Information processing system

Abstract

The present invention provides an information processing system, including a processor configured to: generate prompts for instructing a generative artificial intelligence model to analyze a user's profile, interests, and past behavioral patterns to provide more accurate matching; generate prompts for instructing the generative artificial intelligence model to parse conversations and messages to provide more effective communication support; and generate prompts for instructing the generative artificial intelligence model to identify the user's emotions and adjust matching candidates according to the emotions.

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 matching services and online communication support systems typically rely on static user profiles or simple behavioral statistics for matching recommendations. This fails to reflect users' real and dynamically changing interests and emotional states in a timely and accurate manner, resulting in low matching accuracy and a poor user experience. Furthermore, existing systems often use fixed rules or templates for conversation analysis and response assistance, failing to fully utilize the language understanding and generation capabilities of generative AI models. Consequently, they lack personalization and flexibility in providing communication support. Moreover, current technologies generally cannot automatically identify emotions and dynamically adjust matching candidates to address real-time emotional changes and diverse questions asked by users during the matching process, nor can they generate high-quality, context-sensitive automatic responses to user questions. Therefore, it is necessary to provide a system that utilizes generative AI models to generate multi-purpose prompts, enabling the system to more precisely analyze user profiles, interests, behavioral patterns, and conversation content, and combine this with the user's emotional state for dynamic matching and communication support. This would improve matching accuracy and communication efficiency, ultimately enhancing the overall user experience. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides an information processing system comprising a processor. The processor generates prompts for invoking and controlling a generative artificial intelligence (AI) model, thereby enabling in-depth analysis and intelligent feedback of user-related information. Specifically, the processor is configured to generate prompts instructing the AI ​​model to comprehensively analyze a user's profile, interests, and past behavioral patterns, allowing the AI ​​model to model user preferences in a multi-dimensional feature space and thus provide users with more accurate matching candidates. The processor is also configured to generate prompts instructing the AI ​​model to semantically parse conversations and message content between users, enabling the AI ​​model to output communication support information that helps maintain and advance communication based on the dialogue context, topic type, and interaction intent. Furthermore, the processor is configured to generate prompts instructing the AI ​​model to identify the user's emotional state and adjust matching candidates based on that emotion, allowing the system to dynamically consider user emotional changes during the recommendation process and improve the relevance of the recommendation results. In addition, the processor can also generate prompts instructing the AI ​​model to automatically generate matching candidates based on user behavior patterns, and prompts instructing the AI ​​model to automatically generate answers to questions from users. By generating and combining the above-mentioned various prompts, the system of the present invention can efficiently schedule generative artificial intelligence models to perform unified analysis and response to user data and interaction data, thereby simultaneously improving matching accuracy, communication support quality, and user question response capabilities, effectively solving problems such as inaccurate matching, insufficient communication assistance, and low emotional response capabilities in the prior art.

[0005] "System" refers to a collection of technical devices consisting of at least one processor and cooperating memory, communication interface and / or other hardware and software components, used to perform functions such as matching recommendation, conversation parsing, sentiment recognition and response generation.

[0006] A processor is an electronic processing unit that can execute computer program instructions to perform data processing, logical operations, and flow control. It can be a single physical processor, a combination of multiple processors, or any computing unit including CPU, GPU, and NPU.

[0007] "Generative artificial intelligence model" refers to an artificial intelligence model trained based on deep learning or other machine learning techniques that can automatically generate text, images or other content based on input data. In this invention, it is mainly used to generate analysis results, suggestions or answers in natural language form.

[0008] "Hints" refer to the instructional input information generated by the processor and provided to the generative artificial intelligence model. They are usually described in natural language or structured text, describing the task to be performed, the range of input data, output format requirements and constraints, and are used to guide the generative artificial intelligence model to produce the desired output.

[0009] "User profile" refers to static or semi-static information related to a user's identity and basic characteristics, including but not limited to age, gender, occupation, location, interest tags, self-introduction, and other data that are actively provided by the user or processed by the system.

[0010] "Interests" refers to information that reflects a user's preferences in leisure activities, lifestyle, culture and entertainment, including but not limited to interest tags explicitly filled in by the user and points of interest that can be inferred from their behavior or content preferences.

[0011] "Past behavioral patterns" refer to dynamic behavioral characteristics generated by users during system use and recorded and analyzed by the system, including but not limited to browsing history, liked items, chat frequency, message content characteristics, dwell time, and interaction sequence, which are used to characterize user preferences and usage habits.

