personalized information generation.

[0142] 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".

[0143] In existing technologies, personalized recommendation systems based on user behavior data typically employ pre-trained recommendation models or simple rule engines to filter candidate items from a database. These systems suffer from the following technical problems: First, the server side often only provides structured search results directly to the terminal, lacking unified modeling and dynamic integration of the user's current context, historical behavioral characteristics, and product inventory and promotional status. This makes it difficult for recommendation results to reflect inventory changes and promotional strategy adjustments in a timely manner. Second, while the server can call generative artificial intelligence models to generate natural language descriptions, there is a lack of a technical solution for automatically constructing high-quality prompts within the server. This prevents the effective integration of user behavior information, product information, inventory information, and sales promotion information scattered across multiple data tables into a personalized recommendation system. The input of generative AI models is limited, thus restricting the effectiveness and stability of generative AI models in recommendation scenarios. Third, existing systems typically present the output of generative AI models directly to users as unstructured text. The server lacks a mechanism for parsing and structuring this output at the product level, making it impossible to reliably correlate the product recommendation reasons in the generated text with the backend product data. This results in recommendation explanations being difficult for machines to process and reuse, hindering subsequent automatic optimization and multi-terminal presentation. Fourth, in scenarios with limited communication and computing resources, existing solutions lack a centralized process design on the server side for behavior aggregation, prompt generation, and result parsing. This forces the terminal to handle more logical processing, increasing terminal implementation complexity and cross-platform maintenance costs.

[0144] Therefore, it is necessary to propose a new system and its server-side processing method, enabling the server to: centrally manage user behavior information and organizational-side product / inventory / promotion information; generate prompts adapted to a generative artificial intelligence model by integrating the above data; call the generative artificial intelligence model to obtain personalized recommendation text; parse this text into structured recommendation information corresponding one-to-one with product data on the server side; and then push it to the user's device in a low-load, real-time manner via a communication device. In this way, the efficiency and controllability of the computer system's utilization of generative artificial intelligence models in recommendation tasks are improved from the perspectives of computing architecture and data processing flow, thereby enhancing the computer technical performance of the personalized recommendation system in terms of accuracy, interpretability, and scalability.

[0145] 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.

[0146] In this invention, the server includes: an information collection and recording device for acquiring user information, including behavioral information and search information, from a user device and recording the user information; an information storage device for managing the accumulated user information and the product-related information, inventory-related information, and sales promotion-related information stored in the information storage device, and for retrieving the product-related information, inventory-related information, and sales promotion-related information based on the user information; a prompt statement generation and model invocation device for integrating the user information, product-related information, inventory-related information, and sales promotion-related information obtained through the retrieval, generating prompt statements for inputting into a generative artificial intelligence model, and instructing the generative artificial intelligence model to perform information generation processing; and a generation result parsing and communication device for parsing the generated information output by the generative artificial intelligence model, generating individual recommendation information and recommendation summary information corresponding to the product-related information and inventory-related information, and sending them to the user device. This enables efficient integration of user behavior data with organizational product / inventory / promotion data on the server side, automatic construction of prompts for generative AI models, structured parsing of the generated results, and precise correspondence with the product database. Thus, without increasing the terminal's computing burden, it improves the overall performance of the computer system in terms of accuracy, real-time performance, and interpretability of recommendation processing, thereby improving the efficiency of computer technology utilization for generative AI models in personalized recommendation tasks.

[0147] "User device" refers to an electronic device operated by a user to access information resources, send user information, and receive recommendation information, including but not limited to terminal equipment, computing devices, or communication devices.

[0148] "User information" refers to a collection of data related to a user and sent from the user's device to the server, including behavioral information, search information, and user characteristic information derived therefrom.

[0149] "Behavioral information" refers to the operation records generated by users on their devices, including behavioral data related to interactions with information resources such as page browsing, product clicks, adding items to the shopping cart, and placing orders.

[0150] "Search information" refers to the data related to the query conditions entered and submitted by the user on the user's device, including search keywords, filter conditions, and parameters related to the search behavior.

[0151] "Information storage device" refers to data storage resources used to store and manage user information, product information, inventory information and sales promotion information on the server side, including database systems or other non-volatile storage media.

[0152] "Product-related information" refers to the data set related to the recommended object, including object identification information, name, category, attributes, unit price, and description information.

[0153] "Inventory-related information" refers to inventory status data related to goods, including the available quantity, status, and update time of objects in various storage locations or specifications.

[0154] "Sales promotion related information" refers to data related to promotional strategies set up to promote product sales, including information on discount activities, rules for spending a certain amount to reduce the price, rules for applying coupons, and the effective time range of the activities.

[0155] "Retrieval" refers to the process by which a server performs query processing on product-related information, inventory-related information, and sales promotion-related information stored in an information storage device based on user information, in order to obtain a target data set that meets predetermined conditions.

[0156] "Generative artificial intelligence models" refer to artificial intelligence models that can automatically generate output results such as text based on input prompts, including models trained with machine learning algorithms and capable of generating natural language.

[0157] "Prompt statements" refer to natural language or natural language-like text generated by the server and input into the generative artificial intelligence model. They are used to instruct the generative artificial intelligence model to specify the generation goals, constraints, and contextual information.

[0158] "Generated information" refers to the output information of a generative artificial intelligence model after receiving prompts, including recommended text, explanatory text, or other text content related to the target task expressed in natural language.

[0159] "Individual recommendation information" refers to structured recommendation data generated separately for each recommended object after the generated information is matched with product-related information and inventory-related information, and includes recommendation reasons or attribute descriptions.

[0160] "Recommendation summary information" refers to text or structured data that summarizes and generalizes multiple individual recommendation information based on the generated information, and is used to provide users with an overview of the recommendation results.

[0161] "Information collection and recording device" refers to a functional unit or program processing module on the server side used to receive user information from user devices and store and record such user information.

[0162] "Data management and retrieval device" refers to a functional unit or program processing module on the server side used to organize and maintain the data stored in the information storage device, and to perform query processing to obtain target data based on user information or predetermined conditions.

[0163] "Prompt statement generation and model calling device" refers to a functional unit or program processing module on the server side that integrates user information, product information, inventory information and sales promotion information to construct prompt statements, and sends the prompt statements to the generative artificial intelligence model and requests the generation of information.

[0164] "Generation result parsing and communication device" refers to a functional unit or program processing module on the server side used to parse the generation information output by the generative artificial intelligence model, generate individual recommendation information and recommendation summary information, and send the information to the user device through a communication interface.

[0165] The embodiments of this invention will be described in detail with reference to the server hardware and software structure not shown in the accompanying drawings, illustrating the utilization methods of each functional module of the system and the generative artificial intelligence model. The following embodiments can be used individually or in any combination, as long as they do not contradict each other. In the following description, the subjects are limited to "server," "terminal," and "user."

[0166] In one implementation, a server comprises a computing device including: at least one central processing unit (CPU, such as a multi-core general-purpose processor), an optional graphics processing unit (GPU), main memory (RAM), non-volatile storage (such as a solid-state drive, SSD), and a network interface (supporting Ethernet, wireless LAN, etc.). At the software level, the server runs an operating system (such as a Linux-based server operating system), a database management system (such as a relational database management system, MySQL), an application server (such as an HTTP-based web server), and a backend business application runtime environment (such as a Python runtime environment or other scripting runtime environment). A terminal consists of a user-operated electronic device, such as a smartphone, tablet, or personal computer, running a web browser or native applications. Users interact with the server through the terminal, generating behavioral information and receiving recommendations.

[0167] In one implementation, the program executed by the server is a set of instructions stored in a non-volatile storage device. These instructions are read by the CPU and executed in main memory. The program logic is divided into an information acquisition and recording module, a data management and retrieval module, a prompt statement generation and model invocation module, and a result parsing and communication module. Each module can be physically executed by the same processor or by multiple processing nodes in a distributed manner.

[0168] In the information collection and recording module, the server receives and preprocesses user information from the terminal. When a user interacts with the system, the terminal calls the front-end code to encapsulate user actions such as page browsing, searching, and clicking into data objects with a fixed field structure. These include user identifier, event type, event parameters (e.g., search keywords, product identifiers), and timestamps. Upon receiving this data object, the server uses the CPU to perform format validation, type conversion, and time standardization on the fields, such as converting string timestamps to a unified UTC time format. The server then writes the preprocessed user information into a relational database system as relational data rows, creating an index structure for fields such as user identifier and timestamps. Through this specific data structure and index design, the server can complete historical behavior queries for a single user with logarithmic time complexity, significantly reducing query latency and improving processing speed in large-scale data environments.

[0169] In its data management and retrieval module, the server centrally manages user information, product information, inventory information, and sales promotion information stored in the database. The server sets up data tables for product information, including fields such as product identifier, category, brand, attribute vector, and price; for inventory information, it sets up data tables recording product identifier, specifications (e.g., size, color), and inventory quantity; and for sales promotion information, it sets up data tables recording activity type (discount, minimum spend, etc.), discount parameters, applicable product set, and start and end times. Through predefined foreign key relationships, the server efficiently links the product identifier field in user behavior records to product and inventory information, reducing multiple network accesses and table join overhead during subsequent processing. Because this association processing is centralized on the server side, the terminal only needs to perform lightweight data display, reducing the terminal's computational burden and lowering cross-platform adaptation costs.

[0170] In the prompt generation and model invocation module, the server calculates user features based on the user's recent behavioral information. The server can implement a specific feature extraction algorithm: it selects search and click events within a predetermined time window in chronological order, counts the frequency of search terms, normalizes them into a probability distribution, and selects the keywords with the highest weights as the "search interest vector"; it constructs category and brand count vectors based on the product categories and brands browsed by the user, and uses a weighted average method to obtain the user's "category preference vector" and "brand preference vector"; the server also calculates the median and percentile range of prices based on the distribution of recently viewed product prices, as the user's "price preference range." This feature data is stored in memory as key-value pairs in a feature object for subsequent text generation.

[0171] Within the same module, the server combines user characteristics with product-side data to construct prompts for the generative artificial intelligence model. Using string templates in the application, the server converts recent search keywords, preferred categories, preferred brands, price ranges, and filtered candidate product information into natural language text, then concatenates this text into a structured prompt. In one embodiment, the server can generate the following example prompt: You are an intelligent recommendation consultant for an e-commerce platform.

[0172] The user's recent behavioral characteristics are as follows: - Search keywords: "running shoes", "sports socks" - Preference categories: "Sneakers", "Sportswear" - Preferred brands: "Brand A", "Brand B" Price range: 300-700 yuan The following is information on candidate products (stock and promotional information have been filtered): 1. A lightweight running shoe from a certain brand, priced at 499 yuan, with ample stock, currently 10% off.

[0173] 2. A certain brand of cushioned running shoes, priced at 599 yuan, with moderate stock, participating in a promotion of 50 yuan off for purchases over 400 yuan.

[0174] 3. A certain brand of entry-level running shoes, priced at 299 yuan, with ample stock and no promotions.

[0175] Based on the information above, please recommend no more than 5 of the most suitable products for this user in Chinese: 1. Provide a recommendation reason of no more than 2 sentences for each recommended product (including suitable scenarios, price advantages, promotional information, etc.).

[0176] 2. Output a summary of the overall recommendations in 3-4 sentences, written in a friendly and natural tone.