[0012] "Conversations and messages" refer to text information or text-related interaction records exchanged between users in the system, including instant chat content, historical conversation records, single messages, message sequences and their timestamps, etc.

[0013] "Communication support" refers to the auxiliary information provided by the system to users based on the analysis of conversations and messages, which helps to optimize communication effectiveness, including but not limited to reply suggestions, topic extension suggestions, polite expression references, and emotional calming expressions.

[0014] "User sentiment" refers to the emotional state or tendency inferred by the system based on the user's input text content, interaction behavior or other relevant signals, including but not limited to categories such as happy, nervous, anxious, disappointed, positive, negative, and neutral.

[0015] "Matching candidate" refers to one or more other users selected and recommended by the system from the user group to the target user. These users are identified as potential interaction objects that have a high degree of matching with the target user in terms of interests, behavioral patterns, emotional state, or other characteristics.

[0016] "User behavior patterns" refer to the comprehensive behavioral characteristics obtained by the system through statistical analysis, modeling, and abstraction of multiple user behavior records over a period of time. These characteristics reflect the stable tendencies and patterns of users in terms of object selection, communication methods, and active time periods.

[0017] "Questions from users" refers to inquiries made by users in the system in the form of natural language or structured text, requesting answers, suggestions or explanations from the system or generative artificial intelligence models, including questions about matching, communication, emotions, dating, etc. Attached Figure Description

[0018] Figure 1 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the first embodiment.

[0019] 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.

[0020] Figure 3 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the second embodiment.

[0021] 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.

[0022] Figure 5 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the third embodiment.

[0023] 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.

[0024] Figure 7 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the fourth embodiment.

[0025] 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.

[0026] Figure 9 This represents an emotion map that maps multiple emotions.

[0027] Figure 10 This represents an emotion map that maps multiple emotions.

[0028] Figure 11 This is a sequence diagram illustrating the processing flow of the data processing system of the first embodiment.

[0029] Figure 12 This is a sequence diagram illustrating the processing flow of the data processing system in Application Example 1.

[0030] Figure 13 This is a sequence diagram illustrating the processing flow of the data processing system of the second embodiment.

[0031] Figure 14 This is a sequence diagram illustrating the processing flow of the data processing system in Application Example 2. Detailed Implementation

[0032] Hereinafter, an example of an implementation of the system according to the present disclosure will be described with reference to the accompanying drawings.

[0033] First, let me explain the terminology used in the following instructions.

[0034] 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.

[0035] 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.

[0036] 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.

[0037] 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.

[0038] 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.

[0039] First Implementation Method Figure 1 An example of the configuration of the data processing system 10 according to the first embodiment is shown.

[0040] 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.

[0041] 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).

[0042] 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.

[0043] 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.

[0044] 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.

[0045] 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.

[0046] Figure 2 The diagram shows an example of the main functions of the data processing device 12 and the smart device 14.

[0047] 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.

[0048] 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).

[0049] 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.

[0050] 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.

[0051] 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."

[0052] While existing information processing systems can collect basic user attribute information and some behavioral data, the following technical problems exist: First, servers typically only perform rule matching on user attributes or simple statistical behavioral data, lacking the ability to deeply model multi-source data (attributes, preferences, behaviors, and communication content) using unified data representation and generative artificial intelligence models. This results in coarse user interest profiles, limited matching accuracy, and difficulty in implementing personalized and dynamically optimized matching logic at the computer level. Second, traditional systems mostly treat user communication content as message transmission objects, failing to extract semantic information such as emotional state and interpersonal relationship issues from the communication text into structured features on the server side in a computer-processable manner. These features cannot be fed back to the matching engine, making it difficult to adjust the content and ranking of matching candidates based on real-time emotions and interpersonal states. Third, in existing systems, recommendation algorithms and question-answering or suggestion generation functions are usually independent. The server side lacks a unified mechanism for generating and managing prompt statements, making it impossible to uniformly drive interest inference, combined candidate generation, and interpersonal relationship suggestion information generation through a single generative artificial intelligence model interface. This leads to complex system architecture, high maintenance costs, and difficulty in optimizing multiple tasks in a coordinated manner at the model level. Fourth, in many systems, users' subsequent reactions to recommendation results and suggestions are only used for simple statistics and are not effectively designed as closed-loop signals that can be continuously fed back into the model reasoning and prompt generation process. Therefore, the server cannot continuously and adaptively optimize the amount of interest features and recommendation strategies over time, thus limiting the evolutionary ability of the computer system in terms of recommendation accuracy and interactive experience.