[0177] The server uses the aforementioned prompts to issue explicit generation instructions to the generative AI model, embedding filtered and integrated key data within these prompts. This allows the model to make recommendations within a controlled scope during the generation phase. This server-centralized generation of prompts significantly reduces irrelevant information noise compared to simply using raw behavior logs as model input, improves the relevance and stability of the model's generated results, and reduces the input length required for model inference, thereby increasing overall inference throughput within a given hardware resource limit.

[0178] In terms of model invocation, one implementation uses a sequence-to-sequence neural network based on a self-attention mechanism as the generative artificial intelligence model. This model can employ a multi-layer Transformer encoder and decoder structure, where each layer includes a multi-head self-attention sublayer and a feedforward fully connected sublayer. During the training phase, the server pre-trains the model using a large-scale text corpus and then fine-tunes it on recommendation-related corpora. During fine-tuning, the server uses cross-entropy as the error function and updates the model weights using stochastic gradient descent or adaptive learning rate optimization algorithms. The server can also employ data augmentation strategies, such as randomly shuffling the order of candidate products in the training samples and paraphrasing descriptive text, to improve the model's robustness to different permutations and expressions in the prompts. This specific neural network structure and training method enable the model to effectively capture the high-dimensional non-linear relationships between user preferences, product features, and promotional rules when processing prompts constructed by the server, thus outperforming traditional rule engines in terms of text generation accuracy and diversity.

[0179] In the result parsing and communication module, the server performs programmatic analysis on the generated information output by the generative artificial intelligence model. In one implementation, the server can pre-constrain the output format of the prompt statements, for example, requiring the model to list the reasons for recommending each product by number and explicitly include the product name or product identifier in the text. After obtaining the generated information, the server uses a string matching algorithm or a lightweight natural language parser to extract the product name fragments appearing in each recommendation description and performs fuzzy or exact matching with the product name field in the product-related information table on the server side, thereby determining the specific product record corresponding to the recommendation text. The server binds the matched product identifier with the recommendation text segment, generating a structured data record containing the product identifier, price, promotional tags, and reasons for recommendation, and combines several records into an individual recommendation information list. At the same time, the server saves the overall summary paragraph generated by the model as recommendation summary information. Through this processing method of converting text output into structured recommendation data, the server can cache, evaluate, and reuse the generated results, such as for log analysis and subsequent model training, significantly improving the usability of data management.

[0180] Through the cooperation of the aforementioned modules, the server achieves unified integration and processing of user information and various product-related information at the central level, thereby obtaining high-quality prompts adapted to the generative artificial intelligence model and transforming the model output into machine-processable structured recommendation data. Because the server internally employs specialized feature extraction algorithms, index design, prompt templates, and result parsing methods, this invention not only automates the human decision-making process but also represents a comprehensive improvement in data processing paths, model invocation methods, and result structuring strategies. The technical effects of this improvement include: under the same hardware resources, the server can generate more accurate personalized recommendations in parallel for multiple user requests; since candidate products are filtered based on inventory and promotions before input, the generated results almost entirely exclude unsellable items, reducing erroneous recommendations; through structured parsing, the server can cache and incrementally update recommendation content, reducing the frequency of repeated calls to the generative artificial intelligence model, thereby lowering communication and computational load.

[0181] In another embodiment, the server can employ different database management systems and operating environments, as long as they can support the aforementioned data structures and retrieval logic. For example, the server can use a columnar database management system to optimize aggregate queries on large-scale behavioral logs, or it can use a key-value store system to cache high-frequency user features to reduce database access latency. The server can also employ neural networks of different scales in the implementation of the generative artificial intelligence model, for example, reducing the number of layers and parameter size for deployment on resource-constrained edge servers, or using large-scale models in data centers to improve generation quality. Regardless of the variant, as long as the core processing performed by the server includes: acquiring and recording user information from user devices, managing and retrieving product-related information, inventory-related information, and sales promotion-related information, integrating the above information into prompts for the generative artificial intelligence model, and generating structured results containing individual recommendation information and summary recommendation information based on the model output, it can be considered an embodiment of the present invention.

[0182] In this invention, the terminal primarily handles user interface presentation and lightweight logic processing. After receiving individual and summary recommendation information from the server, the terminal presents the reasons for each product's recommendation, along with its price, promotional tags, and images, to the user through a front-end program. In one embodiment, the terminal can dynamically adjust the number of items displayed or its caching strategy based on network conditions, without needing to execute complex data integration and recommendation logic. This reduces the burden on the terminal's processor and memory, enabling the system of this invention to run smoothly on various terminals with lower hardware configurations.

[0183] In this invention, users generate behavioral information through natural interactions on the terminal. This behavioral information is converted into high-dimensional features by the server and participates in subsequent recommendation generation. Because the server employs a unified algorithm flow and model invocation strategy, users can obtain interpretable recommendation results without needing to understand the internal workings. Through the above technical solution, the server embeds a generative artificial intelligence model into a framework of structured data processing and specific algorithm control, achieving the constraint and utilization of model behavior, thereby technically achieving comprehensive optimization of recommendation accuracy, system throughput, and communication load.

[0184] use Figure 12 The processing procedure is explained.

[0185] Step 1: User actions are generated on the terminal and the terminal packages the action data. Users interact with the application on the terminal by browsing pages, entering search terms, clicking on products, adding items to the shopping cart, and so on. The terminal listens for these operation events through front-end scripts or application logic, and converts each operation into a behavior record.

[0186] The terminal's input consists of real-time user actions (such as clicking on a product, entering "running shoes" in the search box and submitting). Based on this input, the terminal generates a data structure containing fields such as user identifier, event type (browse, search, click, etc.), event parameters (search keywords, product identifier, current page URL, etc.), and timestamp.

[0187] The terminal performs basic validation and completion on these fields (such as generating a local timestamp and supplementing the session identifier), and caches multiple behavior records in memory. When a certain number or time interval is reached, the records are packaged into request data through a secure communication protocol.

[0188] The output of this step is a set of structured behavioral data (e.g., multiple JSON records), ready to be sent to the server.

[0189] Step 2: The terminal sends user information to the server and receives confirmation. After collecting a batch of behavioral data, the terminal initiates a network request to the interface provided by the server, sending the behavioral data as the request body.

[0190] The terminal's input is the set of behavioral data packaged in step 1. The terminal serializes and compresses the data using the network communication stack (e.g., serializing objects into text or binary format, with optional compression), and then sends it to the URL specified by the server via HTTP / HTTPS protocol.

[0191] After sending, the terminal waits for a response from the server. It determines whether the transmission was successful based on the status code and response content returned by the server. If it fails, it marks the message in the local queue and will retry it later.

[0192] The output of this step is the server's response information, including success confirmation or error messages, which the terminal uses to update its local cache status.

[0193] Step 3: The server receives user information, parses it, and stores it. The server receives requests from the terminal through the network interface and then hands the requests over to the backend application for processing.

[0194] The server's input is a behavioral data request sent by the terminal, which contains multiple user information records. The server first parses the request header and request body, and deserializes the transmission format (such as JSON) to obtain an operable memory object.

[0195] The server then performs data processing on each record: checking whether necessary fields exist and whether field types meet expectations, converting timestamps to a standard time format, standardizing user identifiers (e.g., converting them to an internal unified encoding), and filtering or logging abnormal records.

[0196] The server writes qualified records into the information storage device, inserts them into the user behavior table in the relational database, and optimizes subsequent queries using the database index structure (such as an index with user ID and timestamp as the key).

[0197] The output of this step is a collection of user behavior records already stored in the database, along with the corresponding database update status.

[0198] Step 4: The server aggregates user historical behavior and generates user characteristics. When a recommendation is needed, the server reads the target user's historical behavior records from the information storage device and aggregates and extracts their features.

[0199] The server's input consists of behavioral records (including search events, browsing events, click events, etc.) from a database corresponding to a specific user identifier over a given period of time. The server sorts these records by time and uses statistical calculations to count search keywords, selecting the most frequent search terms as the user's primary points of interest.

[0200] The server associates clicked or viewed product identifiers with product-related information tables, counts the frequency of occurrence for each product category and brand, and uses normalization operations to generate category preference vectors and brand preference vectors. The server also calculates the median and quantiles from the price field of the corresponding product to form a price preference range.

[0201] The server encapsulates the above statistical results into a user feature object, including a list of recently searched keywords, a list of preferred categories, a list of preferred brands, and a price range.

[0202] The output of this step is structured user characteristic data, which is used for subsequent prompt message generation.

[0203] Step 5: The server retrieves product-related information, inventory-related information, and sales promotion-related information. The server retrieves and filters product-related information, inventory-related information, and sales promotion-related information from the information storage device based on user characteristics.

[0204] The server's input consists of the user characteristic data generated in step 4, as well as product, inventory, and promotion data tables. First, the server filters the product information tables to select a set of products that meet the user's preferred categories and brands. Then, based on the user's price preference range, it further filters products that are not within that price range.

[0205] The server then uses the product identifier to query the corresponding inventory record in the inventory-related information table, calculates the total inventory of each product for each specification, and filters out products with insufficient or no inventory. Next, the server searches the sales promotion-related information table for ongoing promotional activities based on the current time and product identifier, and attaches corresponding promotional information to each candidate product.

[0206] The server merges the filtered product set with the corresponding inventory status and promotional information into a candidate product list. Each element includes product identifier, name, category, brand, price, inventory quantity, and promotional tags.

[0207] The output of this step is a list of candidate products that match user preferences and system constraints.

[0208] Step 6: The server generates prompts based on user characteristics and candidate items. In the prompt generation and model invocation module, the server converts user characteristics and candidate product lists into natural language prompts that can be understood by the generative artificial intelligence model.

[0209] The server's input consists of the user characteristic data from step 4 and the candidate product list from step 5. The server formats information such as the search keyword list, preference categories, preferred brands, and price ranges into descriptive sentences through string concatenation and template filling. Simultaneously, it organizes the name, brand, price, inventory, and promotional information of each candidate product into a numbered list.

[0210] The server embeds these text fragments into a pre-defined prompt template to construct a complete prompt. For example, the server might generate the following text: You are an intelligent recommendation consultant for an e-commerce platform.

[0211] The user's recent behavioral characteristics are as follows: - Search keywords: "running shoes", "sports socks" - Preference categories: "Sneakers", "Sportswear" - Preferred brands: "Brand A", "Brand B" Price range: 300-700 yuan The following is information on candidate products (stock and promotional information have been filtered): 1. A lightweight running shoe from a certain brand, priced at 499 yuan, with ample stock, currently 10% off.

[0212] 2. A certain brand of cushioned running shoes, priced at 599 yuan, with moderate stock, participating in a promotion of 50 yuan off for purchases over 400 yuan.

[0213] 3. A certain brand of entry-level running shoes, priced at 299 yuan, with ample stock and no promotions.

[0214] Based on the information above, please recommend no more than 5 of the most suitable products for this user in Chinese: 1. Provide a recommendation reason of no more than 2 sentences for each recommended product (including suitable scenarios, price advantages, promotional information, etc.).

[0215] 2. Output a summary of the overall recommendations in 3-4 sentences, written in a friendly and natural tone.

[0216] During the generation process, the server controls the text length and escapes special characters to meet the requirements of the generative artificial intelligence model interface.

[0217] The output of this step is a complete prompt text, which is used as input for a generative artificial intelligence model.

[0218] Step 7: The server invokes the generative artificial intelligence model and obtains the generated information. The server uses the model invocation module to send the prompts generated in step 6 to the generative artificial intelligence model and receive the recommended text generated by the model.

[0219] The server's input consists of the prompt text and model call parameters (such as maximum generation length, temperature, decoding strategy, etc.). The server calls a generative artificial intelligence model service deployed remotely or locally via a network interface. Internally, the model encodes the prompt text based on a pre-trained neural network (such as a multi-layer Transformer structure), maps each word to a vector space, calculates contextual relevance using a multi-head self-attention mechanism, and generates the output text step by step according to the probability distribution during the decoding stage.

[0220] During the call process, the server sends a request and waits for a response. Once the model output is received, the server performs an integrity check on the returned text, such as checking whether it contains the required numbers and keywords. If it does not conform to the agreed format, the server can choose to call it again or make corrections.

[0221] The output of this step is generated information produced by a generative artificial intelligence model, which is natural language text containing multiple reasons for product recommendations and summary recommendations.

[0222] Step 8: The server parses the generated information and constructs structured recommendation results. In the result parsing and communication module, the server parses the generated information and extracts the recommendation reasons and overall recommendation summary for each product.

[0223] The server's input consists of the generated information text obtained in step 7 and the candidate product list from step 5. The server first segments the generated text according to a desired format, such as by line or by number, and attempts to match each recommendation statement with a candidate product.

[0224] The server can use product names or product identifier keywords as anchors to find segments in the recommended text that match the candidate product names through string search or fuzzy matching algorithms (such as edit distance-based or word segmentation matching), thereby determining the product corresponding to that segment of text. The server associates the successfully matched recommended text with the corresponding product identifier, forming a structured record containing the product identifier, name, price, promotional information, and reasons for recommendation.

[0225] The server also extracts a summary paragraph from the generated information and saves it as a recommended summary.

[0226] The main data processing performed by the server in this process includes text segmentation, key phrase extraction, string matching, and index query operations, thereby transforming unstructured text into storable and searchable structured recommendation data.

[0227] The output of this step is a list of individual recommendations and a summary of recommendations, both of which are structured and linked to product-related information.

[0228] Step 9: The server sends the recommendation results to the terminal, which then presents them. After preparing the structured recommendation results, the server sends individual recommendation information and summary recommendation information to the terminal through the communication module.

[0229] The server takes as input the structured recommendation data generated in step 8 and the target user identifier. The server encapsulates this data into a response message, transmits it to the corresponding terminal via network protocol, and includes necessary status information (such as request identifier, generation time, etc.) in the message so that the terminal can display and cache it correctly.

[0230] After receiving the server's response, the terminal parses the recommendation results, combining the recommendation reasons, price, promotional tags, and product images obtained locally or remotely with other resources to render a recommendation list interface. Simultaneously, the terminal displays a summary of recommendations provided by the server on the interface.

[0231] During the display process, the terminal can record the user's further actions on the recommendation results (such as clicking on a recommended product or ignoring a recommendation), and then send the data back to the server, forming a closed loop.

[0232] The output of this step is a personalized recommendation result presented on the terminal interface, as well as new user behavior data input that can be used for subsequent iterations.

[0233] 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.

[0234] 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."

[0235] In existing information provision technologies based on generative artificial intelligence (AI) models, users typically input only a one-time natural language request as a prompt to the model, which then generates recommendations or answers. This approach has several problems: First, the server's utilization of user behavior data (e.g., search behavior, browsing behavior, dwell time) is coarse-grained, often involving only simple statistics or scoring, failing to provide a structured representation of long-term user behavior characteristics at the prompt level. This limits the depth of the generative AI model's understanding of user interest patterns. Second, when constructing prompts, the server lacks a fine-grained correlation between product or service data and user behavior data, failing to accurately define the candidate object set and its attribute range within the prompts. This results in an overly broad output range for the generative AI model, leading to unfocused recommendations and excessive noise. Third, in traditional systems, subsequent user interactions with the system's output (e.g., clicking on a recommended object in the generated results, ignoring a certain type of prompt) are typically only logged and not fed back into the prompt construction and data integration updates in real time. This makes it difficult to adjust the generation logic and output content promptly, resulting in insufficient personalization and slow convergence.

[0236] From a computer technology perspective, the aforementioned problems manifest as follows: the server-side lacks a closed-loop mechanism in its data processing flow—a mechanism for "behavioral data—structured integration—prompt generation—output feedback and re-integration"—for generative AI models. This results in a large amount of behavioral data and product / service data stored in the storage device not being uniformly modeled and accessed in a form suitable for efficient reasoning by generative AI models, leading to low utilization efficiency of server processing and communication resources. Furthermore, due to the lack of a systematic contextual organization in the construction of prompts, the context window of the generative AI model is inefficiently occupied, affecting the quality and real-time performance of generated useful information.

[0237] Therefore, it is necessary to provide a new system and method that unifies and integrates user behavior information and product / service information on the server side to generate prompts that highly match the generative artificial intelligence model. The subsequent interactive behaviors of the model output are then incorporated into the integration process, thereby improving the server's data processing flow and prompt construction method, enhancing the effectiveness and efficiency of the generative artificial intelligence model in personalized information generation scenarios, and ultimately improving computer information processing technology.

[0238] 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.

[0239] In this invention, the server includes means for receiving user behavior information from a program running on an information terminal operated by a user and storing the behavior information in a storage device; means for associating the behavior information stored in the storage device with attribute information about an item or service and provision condition information related to the item or service based on identifiers and time information to generate integrated data; means for constructing prompt statements for inputting into a generative artificial intelligence model based on context information describing the integrated data and output conditions corresponding to the integrated data, and inputting the prompt statements and the integrated data into the generative artificial intelligence model so that the generative artificial intelligence model generates useful information personalized for each user; means for converting the useful information output from the generative artificial intelligence model into display information that can be displayed or notified in the information terminal and sending the display information to the information terminal via a communication network; and means for updating the integrated data and the content of the prompt statements by obtaining the behavior information again through an append operation corresponding to the useful information prompted in the information terminal. This allows for a closed-loop processing flow within the server, encompassing fine-grained behavioral data collection, integrated modeling of behavioral and item / service information, automatic construction of prompts for generative AI models, and dynamic updates of integrated data and prompts based on user-added operations. This enables generative AI models to generate more focused and personalized useful information under a controlled set of candidate objects and their attribute constraints. Consequently, it improves the utilization efficiency of behavioral and item / service data in storage and processing devices, optimizes the server's invocation of generative AI models and context organization, and ultimately enhances the processing performance and resource utilization efficiency of the computer system in personalized information recommendation and notification tasks.

[0240] A "system" refers to an overall technical solution consisting of one or more information processing devices, information terminals, and communication networks, used to collect, store, process user-related data, and provide information to users.

[0241] "Information processing device" refers to an electronic computing device with a processor, storage device and communication interface, used to execute programs to collect, store, analyze and transmit various types of data, including but not limited to servers or computing nodes.

[0242] "Information terminal" refers to a device operated by a user and connected to an information processing device via a communication network for displaying information and receiving user input, including but not limited to computer terminals, mobile terminals, or other devices with human-computer interaction functions.

[0243] "User" refers to an entity that interacts with the system through an information terminal, whose behavioral information is collected and used to generate personalized and useful information. It can be a natural person or its agent.

[0244] "Behavioral information" refers to data obtained based on a user's actions on an information terminal, which reflects the user's interests or usage patterns, including but not limited to search operations, browsing operations, click operations, dwell time, and related records of content interaction.

[0245] "Storage device" refers to hardware resources used to retain data in a readable and writable manner, including but not limited to semiconductor memory, magnetic storage media, optical storage media, and storage systems formed by combinations thereof.

[0246] "Items" refers to any tangible or intangible object that can be used or consumed by users through trade or provision, including but not limited to goods, digital content, and other recommendable objects.

[0247] "Services" refers to various intangible offerings that can be provided to users, which consist of behaviors or functions, including but not limited to online services, offline services, or value-added services associated with goods.

[0248] "Attribute information" refers to structured or semi-structured data used to describe the characteristics of an item or service, including but not limited to category information, functional information, price information, inventory information, scope of application information, and label information.

[0249] "Conditions information" refers to constraints or preferential data related to the manner in which goods or services are provided, including but not limited to price conditions, discount conditions, combination conditions, time restrictions, and regional restrictions.

[0250] "Identifier" refers to a numerical or string value used to uniquely or distinguish an object in a system, including but not limited to user identifiers, session identifiers, item identifiers, service identifiers, and data record identifiers.

[0251] "Time information" refers to time data used to characterize the time of occurrence or effective time range of an event or data record, including but not limited to timestamps, dates, time intervals, and markers related to time sequence.

[0252] "Integrated data" refers to a comprehensive set of data used to describe the relationship between a specific user and a specific item or service. This data is obtained by associating, classifying, and organizing behavioral, attribute, and conditional information based on identifiers and time information in an information processing device.

[0253] "Contextual information" refers to supplementary information used to explain the meaning of the integrated data and the background of its generation, including but not limited to user preference information, behavioral summaries, descriptions of the candidate object range, and environmental constraint information.

[0254] "Output conditions" refers to a set of parameters that control or limit the content output by a generative artificial intelligence model, including but not limited to output length, output format, language style, object range, and granularity requirements.

[0255] "Generative artificial intelligence model" refers to an artificial intelligence model that is trained on large-scale data based on machine learning and deep learning technologies to generate text or other content based on input. In this invention, it mainly refers to a model that can generate useful information based on prompts and integrated data.

[0256] "Prompt statements" refer to input text or structured instructions constructed to guide generative artificial intelligence models to perform specific generative tasks. These include contextual information, integrated data summaries, output conditions, and descriptions of the generation goals.

[0257] "Useful information" refers to information generated by generative artificial intelligence models based on prompts and integrated data that is relevant to user needs or interests and has application value, including but not limited to recommendations, suggestions, discount tips, and guiding text.

[0258] "Display information" refers to output content that, after being converted by an information processing device, can be presented on an information terminal in the form of a graphical interface, text, or notification to prompt the user, including but not limited to page content, pop-up notifications, and message prompts.

[0259] "Additional actions" refer to the subsequent interactive behaviors that users perform after seeing useful information displayed on the information terminal, including but not limited to clicking on recommended objects, expanding or collapsing content, ignoring a prompt, adding to favorites, or initiating a purchase.

[0260] "Candidate object set" refers to a group of items or services selected by the information processing device based on attribute information and provided condition information before generating useful information, and used as candidates for output objects of the generative artificial intelligence model.

[0261] "Preference information" refers to statistical results or pattern descriptions that represent users' preferences in terms of categories, price ranges, and functional features, obtained by summarizing and analyzing behavioral information within a predetermined time period.

[0262] In this invention, the server, deployed as an information processing device in a data center or cloud computing environment, includes a processor, main memory, non-volatile storage, and a network interface. The server's processor can be a multi-core general-purpose processor, and its storage device can be a disk array or a solid-state drive array. The server communicates with multiple terminals and external generative artificial intelligence model services via the network interface. The terminals, as information terminals, can be smartphones, tablets, or desktop computing devices, and communicate with the server through a browser or local applications. Users perform operations such as searching, browsing, and clicking through the terminals.

[0263] In this invention, the server executes a pre-installed application to implement logical functions such as a behavior information collection interface, a data storage and integration module, a prompt statement construction module, a generative artificial intelligence model invocation module, and a result post-processing module. At the software level, the server can run an operating system, a database management system, and application server software. The database management system can be a relational database management system, and the application server software can be a web application runtime environment implemented using a scripting language or a compiled language.