[0053] The problem this invention aims to solve is to provide a technical solution for improving data representation and processing on the server side. By unifying and structuring user attribute information, preference information, behavioral history information, and communication content, and combining this with an automatic generation and feedback mechanism for prompt statements for generative artificial intelligence models, the server can use computer technology to: on the one hand, extract user interest and emotional features with high precision and generate matching candidates; on the other hand, dynamically adjust recommendations and interpersonal relationship suggestions based on conversation semantics and user behavior loops, thereby achieving a substantial improvement over existing computer recommendation and dialogue support technologies.

[0054] 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.

[0055] In this invention, the server includes: a device for acquiring user attribute information, preference information, and behavioral history information, and storing the information as standardized structured data; a device for automatically generating prompt statements for a generative artificial intelligence model based on the structured data, inputting the prompt statements and the structured data into the generative artificial intelligence model to perform inference processing, thereby calculating the feature quantities of user interests and preferences; a device for performing similarity calculations on the feature quantities of other objects or service objects based on the feature quantities and selecting and ranking candidate combinations; a device for acquiring communication content between users or between users and service objects, inputting the communication content and corresponding prompt statements into the generative artificial intelligence model to parse and extract topics and emotional states related to interpersonal relationships; and a device for dynamically adjusting the content or ranking of the candidate combinations based on the relationship between the topics and emotional states and the candidate combinations, generating interpersonal relationship suggestion information for users, and feeding back the user's reaction behavior to the candidate combinations and suggestion information as new behavioral history information to update at least one of the feature quantities, candidate combinations, and suggestion information. This allows for the formation of a closed-loop computing architecture within the server, centered on generative artificial intelligence models and prompts. This enables the computer system to simultaneously perform unified modeling of multi-source user data and communication semantics, achieving collaborative optimization between interest profile construction, matching candidate generation, and interpersonal relationship suggestion provision. Consequently, it significantly improves the performance of existing computer technologies in terms of algorithm accuracy, adaptability, and system scalability.

[0056] "User attribute information" refers to structured data used to represent the basic characteristics of users, including but not limited to descriptive information such as age, gender, occupation, and location, which are provided by users or inferred by the system.

[0057] "Preference information" refers to data used to represent a user's interests and preferences, including but not limited to interest tags actively selected by the user and interest categories and preference weights inferred from behavioral history information and communication content.

[0058] "Behavioral history information" refers to behavioral data generated and recorded during a user's use of the system, including but not limited to browsing operation information, evaluation operation information, like operation information, click information, participation in activities information, and their corresponding time information and object identification information.

[0059] "Standardized structured data" refers to structured data representation that is easy for computers to store, retrieve, and process, obtained by cleaning, transforming, and encoding user attribute information, preference information, behavioral history information, and communication content through predetermined data formats and field rules.

[0060] "Generative artificial intelligence models" refer to computational models built on machine learning algorithms that can automatically generate analysis results or text outputs based on input data and prompts, including but not limited to text generation models and feature inference models based on deep learning networks, attention mechanisms, or transformation structures.

[0061] "Prompt statements" refer to information that is automatically generated by the system or partially set manually, describing task requirements, constraints, and input context in natural or quasi-natural language. These statements are used as instructions or guidance to be input into generative artificial intelligence models to control the models to perform processes such as interest inference, candidate combination generation, or suggestion information generation.

[0062] "Feature quantity" refers to a numerical vector or scalar that is calculated by a generative artificial intelligence model or other computing module based on standardized structured data, and is used to represent the characteristics of a user or object in terms of interests, preferences, attributes, or emotional states.

[0063] "Combined candidates" refers to a set of candidate matching objects or candidate service schemes generated for a target user based on the similarity calculation results of user feature quantities and other object or service object feature quantities, which are used to recommend and display to the user on the terminal.

[0064] "Similarity calculation" refers to the mathematical calculation process used to measure the degree of similarity between two or more features, including but not limited to cosine similarity, Euclidean distance, Manhattan distance or other statistical similarity measures.

[0065] "Communication content" refers to the message content sent and received by users or between users and service objects through the system, including but not limited to text messages, emoticons and their metadata, and is used as object data for semantic analysis and sentiment analysis in the system.