[0264] In this invention, the terminal executes a front-end script program to monitor user input and operations on a webpage or application interface, and then packages the collected behavioral information and sends it to the server. The script executed by the terminal on the front-end page can be based on a web scripting language and uses an event listening mechanism to collect behavioral characteristics such as search term input, page browsing, product clicks, and dwell time. In this invention, the terminal utilizes local storage mechanisms or memory objects to cache temporary behavioral records, enabling batch uploading and thus reducing the number of communications with the server.

[0265] After receiving behavioral information from the terminal, the server writes this information into a predefined data table in the database management system. The server establishes structured storage for the behavioral information in the database, including user identifier fields, session identifier fields, event type fields, object identifier fields, timestamp fields, and numerical characteristic fields. The server also stores item attribute information and service attribute information in the same database, including category fields, price fields, inventory fields, tag fields, and scope of application fields, as well as related provisioning conditions, including discount conditions, combination conditions, time restrictions, and regional restrictions.

[0266] In one embodiment of this invention, the server uses a database query language to perform join and filtering operations based on identifiers and time information between the behavior information table, the item / service information table, and the provision condition information table, thereby generating integrated data. Within this integrated data, the server constructs a comprehensive data structure for each user for subsequent AI processing. This data structure includes: fields representing the user's recent search keyword sequences; fields representing the user's browsing events and their duration; fields representing the item prices, inventory, and categories corresponding to these browsed items; and fields representing currently available promotions or provision conditions. The server performs aggregation operations on this integrated data as needed, such as counting the number of views within a specific price range or the frequency of visits to a certain category, to form preference information.

[0267] In this invention, the server extracts contextual and output condition information from the integrated data through application logic, using this as input for constructing the prompt statement. In a typical implementation, the server converts a summary of the user's recent behavior within a certain time window (e.g., "user's recent search keywords: smartphone case," "user's preferred price range: 50~150 yuan") into natural language fragments, and simultaneously generates a summary of the candidate object set (e.g., "candidate items include a combination of several shockproof phone cases and high-definition tempered glass screen protectors at discounted prices") from attribute information and provided condition information. When constructing the prompt statement, the server combines the system role description, user behavior summary, candidate object summary, and output format requirements into a continuous text according to a predetermined template.

[0268] In one embodiment, the server constructs prompts of the following type as input to a generative artificial intelligence model: "System: You are an intelligent recommendation assistant for an e-commerce platform, and you need to generate personalized recommendation content for users based on user behavior data and product information."

[0269] User behavior summary: - Recent user search keywords: smartphone case - Most viewed product category in the last 7 days: Phone cases - User preferred price range: 50~150 yuan - The user has browsed wireless chargers and tempered glass screen protectors, but has not yet made a purchase.

[0270] Candidate Product Information (Partial): 1) Product ID: 12345, Name: Brand Name Anti-drop Phone Case, Price: 99 RMB, Stock: Sufficient 2) Product ID: 23456, Name: Ultra-thin phone case of a certain brand, Price: 129 yuan, Stock: Limited 3) Product ID: 34567, Name: High-definition tempered glass screen protector, Price: 29 yuan, Stock: Sufficient 4) Product ID: 45678, Name: Wireless Charger, Price: 159 yuan, Stock: Average Promotional information: - Promotion 1: Get ¥20 off any phone case + tempered glass screen protector combo. - Activity 2: Wireless chargers are eligible for a ¥30 discount on purchases over ¥200. Task: Based on the above information, please generate a recommendation message for this user in Simplified Chinese, suitable for display in the "Recommended for You" section of an e-commerce website. The message should be no more than 150 characters long and clearly state the applicable promotional offers. In another embodiment, the server may also be configured to generate prompts that require the model to generate multiple title phrases and brief descriptions, for example: Please refer to the following user behavior data: - Recent search keywords in the last 7 days: 'smartphone case', 'wireless charger' - Product categories where you spent the most time browsing: Phone cases, phone stands - Price range most frequently chosen by users: 50~150 yuan Based on the following product and promotional information: - Several brands of shockproof phone cases, with prices ranging from 89 to 139 yuan; - Special offers on tempered glass screen protectors and phone cases; - Limited-time discounts on phone holders and car mounts.

[0271] Please generate: 1) Three short sentences suitable for use as titles in the recommendation module; 2) A comprehensive recommendation description of no more than 150 words, highlighting product combinations and special offers suitable for the user; 3) The reasons for recommendation should, as far as possible, correspond to the user's browsing preferences (price range, brand, category). When the server invokes the generative AI model, it can send HTTP requests to external AI services via a network interface or use the official SDK interface. In this invention, the generative AI model can employ a deep neural network based on a transformer structure. This model includes multi-layer self-attention encoding and decoding modules, possessing a multi-head attention mechanism and a feedforward network structure. During the training phase, the generative AI model uses large-scale text data as corpus, optimizes parameters using an autoregressive objective, employs a cross-entropy loss function as the error function, and iteratively updates the weight parameters in the network using a stochastic gradient descent-like optimization method. During the usage phase, the server does not perform model training. Instead, it uses the prompt statements as the model input sequence, encodes them into discrete tags, maps them to vectors through an embedding layer, and decodes them after inference through a multi-layer self-attention network to generate the output tag sequence.

[0272] In this invention, the server employs a structured design for the prompt statements, enabling the attention layer of the generative AI model to explicitly align user behavior features with candidate item / service features within the same context window. The server incorporates the identifier information and concise attribute summaries of the candidate object set into the prompt statements. During self-attention computation, the model can assign higher attention weights to the vector representations of these candidate objects, thereby limiting the output content to a pre-selected set. By imposing constraints on the input side in this way, rather than simply filtering after output, the server reduces the generation of irrelevant text, lowers the computational burden of post-processing, and improves the consistency between the generated results and the candidate set.

[0273] In this invention, to improve processing efficiency, the server optimizes the storage structure of behavioral information and the generation logic of integrated data. The server uses an index structure (such as a composite index based on time and user identifier) ​​in the behavioral information table to accelerate time-series query operations. When generating integrated data, it only processes relevant records within the most recent time window, thereby reducing unnecessary disk reads and memory usage. The server employs a batch processing and caching strategy in the integrated data generation module, caching frequently accessed user preference information. When a user requests recommendations multiple times within a short period, the server can directly reuse the cached preference summary, incrementally updating only the most recent behavioral records, thus reducing the overall computational load.

[0274] In this invention, the server collects and stores user actions on the recommendation results (such as clicking on a recommended item or ignoring a certain type of recommendation block) as behavioral information, thereby introducing feedback information into the subsequent generation of integrated data and construction of prompt statements. In one embodiment, the server assigns a higher weight to the appended actions than to historical behaviors, employing a weighted statistical method when generating preference information, so that recent feedback behaviors account for a larger proportion in feature aggregation. Through this weighted aggregation strategy, the server enables the system to react more quickly to changes in user interests, technically shortening the number of behavioral samples required for recommendation result convergence and improving the matching degree between the generated results and the user's current intent.

[0275] In another embodiment of the invention, the server can further convert the integrated data into sparse feature vectors or compact semantic summaries via a preprocessing module before invoking the generative AI model, and include summary descriptions of these vectors in the prompt statement. The server can utilize traditional machine learning models, such as gradient boosting tree models or factorization machine models, to score the interaction features between users and items / services, inputting high-scoring objects as a candidate set into the generative AI model. Through this hybrid structure, the server enables the generative AI model to generate natural language within a more refined candidate space, thereby reducing the invalid search space during the model inference stage and improving generation speed and output accuracy.

[0276] In this invention, the terminal serves as both the interactive interface and the result display. After receiving the display information returned by the server, the terminal uses an interface rendering engine to display the recommended text and a list of candidate items / services to the user. In one embodiment, the terminal displays the natural language text output by the generative artificial intelligence model in the "Recommended for You" area, below which corresponding items or services are displayed in the form of image and text cards, including elements such as images, names, prices, and tags. When the user interacts with these cards by clicking, swiping, or adding them to favorites, the terminal records these interaction events as behavioral information and uploads them to the server, thus forming a closed loop.

[0277] In this invention, users do not need to understand the internal data processing and model structure; they participate in the system solely through natural browsing and clicking behaviors. User behavior is automatically converted into structured behavioral information on the terminal side, which is then processed and utilized by the server. During use, users receive recommendations that better match their current interests and price preferences, improving the timeliness and relevance of the recommended content.

[0278] In various embodiments of this invention, the server can adopt different modular deployment schemes based on hardware resources and business scale. For example, the behavior information collection interface module can be deployed on a separate access server, the integrated data generation module and the prompt statement construction module can be deployed on an application server cluster, and the generative artificial intelligence model invocation module can interface with external AI services through a dedicated gateway server. Through this layered architecture, the server can maintain the stability and real-time performance of behavior data writing, integrated data generation, and AI invocation even in high-concurrency request scenarios.

[0279] By employing the integrated data modeling, structured prompt design, and feedback-based dynamic update mechanism described in this invention, the server elevates the generative AI model from a passive response to single input requests to a highly efficient generation module constrained by controlled data structures and specific candidate sets. This design not only improves the relevance and accuracy of recommended content but also reduces data transfer volume and the number of calls between the server and external AI services by minimizing redundant computation and irrelevant generation, thereby achieving improvements in both computational efficiency and communication load.

[0280] In an alternative embodiment of this invention, the server can also employ a self-built generative artificial intelligence model to meet higher requirements for data security and latency. In this case, the server locally deploys a neural network model based on a transformer structure, fine-tunes the model using a pre-prepared training dataset, and employs batch gradient descent and learning rate scheduling strategies during training. The loss function can be a labeled smoothed cross-entropy function. During the training phase, the server can use data augmentation methods, such as time window resampling of user behavior sequences and synonymous representation expansion of item attribute descriptions, thereby improving the model's robustness to different behavior patterns and product descriptions. During the inference phase, the server utilizes efficient matrix operation libraries and hardware acceleration resources (such as graphics processing units) to perform multi-head attention and feedforward network operations, thereby shortening response time while maintaining output quality.

[0281] In summary, through the aforementioned hardware configuration, data structure design, integrated processing algorithm, prompt statement construction method, and controlled invocation of the generative artificial intelligence model, the server achieves efficient utilization of user behavior information and item / service information. This invention establishes a closed-loop technical solution within the computer, encompassing "behavioral data—integrated data—prompt statements—model output—feedback behavior," making the contextual information organization of the generative artificial intelligence model more efficient and the output range more controllable. This significantly improves processing speed, recommendation accuracy, and overall system resource utilization efficiency in personalized information generation scenarios.

[0282] use Figure 13 The processing procedure is explained.

[0283] Step 1: When a user accesses a service page, the terminal loads the front-end program and begins collecting behavioral information.

[0284] Input: The actual actions a user performs on the terminal (such as typing, clicking, scrolling, or pausing).

[0285] The terminal executes scripts in a browser or local application, registering event listeners on page elements (search box, product list, buttons, etc.). When a user enters keywords in the search box, the terminal records the input and a timestamp; when a user clicks on an object, the terminal records the object identifier, click location, and current page type; when a user opens or closes a page, the terminal calculates the dwell time based on the entry and exit times. The terminal encapsulates these events into structured records and temporarily stores them in memory or local storage.

[0286] Output: A set of raw behavioral event records (including event type, user ID, object ID, timestamp, dwell time, etc.).

[0287] Step 2: The terminal packages the cached behavioral events and uploads them to the server.

[0288] Input: The set of original behavioral event records obtained in step 1.

[0289] When the upload conditions are met (the cumulative number of events reaches a threshold or the time interval is reached), the terminal groups the locally cached behavior events by user identifier and session identifier, and serializes them into request body data. The terminal constructs a request through the network interface, sets the target URL and necessary headers (including user identification information), and sends the request containing the behavior events to the server using the network protocol.

[0290] Output: Data packets sent to the server (containing structured data of user behavior events).

[0291] Step 3: The server receives behavioral information uploaded by the terminal, and then parses and stores it.

[0292] Input: Behavior event data packets received from the terminal.

[0293] The server receives requests via a network interface and parses the request body in the application, mapping the behavioral events within it to internal data structures by field. The server performs validity checks on the data fields, such as verifying the user identifier format, whether the event type is within a predefined set, and whether the timestamp is within a reasonable range. For the validated behavioral events, the server constructs a database insert statement, calls the database management system, writes the events to the corresponding data tables, and creates indexes by user identifier and time.

[0294] Output: Standardized behavioral information records stored in the database, and an index structure that can be used for subsequent queries.

[0295] Step 4: The server reads behavioral and item / service information from the database and generates integrated data.

[0296] Input: Records of behavioral information, item / service attribute information, and provision conditions in the database.

[0297] The server generates a personalized user identifier based on current needs and retrieves the user's behavioral information within a predetermined time window from the database. Simultaneously, the server retrieves item / service records associated with these behaviors from the attribute information table and currently valid offers or restrictions from the provision conditions information table. The server uses join operations to associate behavioral records with item / service records by identifier and filters expired or invalid conditions based on time information. The server aggregates and statistically analyzes the association results, such as calculating the number of times the user browses across price ranges, the browsing frequency of each category, and the intensity of recent interactions. These statistics are then packaged with the original records to form a unified data structure describing the relationship between the user and related items / services.

[0298] Output: Aggregated data for a specific user (including behavioral sequences, attribute fields, conditional information, and preference statistics).

[0299] Step 5: The server constructs context information and output conditions based on the integrated data, and generates prompt statements.

[0300] Input: The aggregated data obtained in step 4.

[0301] The server analyzes and integrates behavioral sequences and statistical results from the data, extracting representative information such as "recent search keywords," "most frequently viewed categories," "preferred price ranges," and "types of recently viewed but not yet purchased items." The server transcribes this information into short natural language sentences and simultaneously selects a set of candidate items from item / service attributes and provision conditions, generating text fragments containing item identifiers, brief attributes, and discount summaries. Based on a preset template, the server sequentially concatenates system role descriptions, user behavior summaries, candidate item summaries, and output requirements to form a complete prompt statement, guiding the generative AI model to generate results within a defined scope.

[0302] Output: Prompt text for generative artificial intelligence models, along with corresponding integrated data reference information.

[0303] Step 6: The server invokes a generative artificial intelligence model to generate useful information based on prompts and integrated data.

[0304] Input: The prompt text generated in step 5 and the associated integrated data.

[0305] The server encodes the prompt statement as an input sequence into tags and invokes a generative artificial intelligence model deployed locally or externally. Internally, this model employs a transformer-based neural network, using multi-head self-attention layers to encode the input tag sequence and modeling contextual relationships through feedforward networks and residual connections. During training, the model adjusts its weights by minimizing cross-entropy loss. In the current inference phase, it predicts the conditional probability distribution of the next tag based on the prompt statement and progressively generates the output tag sequence according to a set temperature and sampling strategy. During the invocation process, the server controls the termination of generation based on the output length and a termination tag, and then decodes the output tag sequence into natural language text.

[0306] Output: Useful information text for the target user (such as personalized recommendation copy, candidate ranking instructions, etc.).

[0307] Step 7: The server performs post-processing on the generated results and converts them into display information before sending them to the terminal.

[0308] Input: Useful information text generated in step 6 and candidate object identifiers.

[0309] The server performs format checks and length truncation on the generated text, filtering out segments irrelevant to the intended topic. Based on candidate object identifiers, the server re-extracts necessary display fields from the database, such as name, image link, price, and brief tags, and combines the generated text with these fields to form the interface display structure. The server encapsulates this structure as response data and sends it to the requesting terminal via a network interface.

[0310] Output: Display information data for the terminal (including recommended text and display data of corresponding objects).

[0311] Step 8: The terminal receives display information sent by the server and displays it on the interface, while also collecting user actions to add useful information.

[0312] Input: The display information data output from step 7.

[0313] After receiving the display information, the terminal uses its interface rendering engine to display the generated recommendation text in the designated area and generates a list of images and text or cards according to the order and attributes provided by the server. The terminal then registers event listeners for clicks, swipes, expand / collapse, etc., on these display elements. When a user clicks, adds to favorites, or ignores a recommendation, the terminal records these actions as new behavior events, attaching an identifier from the original recommendation content or location to indicate which generation result the action is associated with. The terminal then packages these new behavior events again and uploads them to the server in the same way.

[0314] Output: New behavioral event records (including appended operations related to useful information from the previous round), which the server uses for the next round of data integration updates and prompt adjustments.

[0315] 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".

[0316] In a network environment, when users access information resources through their devices, they often generate a large amount of scattered behavioral and browsing data. Existing technologies typically employ rule-based matching or simple statistical models, making recommendations based solely on static features such as browsing history or click counts. This makes it difficult to capture dynamic changes in user interests in a timely manner, and even more difficult to fully utilize the implicit information of the user's current emotional state. Therefore, the following technical problems exist: (1) Behavioral data and browsing data are mostly stored in isolation in the form of logs, lacking time-series structured processing for real-time recommendations, resulting in a large delay in subsequent analysis and generation processing, and failing to achieve highly real-time personalized information push. (2) Existing systems fail to integrate user behavior features, candidate content features and sentiment features in a unified data structure, making the recommendation results reflect only "what users have seen" and not "what users really want in the current context"; (3) Even when generative artificial intelligence models are introduced, they usually only take a small amount of static information or simplified keywords from users as input directly. They do not use a systematic prompt statement construction process to convert multi-source and multi-dimensional features into natural language instructions that are optimal for the model, resulting in unstable quality of generated results, insufficient relevance, and low efficiency in the use of computing resources. (4) The user’s interaction results with the system (such as responses to recommended content and subsequent behaviors) are not effectively fed back into the process of constructing behavioral data and prompt statements. As a result, the system is unable to form an adaptive closed-loop optimization mechanism and cannot continuously improve recommendation accuracy and response speed at the computational structure level.

[0317] Therefore, there is an urgent need for an improved computer implementation scheme: within the same server-side data processing framework, user behavior information and browsing information are accumulated and organized in a time sequence and structured manner, combined with the features of candidate information to form a unified feature quantity, and then sentiment information is generated through sentiment analysis. Based on this, high-quality prompt statements are automatically constructed for use by generative artificial intelligence models, thereby improving the real-time performance, accuracy and resource utilization efficiency of personalized information generation and push at the system level, and achieving an improvement in computer technology itself.

[0318] 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.

[0319] In this invention, the server includes means for receiving user behavior information and browsing information from a user terminal, accumulating and storing the user behavior information and browsing information in chronological order, and organizing the user behavior information and browsing information into a data structure suitable for subsequent calculations; means for retrieving candidate information from information resources based on the accumulated user behavior information and browsing information, and associating and integrating the candidate information with the user behavior information and browsing information to generate a feature quantity representing the user's interest level; means for performing sentiment analysis processing based on the user behavior information, the browsing information, and the feature quantity to generate sentiment information representing the user's emotional state; and means for combining the user behavior information and the browsing information... The system includes: a device for converting the feature quantities and emotional information into natural language and constructing prompt statements to instruct a generative artificial intelligence model to generate relevant information; a device for inputting the prompt statements into the generative artificial intelligence model and obtaining the generation result, including relevant information and explanatory text in response to user interests and emotional states; a device for extracting personalized information for presentation to the user from the generation result, converting it into a predetermined output format, and sending it to the user terminal in real time via a communication path; and a control device for feeding back additional user behavior information and response information from the user terminal into the accumulated stored user behavior information and updating the generation of the prompt statements and the input processing to the generative artificial intelligence model accordingly. This allows for the formation of a unified data processing and prompt statement construction mechanism within the server for generative artificial intelligence models, achieving tight coupling processing of user behavior data, candidate content features, and emotional features. While ensuring real-time performance, this significantly improves the matching degree between the generated results and the user's current interests and emotional states, reduces invalid calculations and repetitive data processing, and enhances the overall processing efficiency and technical performance of the computer system in personalized information generation and push tasks.

[0320] "User terminal" refers to an information processing device operated by a user to access network services and interact with the system, including but not limited to computer equipment, mobile terminal devices, and applications or browsers running on them.

[0321] "User behavior information" refers to data recorded on a user's terminal that reflects the user's operational behavior, including but not limited to page access records, click records, scrolling operation records, search input records, shopping cart records, and purchase records.

[0322] "Browsing information" refers to data related to the content viewed by a user on their terminal, including but not limited to the identifier of the accessed page, resource address, content category, display duration, and the display status of page elements.

[0323] "Information resources" refers to data sets that can be accessed by servers through communication networks and used to generate related information for users, including but not limited to product information sets, document information sets, multimedia information sets, and other structured or unstructured data sets.

[0324] "Candidate information" refers to information units retrieved by the server from information resources that may be selected as content to be presented to the user, including but not limited to candidate products, candidate documents, candidate multimedia content and their related attribute information.

[0325] "Feature quantity" refers to quantitative or qualitative feature data calculated by the server based on the correlation between user behavior information, browsing information and candidate information, and used to represent the degree of user interest or content relevance.

[0326] "Sentiment analysis processing" refers to the data processing process in which the server analyzes text information and operation patterns obtained from user behavior information and browsing information in order to identify the user's emotional state and output corresponding emotional information.

[0327] "Emotional information" refers to data obtained through sentiment analysis that represents a user's current or specific emotional state over a period of time, including but not limited to sentiment category, sentiment intensity, and confidence level related to the sentiment.

[0328] "Natural language" refers to the language forms used by humans in everyday communication, including but not limited to text expressed in the form of sentences, phrases or paragraphs, excluding purely coded or markup languages ​​used exclusively for internal machine processing.

[0329] "Prompt statements" refer to input text or instruction text that is constructed by the server based on user behavior information, browsing information, feature quantities, and emotional information, expressed in natural language, and used to instruct generative artificial intelligence models to generate target information content.

[0330] "Generative artificial intelligence models" refer to artificial intelligence models that can automatically generate text information, explanatory information, or other forms of output information based on input prompts, including but not limited to text generation models built based on machine learning or deep learning technologies.

[0331] "Generated results" refers to the output data of the generative artificial intelligence model after receiving a prompt statement, including but not limited to related information, explanatory text, recommendation reasons, and other content that can be used to constitute personalized information.

[0332] "Personalized information" refers to information content that the server selects or combines from the generated results and customizes based on the interests and emotional state of a specific user, including but not limited to recommendation lists, explanatory text, promotional tips, and guidance information.

[0333] "Communication path" refers to the network communication channel used to transmit data between the server and the user terminal, including but not limited to wired networks, wireless networks, and the communication protocols and connection methods used thereon.