[0066] "Interpersonal relationship issues" refer to problems, confusions, or concerns related to communication, collaboration, conflict, misunderstanding, or relationship maintenance between users, identified through semantic analysis of communication content.

[0067] "Emotional state" refers to the emotional or attitudinal characteristics of a user over a specific period of time, inferred from communication content and behavioral history information, including but not limited to positive, neutral, negative, anxious, angry, and frustrated emotional categories or their intensity.

[0068] "Suggested information" refers to user-oriented suggestions generated by a generative artificial intelligence model based on user characteristics, emotional state, and interpersonal relationship-related topics, presented in natural language. These suggestions include, but are not limited to, suggestions on communication strategies, behavioral adjustments, and relationship improvement.

[0069] "Other objects" refers to entities other than the target user that are managed by the system and can be used as matching or recommendation objects, including but not limited to other users, virtual user profiles, or abstract interest groups.

[0070] "Service object" refers to an entity provided by or associated with the system that can provide activities, content or functions to users, including but not limited to abstract representations of activity items, online services, offline services or functional modules.

[0071] "Evaluation operation information" refers to the record of operations performed by users on other objects or service objects that have a clear positive or negative meaning, including but not limited to likes, ratings, positive reviews, negative reviews, favorites or blocking, and their related metadata.

[0072] "Browsing operation information" refers to the operation records generated by users in the system when they view, stay, swipe, or enter detail pages of other objects or services. It is used to reflect the user's level of attention and interest behavior patterns.

[0073] "Output information" refers to the result data generated by the system based on feature calculation and similarity operation results, which is provided to users or other modules, including but not limited to the sorted combined candidate list, matching score and corresponding explanation.

[0074] "Dynamic adjustment" refers to the process by which the system automatically updates, reorders, or modifies existing candidate combinations, feature quantities, or suggestion information after receiving new behavioral history information, communication content, or changes in emotional state, without requiring manual intervention item by item.

[0075] "Input information for learning" refers to the data input used to train or retrain generative artificial intelligence models, including but not limited to the latest user behavior history, feedback information, and annotation information, which are used to update model parameters to improve model performance.

[0076] "Input information for reasoning" refers to the data input into a generative artificial intelligence model after the model has been trained to obtain a specific output result, including but not limited to the current user's structured data, prompts, and related contextual information.

[0077] In one embodiment of the present invention, a server, a terminal, and a user collaboratively constitute an overall system. The server serves as the core of data processing, the terminal serves as the human-computer interaction and data acquisition interface, and the user interacts with the server through the terminal.

[0078] In a preferred embodiment, the server is implemented using a hardware platform that includes a multi-core central processing unit and a graphics processing unit, such as a multi-core general-purpose processor, a graphics processing unit with parallel computing capabilities, a large-capacity random access memory, and a solid-state storage device. The server runs a general-purpose operating system (e.g., a UNIX-like operating system) and deploys a database management system, a web server program, and a machine learning framework. The database management system can be a relational database management system, the web server program can be a combination of a reverse proxy and an application server, and the machine learning framework can include a framework that supports building deep neural networks, such as using an open-source framework capable of building and inferring neural networks.

[0079] In one embodiment, the terminal is a mobile computing device, such as a smart mobile terminal or a portable information processing device. The terminal is equipped with a mobile operating system and runs dedicated applications of this system. The terminal uses input / output components such as a touch screen, keyboard, and speaker to collect user input and display data returned by the server. The terminal sends and receives data with the server through a secure communication protocol.

[0080] In one implementation, the server uses a database to manage user attribute information, preference information, behavioral history information, and communication content in a unified manner. The server converts user attribute information, preference information, and behavioral history information into a standardized structured data format. The server can use a data analysis library to clean, normalize, and encode the raw data in memory, converting text fields into discrete identifiers, mapping tag information into multidimensional sparse vectors, and performing range scaling or segmented encoding on numerical features.

[0081] In one implementation, the server uses a generative artificial intelligence model as its core analysis module. The server constructs a multi-layered neural network using a machine learning framework. This neural network may include an embedding layer, several fully connected layers, and a transformation structure. The embedding layer maps discrete identifiers (e.g., interest tags, occupation categories) into low-dimensional dense vectors; the transformation structure models multi-source user features using a multi-head attention mechanism; and the fully connected layers jointly map different feature subspaces. The server trains the model using a loss function (e.g., cross-entropy loss or mean squared error loss) and updates the network weights using a gradient descent-based optimization method. The server can employ batch training, performing forward computation and backpropagation on multiple user samples simultaneously, thereby improving training efficiency.