[0334] "Additional user behavior information" refers to user behavior information generated by the user on the user's terminal and collected again after the server sends personalized information to the user's terminal. This includes clicking, ignoring, and responding to recommended content, as well as subsequent browsing and purchasing behaviors.

[0335] "Response information" refers to explicit or implicit feedback data from users in response to personalized information provided by the server, including but not limited to evaluation information, interest tags, rejection tags, dwell time, and interaction frequency.

[0336] "Control device" refers to a functional unit implemented by the processor in the server and the program it executes, used to update and schedule the generation of prompt statements and the input processing of generative artificial intelligence models based on accumulated user behavior information, emotional information and response information.

[0337] The embodiments of the present invention will be described in accordance with the system structure defined in the claims. The following embodiments are merely examples, and those skilled in the art can make various modifications and substitutions without departing from the concept of the present invention.

[0338] In the following description, a server is composed of one or more processors, memory, and communication interfaces, running an operating system (such as a general-purpose server operating system), as well as middleware and applications; a terminal is an information processing device operated by a user; and a user is the entity that uses the terminal to access network services. For ease of explanation, the following descriptions of each part of the processing will use "server," "terminal," and "user" as the subjects.

[0339] I. Overall Hardware and Software Composition of the System In one implementation, the server includes: a multi-core central processing unit, main memory, a mass storage device, a network interface controller, and an optional graphics processing unit. The server runs application server software components on an operating system, including: Server-side network communication frameworks (e.g., general HTTP server frameworks), application layer frameworks (e.g., general scripting language frameworks), data processing libraries (e.g., data table processing libraries), machine learning frameworks (e.g., general tensor computation frameworks or tensor computation libraries), message queue system client components, database access components, etc.

[0340] In one implementation, a terminal includes a mobile terminal device or a fixed terminal device, on which an operating system (such as a mobile operating system or a desktop operating system), a web browser, or a local application runs. The terminal runs a script engine (such as a JavaScript engine) in the browser or a data tracking SDK in the local application to collect user behavior data.

[0341] Users access the server through a terminal and interact with the server through a communication network.

[0342] II. Data Acquisition and Preprocessing on the Terminal Side In this invention, the terminal acquires user behavior information and browsing information, and performs preliminary structured organization locally.

[0343] When a user accesses a web page, the terminal intercepts interface events and page state information through browser scripts or application-level embedded components. The terminal categorizes the following information as user behavior information: - User click events, swipe events, and scroll events on the interface; - Keywords entered by the user in the search input box; - Users add items to their favorites, shopping cart, etc.; - A user's click or ignore of specific recommended content.

[0344] The terminal categorizes the following information as browsing information: - Uniform Resource Identifier for the current page or resource; - The content category identifier to which the page belongs; - Page load time and page close time, page dwell time; - Markup or structural information for the main content area of ​​the page.

[0345] The terminal generates a record entry for each behavioral event, containing fields such as timestamp, user identifier, session identifier, event type, and resource identifier, through script logic or SDK. The terminal organizes multiple record entries into a list-style data structure and sorts them in chronological order. The terminal temporarily stores this data locally and sends it to the server through the communication interface according to preset conditions (such as the number of records reaching a threshold or the time interval reaching a predetermined value).

[0346] By performing time-sequence sorting and field standardization on the terminal side, the parsing burden on the server side can be reduced, and duplicate field transmissions can be decreased, thereby technically reducing the communication load and improving the processing throughput of the server backend.

[0347] III. Server-side data accumulation, structuring, and feature generation After receiving user behavior and browsing information from the terminal, the server transforms it into an internal data structure in its memory. The server accumulates and stores the data for each user in a hierarchical key-value or tabular row format. For example, the server can maintain a time-series array for each user, with array elements being structured event record units, indexed by user identifiers and timestamps.

[0348] The server uses a data processing library to map user behavior and browsing information into numerical and categorical features. Based on content identifiers in pre-stored information resources (such as product information sets and document information sets), the server matches the resource identifiers browsed by the user with corresponding entries in the information resources to determine a candidate information set. The server further calculates the relevance between each candidate information and the user's recent behavior within the candidate information set. This relevance can be constructed using the following features: - The degree of match between the category of the candidate information and the categories recently viewed by the user; - Frequency of user visits to this category; - The average time users spend on this category page; - The depth of user interaction with this category of content (e.g., the number of drill-down levels).

[0349] The server combines these features into a feature vector, or feature quantity, which quantifies the user's interest in candidate information. Feature quantities can include continuous features (such as frequency and duration), discrete features (such as category encoding), and derived features that have undergone normalization and standardization. Based on the feature quantities, the server constructs data structures, such as matrices or tensors, in memory for subsequent model calls to reduce the overhead of subsequent feature recalculation, thereby improving overall computational efficiency.

[0350] IV. Sentiment Analysis and Processing and Sentiment Information Generation on the Server To reflect the user's current emotional state, the server incorporates a sentiment analysis module. The server extracts semantically relevant text fragments from user behavior and browsing information, such as: - The user's search keyword sequence; - Natural language text submitted by users in comments, Q&A, or input boxes; - Description text attached when a user selects or rejects specific content.

[0351] The server preprocesses the text, including word segmentation, stop word removal, and lemmatization, and then uses a sentiment analysis module to output sentiment information. In one implementation, the sentiment analysis module can be a neural network model, such as a multi-layer bidirectional recurrent network or an attention-based text classification network. Its input is a sequence of word vectors or sub-word vector sequences, and its output is a probability distribution of several predefined sentiment categories, such as "joy," "curiosity," "hesitation," and "annoyance."

[0352] The server selects the sentiment category with the highest probability from the output probability distribution as the user's current sentiment tag, while simultaneously retaining the probability value of that category as a measure of sentiment intensity. The server adds the sentiment tag and sentiment intensity fields to the data structure associated with the user, managing them uniformly with behavioral features.

[0353] By combining user behavior features and emotional features in the same data structure, the server can provide more multidimensional information input to the generative artificial intelligence model at the computational level, thereby improving the matching degree between the generated results and the user's actual needs.

[0354] V. Constructing Server-Side Messages and Invoking Generative Artificial Intelligence Models In this invention, the server constructs prompt statements to input user behavior information, browsing information, feature values, and emotional information into a generative artificial intelligence model in natural language form. In one implementation, the generative artificial intelligence model can be a multi-layer sequence-to-sequence network based on a self-attention mechanism, with its parameters obtained through pre-training on a large-scale text corpus. The model's internal structure may include embedding layers, multi-layer encoder-decoder stacks, feedforward networks, and multi-head attention mechanisms, utilizing positional encoding to represent sequence order.

[0355] The server performs the following data processing operations when constructing the prompt statement: The server selects several recent browsing and search records from the user, extracting key content categories and keywords. Based on the feature calculation results, the server selects several candidate information with the highest interest scores and converts their basic attributes (such as category, price range, and discount tags) into natural language descriptions. The server converts sentiment tags and sentiment intensity into textual expressions describing the user's current state of mind. Then, the server combines this information into coherent natural language paragraphs according to predefined templates.

[0356] For example, the server can generate the following prompt: "The user recently searched for 'smartphone case', browsed the mobile phone accessories category page multiple times, and their current sentiment is 'curiosity'. Based on this behavioral and sentiment information, please select the 5 most suitable products to recommend from the candidate products, and generate a brief Chinese recommendation reason for each product." For example, the server can generate the following prompt: "Based on the user's browsing history and shopping cart contents over the past week, we analyze the promotional activities or discount information that the user may have missed, and explain in 3 to 5 sentences in Chinese how these offers can help the user save money." The server encodes the constructed prompts into a text sequence and inputs it into the generative AI model through its interface. The server sets model inference parameters, such as the maximum generation length, random sampling temperature, and candidate sampling threshold, to strike a balance between generation quality and computational cost. Based on the input prompts, the generative AI model internally calculates the conditional probability distribution of the sequence using attention mechanisms and hierarchical feedforward networks, and outputs natural language text through progressive decoding.

[0357] After receiving the generated results from the model, the server parses the results into a structured form for subsequent filtering and display. For example, a continuous text can be parsed into several "recommended items," each containing a product name, a description of key features, and a reason for recommendation.

[0358] VI. Personalized Information Filtering and Output Formatting on the Server After receiving the generated results, the server does not directly send all the information to the terminal. Instead, it performs a secondary filtering by combining feature data and sentiment information. The server processes each generated recommendation item as follows: - Check the availability status of the corresponding candidate information in the information resources, such as whether it has been removed from the shelves or whether it is in stock; - Calculate the interest score and sentiment fit of the candidate information corresponding to this item; - If the text description of an entry does not match the sentiment tag (e.g., if the tone is overly inflammatory when the user's sentiment is "hesitant"), the server may lower the entry's ranking or remove it.

[0359] The server processes the data using the aforementioned techniques, matching the generated results with the user's current context from multiple dimensions. This reduces information irrelevant to the user's needs or that may cause offense, thereby improving recommendation quality. The server then converts the retained entries into a predetermined output format, such as a key-value structure or a tag structure, and adds necessary metadata (such as display order and jump link identifiers).

[0360] The server uses a lightweight data format during the output phase to reduce data packet size and parsing complexity, thereby reducing network transmission latency and terminal parsing time, and shortening the overall response time.

[0361] VII. Terminal reception and display, and user interaction After receiving personalized information from the server, the terminal uses a local parsing component to convert structured data into interface elements. Based on the display order and recommendation reasons provided by the server, the terminal generates a list view or card view on the interface, displaying product thumbnails, names, prices, and recommendation text output by a generative AI model. The terminal can trim the length of the recommendation reasons to fit different screen sizes.

[0362] Users can view these recommended items on the terminal interface and choose to click to view details, add to favorites or shopping cart, or ignore items they are not interested in. This new user behavior information is then collected by the terminal and fed back through the server loop, allowing the system to consider this new information in subsequent prompt construction and feature calculations, forming a dynamically updated closed loop.

[0363] VIII. Technical Effects and Their Causal Relationship Through the specific data structure design, feature calculation methods, sentiment analysis module, and prompt statement construction process described above, the server enables the generative artificial intelligence model to no longer rely solely on simple keywords during reasoning, but rather on preprocessed and fused high-dimensional structured features. Therefore: - Because behavioral information is accumulated and stored in a structured manner according to time series, the server can quickly access the data in the most recent window through the index when calculating feature values, reducing the overhead of scanning the entire log and thus improving processing speed; - Because candidate information features and user sentiment features are uniformly embedded in the prompt statements, generative artificial intelligence models can utilize multiple information sources in a single reasoning process, which helps to improve the consistency between the generated results and the user's current state, thereby improving recommendation accuracy; - Because the server filters and optimizes the format of the results before sending them to the terminal, redundant content transmission is avoided, thereby reducing communication load and terminal rendering time.

[0364] Furthermore, during the model training phase, the server can perform supervised or self-supervised learning on generative AI models and sentiment analysis models. The server uses a loss function (such as cross-entropy loss) to measure the difference between the model output and the target output, and updates the model weights through backpropagation. The server can employ data augmentation methods, such as synonym replacement, sentence perturbation, and behavior sequence resampling, to improve the model's robustness in different scenarios. Through these techniques, the model internally develops a complex parameter space representation capability, thereby achieving abstract modeling of user behavior patterns and emotional states—something difficult to achieve in traditional systems primarily based on manual rules.

[0365] In this invention, the server employs a template with a specific logical order when constructing prompt statements. It aggregates user behavior characteristics, candidate information characteristics, and emotional information using a high-dimensional rule combination method that is not feasible for manual human intervention. This allows the generative artificial intelligence model to operate in a higher-dimensional feature space. This process does not simply simulate human editing of recommendation statements; rather, it implements a unique data fusion and instruction generation mechanism within the computer, enabling high-concurrency operation in environments with large-scale users and information resources. This represents an improvement to computer technology itself.

[0366] IX. Optional Implementation Forms and Variations In one variant implementation, the server can adopt different generative artificial intelligence model structures, for example: - Use deeper self-attention networks to improve long text modeling capabilities; - Use a lightweight network architecture to accommodate deployments in edge server or resource-constrained environments.

[0367] In another implementation, the server can combine or replace the sentiment analysis module according to different application scenarios. For example, it can use a text sentiment classifier based on convolutional networks or a multimodal sentiment analysis network that incorporates physiological signals. The server can also adjust the composition of features, such as adding features related to time intervals, device types, and geographical locations, to further improve recommendation accuracy.

[0368] In another implementation, the terminal can also locally cache some model inference results or feature values. When users repeatedly access the same type of content in a short period of time, the terminal can reduce the load on the server, realize collaborative computing between the front end and the back end, and thus further improve the overall system performance.

[0369] Through the above embodiments, this invention not only provides a personalized information generation scheme based on a generative artificial intelligence model and prompt statements, but also achieves comprehensive optimization of server-side computing efficiency, recommendation result accuracy, and communication load through specific designs for data structure, feature generation, sentiment analysis, prompt statement construction, and result filtering, thereby producing significant technical effects at the computer technology level.

[0370] use Figure 14 The processing procedure is explained.

[0371] Step 1: Users access services and generate actions using terminals. Users access web pages through a browser or application on their device to perform actions such as searching, browsing, clicking, and adding items to their shopping cart.

[0372] Input: User actions (clicks, input, scrolling, etc.).

[0373] Output: Raw interactive events (unstructured data) that occur locally on the terminal.

[0374] The terminal captures these events at the interface layer, such as listening for click events, input box change events, and page load / unload events, and temporarily stores them in memory as event objects for subsequent structured processing.

[0375] Step 2: Terminal collects and structures user behavior and browsing information The terminal converts the raw events captured in step 1 into structured records, including fields such as user identifier, session identifier, event type, resource identifier, and timestamp, and distinguishes them into "user behavior information" and "browsing information".

[0376] Input: Raw interaction events (such as click coordinates, input text, page URL, time point).

[0377] Output: A structured list of events arranged in chronological order (data structures containing user behavior and browsing information).

[0378] The terminal generates a record object in a uniform format for each event, groups multiple records into an array or list, sorts them by timestamp, and temporarily stores them in local cache or session storage for batch sending to the server.

[0379] Step 3: The terminal sends behavioral and browsing information to the server. When the terminal meets the predetermined conditions (such as the number of records reaching a threshold or the time interval arriving), it packages the structured event list into a request message and sends it to the server via the network protocol.

[0380] Input: A structured list of events (containing multiple user behavior and browsing information).

[0381] Output: A network request containing structured event data (a message transmitted to the server).

[0382] The terminal serializes the event list into a text format, writes the user identifier, session identifier, and client type into the request header or request body, and sends the request to the specified interface of the server through the network stack.

[0383] Step 4: The server receives and parses the data sent by the terminal. The server's communication module receives requests from the terminal, parses the event list from the request body, and verifies the correctness and integrity of the data.

[0384] Input: A network request sent by the terminal (containing structured event data).

[0385] Output: Event data structures available in server memory (sets of user behavior information and browsing information).

[0386] The server calls a parsing function to convert text-formatted data into an internal object or recordset, checks if necessary fields exist, discards invalid records or logs errors, and stores valid records in a temporary buffer.

[0387] Step 5: The server accumulates and stores behavioral and browsing information in chronological order. The server inserts the event records parsed in step 4 into persistent storage, while maintaining a time-ordered data structure for each user.

[0388] Input: A valid collection of event records (with fields such as user ID and timestamp).

[0389] Output: A dataset of behavioral logs organized by user and time index.

[0390] The server sorts the event set by user ID and timestamp, writes it to a database or time series table, and creates an index to enable quick retrieval of events by time window and user.

[0391] Step 6: The server retrieves candidate information from information resources. Based on accumulated behavioral and browsing information, the server retrieves potential candidate information, such as product entries or content entries, from the information resource database.

[0392] Input: User's recent behavior and browsing information, and information resource database (containing multiple information records).

[0393] Output: A list of candidate information related to the user's recent behavior.

[0394] The server performs a search in the information resource database based on query conditions (such as category matching, keyword matching, and historical click relevance), marks entries that meet the conditions as candidate information, and encapsulates them into a list of records containing identifiers, categories, attributes, etc.

[0395] Step 7: The server generates features representing the degree of interest. The server correlates behavioral information, browsing information, and candidate information to generate a feature vector representing the intensity of a user's interest in each candidate information.

[0396] Input: Candidate information list, user behavior information (number of views, number of clicks, duration of stay, etc.), browsing information (page category, time distribution, etc.).

[0397] Output: A set of features for each candidate (containing multiple numerical and categorical features).

[0398] The server calculates frequency statistics, time weights, and category similarity for each candidate information. After normalizing and other data processing, these values ​​are combined into a fixed-dimensional feature vector and stored in memory or a feature storage module.

[0399] Step 8: The server performs sentiment analysis processing to generate sentiment information. The server extracts relevant text or operation patterns from user behavior and browsing information, and identifies the user's current emotional state through the sentiment analysis module.

[0400] Input: Recent text content of the user (search terms, comments, etc.) and operation pattern data (repeated browsing, quick exit, etc.).

[0401] Output: Emotional information (including emotion category and emotion intensity).

[0402] The server segments and encodes the text, inputs the operation mode features and text features into the sentiment analysis model, calculates the probability of each sentiment category, selects the category with the highest probability as the sentiment label, and retains the corresponding probability value as the sentiment intensity.

[0403] Step 9: Server integrates behavioral characteristics and emotional information The server merges the features representing the degree of interest with the emotional information to form a unified data structure, which serves as the basis for subsequent prompt construction and generative artificial intelligence model invocation.

[0404] Input: Feature set (interest features for candidate information), sentiment information (sentiment labels and sentiment intensity).

[0405] Output: A data structure of user interest features with an emotional dimension.

[0406] The server adds a sentiment-related field to the user feature structure, adjusts the interest scores in the feature values ​​according to sentiment intensity (e.g., weighting content related to positive sentiment), and writes the results to a cache area used for recommendation calculation.

[0407] Step 10: Server construct prompt statement Based on the fused feature data, the server converts user behavior information, browsing information, feature quantities, and sentiment information into prompts in natural language form, which are used to instruct the generative artificial intelligence model to generate related information.

[0408] Input: Summary of recent user behavior, summary of candidate information, and fused interest and sentiment features.

[0409] Output: One or more natural language prompts.

[0410] The server extracts representative behaviors (such as recent search terms and browsing categories) from feature data, selects candidate information with the highest interest scores, describes their attributes in natural language, adds textual descriptions of the user's current mindset based on sentiment tags, and then combines them into a complete prompt statement using a fixed template.

[0411] Step 11: The server will input the prompt into the generative artificial intelligence model and obtain the generated results. The server calls the generative artificial intelligence model interface, sends the prompt statement generated in step 10 to the model, and obtains the associated information and explanatory text generated by the model.

[0412] Input: Natural language prompts (including descriptions of behavioral information, candidate information, and sentiment information).

[0413] Output: The generated result text (including related information, recommendations, copy, etc.).

[0414] The server encodes the prompt statement into an input format that the model can accept, calls the model's inference function, the model calculates the input based on its internal parameters and outputs a text sequence, and the server receives the output and converts it into readable text.

[0415] Step 12: The server parses the generated results and extracts personalized information. The server parses the output text of the generative artificial intelligence model, breaks it down into multiple candidate recommendation items, and extracts personalized information to be presented to the user.

[0416] Input: The resulting text (continuous natural language content).

[0417] Output: A structured collection of personalized information (including item titles, content, reasons for recommendation, etc.).

[0418] The server identifies the segmentation structure, sequence number, or marker in the text, divides it into multiple independent recommendations, analyzes the object identifiers or category information involved in each content, and generates a structured record for each recommendation.

[0419] Step 13: The server filters and ranks personalized information based on feature values ​​and sentiment information. Based on the structured and personalized information set, the server combines previously generated feature data and sentiment information to select the most suitable items for the current user and determine their display order.

[0420] Input: Personalized information set (multiple recommendation items generated by the model), features (interest level), and sentiment information.

[0421] Output: A filtered and sorted list of personalized information.

[0422] The server calculates a comprehensive score for each recommendation (combining interest score, sentiment fit, and content relevance), filters out items with low scores or those that conflict with sentiment tags, and sorts them from high to low according to the comprehensive score, keeping the top few for display.

[0423] Step 14: The server converts personalized information into a predetermined output format. The server converts the filtered list of personalized information into a data format agreed upon with the terminal and adds the necessary metadata to support client display.

[0424] Input: A list of filtered and sorted personalized information.

[0425] Output: Output data structure conforming to the terminal protocol (e.g., a data object containing an array of entries and metadata).

[0426] The server encapsulates the title, description, jump marker, and display order of each recommended item into a unified structure, adds auxiliary fields such as version number and timestamp, and serializes it into a compact transmission format to reduce data size.

[0427] Step 15: The server sends personalized information to the terminal in real time via a communication path. The server uses a network communication module to send the data generated in step 14 to the corresponding user terminal via the network, thereby achieving real-time push.

[0428] Input: A data object with a predefined output format.

[0429] Output: Response messages or push messages sent to the terminal.

[0430] The server selects the sending method (e.g., synchronous response, asynchronous push) based on the connection status and monitors for errors during the sending process, such as network anomalies or timeouts, and performs retry or degradation processing when necessary.

[0431] Step 16: The terminal receives and parses the personalized information sent by the server. The terminal obtains messages sent by the server from the network stack, parses the data portion, and restores it into a local data object.

[0432] Input: A response message or push message from the server.

[0433] Output: A list of recommended items local to the terminal.

[0434] The terminal calls the parsing logic to convert the text format into an internal structure, verify field integrity and version compatibility, and cache these recommended items in memory to prepare for display on the interface.

[0435] Step 17: The terminal displays personalized information in the user interface. Based on the parsing results, the terminal dynamically generates a recommendation area in the user interface layout, displaying the title, description text, icon, or thumbnail of the recommended items.

[0436] Input: A list of local recommended items.

[0437] Output: An interactive recommendation interface displayed on the screen.

[0438] The terminal binds a click event listener to each entry, sets appropriate font size and line break rules for the explanatory text, and arranges them according to the display order provided by the server, so that users can browse and operate them intuitively.

[0439] Step 18: Users interact with recommended content and generate subsequent behaviors. When browsing recommended content, users can click on items that interest them, go to the details page, add items to their shopping cart, or ignore items that do not interest them.

[0440] Input: Recommended content and interactive controls displayed on the screen.

[0441] Output: New user action events (click, jump, ignore, etc.).

[0442] These user actions are captured and recorded again by the terminal, forming new user behavior information for subsequent feedback and updates.

[0443] Step 19: The terminal sends additional user behavior information and response information back to the server. The terminal records the user's actions after the recommendation is displayed as additional user behavior information and response information, such as clicking on a recommendation, dwell time, closing action, etc., and sends them to the server.

[0444] Input: User actions on the recommendation screen.

[0445] Output: A network request containing additional user behavior information and response information.

[0446] The terminal processes these events in a structured manner, adds identifiers associated with the recommended items, and then packages and sends them to the server's designated feedback interface via the network interface.

[0447] Step 20: The server updates the accumulated behavior information and prompt statement generation strategy based on feedback information. The server receives additional user behavior and response information from the terminal, merges it into the existing behavior log, and adjusts the feature values ​​and prompt statement construction rules.

[0448] Input: Additional user behavior and response information, existing behavior logs, and feature data.

[0449] Output: Updated cumulative behavior information, updated feature values, and updated prompt statement construction parameters.

[0450] The server integrates the feedback data, updates the user interest model (e.g., increases the weight of the category to which the clicked item belongs and decreases the weight of categories that are ignored multiple times), and adjusts the behavioral summary strategy and sentiment description weights used when constructing the prompts accordingly, so that prompts that are more in line with user preferences can be generated the next time the generative AI model is called.

[0451] 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.

[0452] 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.

[0453] 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.

[0454] 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.

[0455] 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.

[0456] Second Implementation Method Figure 3 An example of the configuration of the data processing system 210 according to the second embodiment is shown.

[0457] 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.

[0458] 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).

[0459] 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.

[0460] 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.

[0461] 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).

[0462] 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.

[0463] 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.

[0464] 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.

[0465] 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).

[0466] 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.

[0467] 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".

[0468] 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.

[0469] 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.

[0470] 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.

[0471] 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.

[0472] 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.

[0473] 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.

[0474] 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.

[0475] 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.

[0476] 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.

[0477] Third Implementation Method Figure 5 An example of the configuration of the data processing system 310 according to the third embodiment is shown.

[0478] 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.

[0479] 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).

[0480] 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.

[0481] 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.

[0482] 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).

[0483] 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.

[0484] 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.

[0485] 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.

[0486] 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.

[0487] 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.

[0488] 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".

[0489] 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.

[0490] 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.

[0491] 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.

[0492] 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.

[0493] 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.

[0494] 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.

[0495] 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.

[0496] 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.

[0497] 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.

[0498] Fourth Implementation Method Figure 7 An example of the configuration of the data processing system 410 according to the fourth embodiment is shown.

[0499] 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.

[0500] 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).

[0501] 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.

[0502] 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.

[0503] 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).

[0504] 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.

[0505] 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.

[0506] 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.

[0507] 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.

[0508] 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.

[0509] 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.

[0510] 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".

[0511] 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.

[0512] 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.

[0513] 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.

[0514] 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.

[0515] 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.

[0516] 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.

[0517] 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.

[0518] 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.

[0519] 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.

[0520] Furthermore, the emotion-specific model 59, acting as an emotion engine, can determine a user's emotion based on a specific mapping. Specifically, the emotion-specific model 59 can determine a user's emotion based on an emotion graph that serves as a specific mapping (see [reference]). Figure 9 The emotion-specific model 59 can also determine the robot's emotion, and the specific processing unit 290 performs specific processing based on the robot's emotions.

[0521] 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.

[0522] 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.

[0523] 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).

[0524] 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.

[0525] 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."

[0526] 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.

[0527] 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).

[0528] 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.

[0529] 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.

[0530] 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.

[0531] 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.

[0532] 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.

[0533] 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.

[0534] 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.

[0535] 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.

[0536] 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.

[0537] 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.

[0538] In addition, the following notes are provided in response to the above explanation.

[0539] Example 1 (Note 1) An information processing system, characterized in that it comprises: A device for obtaining usage information, including user behavior history, retrieval history, and currently viewed information resources, from a user-operated terminal. An apparatus for generating characteristic information representing user interests based on the usage information and the information resources being browsed, and describing the characteristic information in natural language as prompts for use by a generative artificial intelligence model, thereby generating input instruction information; An apparatus for obtaining information held by an organization from an information management device and extracting the information held by the organization through retrieval processing; A device for inputting the prompt statement and the information held by the organization into the generative artificial intelligence model, so that the generative artificial intelligence model generates personalized information that integrates the information resources being browsed with the information held by the organization based on the user's interests; A device for sending the generated personalized information to the terminal and for the terminal to associate the personalized information with the information resource being browsed and display it. An apparatus for obtaining user evaluation information regarding the personalized information from the terminal, and updating the feature information and the generation conditions of the prompt statement based on the evaluation information.

[0540] (Note 2) The information processing system according to Appendix 1 is characterized in that, The apparatus for generating characteristic information representing user interests is configured to, when generating prompt statements representing the characteristic information, extract multiple topics and their importance information from the user's behavior history and search history, generate a data structure containing the multiple topics and the importance information, and summarize the data structure in natural language as a prompt statement for use within the generative artificial intelligence model.

[0541] (Note 3) The information processing system according to Appendix 1 is characterized in that, The system is configured to parse the content output by the generative artificial intelligence model, automatically generate secondary prompts indicating retrieval conditions and information integration policies for the information held by the organization, and control the retrieval processing of the information management device or the re-input processing of the generative artificial intelligence model based on the secondary prompts.

[0542] Application Example 1 (Note 1) An information processing system, characterized in that it comprises: A device for acquiring user information, including behavioral information and search information, from a user device and for recording the user information; An apparatus for managing an information storage device that accumulates the user information, as well as product-related information, inventory-related information, and sales promotion-related information stored in the information storage device, and for retrieving the product-related information, inventory-related information, and sales promotion-related information based on the user information; A device for integrating the user information, product information, inventory information and sales promotion information obtained through the retrieval, generating prompt statements for input to a generative artificial intelligence model, and instructing the generative artificial intelligence model to perform information generation processing; A device for parsing the generated information output by the generative artificial intelligence model and generating individual recommendation information and recommendation summary information corresponding to the product-related information and the inventory-related information; A means for sending the individual recommendation information and the summary recommendation information to the user device and for performing communication for presenting the information in real time on the user device.

[0543] (Note 2) The information processing system according to Appendix 1 is characterized in that, The device for generating prompt statements is configured to convert search terms extracted from the user information, the categories, price ranges, and preferences of the browsed information resources, as well as product-related information, inventory-related information, and sales promotion-related information, into a document in natural language form and include it in the prompt statement, thereby specifying the conditions for generating the individual recommendation information and the summary recommendation information to the generative artificial intelligence model.

[0544] (Note 3) The information processing system according to Appendix 1 is characterized in that, The device for parsing generated information and generating recommendation information is configured to extract product identification information or product names from the generated information and match them with the product-related information, convert the descriptions contained in the generated information into structured data that serves as reasons for recommending each product, and generate the structured data as a sending object for the user device.

[0545] Example 2 (Note 1) An information processing system, characterized in that it comprises: A means for acquiring behavioral information based on user operations in an information processing device, the means being configured to receive the behavioral information from a program running on an information terminal operated by the user, and to store the behavioral information in a storage device. A means for associating the behavioral information stored in the storage device with attribute information about an item or service and provision conditions information related to the item or service, based on identifiers and time information, to generate integrated data; An apparatus for constructing prompt statements for inputting into a generative artificial intelligence model based on contextual information describing the integrated data and output conditions corresponding to the integrated data, and inputting the prompt statements and the integrated data into the generative artificial intelligence model so that the generative artificial intelligence model generates useful information personalized for each user. A device for converting the useful information output from the generative artificial intelligence model into display information that can be displayed or notified in the information terminal, and for sending the display information to the information terminal via a communication network; An apparatus for updating the integrated data and the content of the prompt statement by acquiring the behavior information again through an append operation corresponding to the useful information prompted in the information terminal.

[0546] (Note 2) According to the information processing system described in Appendix 1, the information processing device is configured to, when generating the integrated data, summarize the behavioral information within a predetermined time period to extract preference information for each user, and control the content of the useful information generated by the generative artificial intelligence model by including a summary of the preference information in the prompt statement.

[0547] (Note 3) According to the information processing system described in Appendix 1, the information processing device is configured to select a set of candidate objects based on attribute information about the item or the service and the provision condition information, and to limit the output object range of the generative artificial intelligence model by including the identification information and attribute summary of the candidate object set in the prompt statement.

[0548] Application Example 2 (Note 1) An information processing system, characterized in that it comprises: An apparatus for receiving user behavior information and browsing information from a user terminal, accumulating and storing the user behavior information and browsing information in chronological order, and organizing the user behavior information and browsing information into a data structure and saving it. An apparatus for retrieving candidate information from information resources based on the accumulated user behavior information and browsing information, and for associating and integrating the candidate information with the user behavior information and the browsing information to generate a feature quantity representing the degree of user interest; An apparatus for performing sentiment analysis processing based on the user behavior information, the browsing information, and the feature quantity, thereby generating sentiment information representing the user's emotional state; An apparatus for converting the user behavior information, the browsing information, the feature quantity, and the emotional information into natural language and constructing them into prompt statements for instructing a generative artificial intelligence model to generate related information; An apparatus for inputting the prompt statement into the generative artificial intelligence model and obtaining the generation result, including related information on user interests and user emotional state and explanatory text; A device for extracting personalized information for presentation to the user from the generated results, converting it into a predetermined output format, and sending it to the user terminal in real time via a communication path; Control devices for feeding back additional user behavior information and response information from the user terminal to the accumulated stored user behavior information, and updating the generation of the prompt statement and input processing to the generative artificial intelligence model accordingly.

[0549] (Note 2) The information processing system according to Appendix 1 is characterized in that, The device for converting the user behavior information, the browsing information, the feature quantity, and the sentiment information into natural language and constructing a prompt statement for instructing a generative artificial intelligence model to generate relevant information is configured to classify the user search history and browsing history, which are included in the user behavior information, according to information categories and information content with high relevance, and embed the summary of the classification results into the prompt statement, thereby instructing the generative artificial intelligence model to generate personalized information.

[0550] (Note 3) The information processing system according to Appendix 1 is characterized in that, The apparatus for performing sentiment analysis processing based on the user behavior information, the browsing information, and the feature quantity to generate sentiment information representing the user's emotional state is configured to perform sentiment analysis processing based on text information extracted from the user behavior information and the browsing information, as well as the user's operation pattern, to obtain the user's emotional state, and to include the emotional state in the prompt statement, thereby instructing the generative artificial intelligence model to generate information corresponding to the emotional state, and selecting only the information matching the emotional state from the generated result as the personalized information.

Claims

1. An information processing system, characterized in that, include: At least one processor; The processor is configured as follows: Provide the user with an interface for receiving information from the user; The received information is analyzed using a generative artificial intelligence model. Based on prompt words that instruct the generative artificial intelligence model to predict user interests and emotions and generate highly relevant information, information that is highly relevant to the user is generated. It also communicates with the user to notify the generated information in real time.

2. The information processing system according to claim 1, characterized in that, The processor is configured to input the user's behavioral history and search history as prompts into the generative artificial intelligence model to instruct the generation of personalized information.

3. The information processing system according to claim 1, characterized in that, The processor is configured to analyze the user's emotional state and provide the obtained analysis results as prompts to the generative artificial intelligence model to instruct the generation of corresponding information.

Images