[0082] In one implementation, the server constructs prompts for the generative artificial intelligence model. The server preprocesses user attribute information, preference information, behavioral history information, and communication content, embedding them into natural language text so that the prompts simultaneously include task description and input context information. For example, the server can generate prompts for preference analysis and matching recommendations as follows: "Based on the user's attribute information, interest tags, and behavior records over the past 30 days, please analyze the user's interests and hobbies, and provide 3 to 5 interest keywords. Also, recommend several highly relevant candidate combinations for the user:" Attribute information: Age=29, Gender=Female, Occupation=Engineering professional, City=A certain city; Interest tags: outdoor activities, camping, hiking; Behavioral history: Liked 30 items tagged "Outdoor Activities" and 3 items tagged "Indoor Recreation" in the past 30 days. The server can also generate helpful prompts for interpersonal relationships, such as: "The following is a description of a problem a user encountered in work communication: 'Recently, I've been having trouble communicating with my colleagues. They always misunderstand me, which makes me confused and a bit frustrated.'" Please analyze the main problems this user may have in interpersonal communication, and provide 3 to 5 specific, actionable communication suggestions. Use gentle, empathetic language and avoid a commanding tone. In one implementation, the server simultaneously inputs the aforementioned prompt statements and structured vector features into a generative artificial intelligence model. The server sets up a text encoding module at the model's front end to tokenize and embed the prompt statements. In another branch of the model, a numerical feature encoding module performs linear transformations on the user attribute vectors, preference vectors, and behavior vectors. Then, a feature fusion layer concatenates or weights the text representation and the numerical feature representation. The server performs several layers of transformation on the fused representation, enabling the model to simultaneously capture natural language semantics and behavioral pattern features in a unified representation space.

[0083] In one implementation, the server generates interest and sentiment features based on the model output. The server normalizes the continuous vectors output by the model, interpreting the output of the interest dimension as a probability distribution or weight of different interest categories, and the output of the sentiment-related dimension as the confidence level of different emotion categories. The server stores these features in a user profile data table, providing fast access for subsequent similarity calculations and ranking.

[0084] In one implementation, the server performs similarity calculations on the combined candidates. The server pre-constructs a corresponding feature vector for each candidate, including attribute features, interest features, and behavioral features. The server calculates the similarity value between the target user and each candidate using vector dot product, cosine similarity, or a learned metric function (e.g., a distance function parameterized using a small neural network). The server employs vectorized operations and leverages the parallel capabilities of the graphics processing unit during the calculation process, thereby reducing computation time. The server sorts the candidates based on the similarity values ​​and may apply threshold filtering, geographical constraints, or other rules to ensure that the final candidate set satisfies both interest matching and geographical location or other conditions.

[0085] In one implementation, the server performs semantic and sentiment analysis on the communication content. The server extracts conversation fragments from a message database, processes the text using word segmentation and sub-word tokenization methods, and inputs the word sequences into a text encoding module. The server captures keywords and contextual relationships within sentences using an attention mechanism, assigning higher attention weights to emotion-related words (e.g., words expressing dissatisfaction, confusion, or stress). The server sets up classification units at the output layer to perform multi-label classification of emotion categories and interpersonal relationship topic categories, thereby obtaining structured representations of "interpersonal relationship related topics" and "emotional states." In this way, the server transforms free text, originally intended for human reading, into feature data that can be directly used by the algorithm, reducing reliance on manual annotation and rule design.

[0086] In one implementation, the server dynamically adjusts the candidate combinations based on extracted topics and emotional states. The server configures multiple scoring components for the recommendation ranking module, including interest matching score, behavioral similarity score, and emotional fit score. The emotional fit score can be calculated based on statistical results of the user's current emotion and past interactions with different types of objects under similar emotional states. When synthesizing the total score, the server applies learnable weight coefficients to each component and adjusts them based on long-term user feedback. Thus, the server considers not only the interest dimension but also emotional and interpersonal needs, making the recommendation results more consistent with the user's receiving tendencies in the current context, thereby reducing invalid communication and inappropriate matching.

[0087] In one implementation, the server generates suggested information using a generative artificial intelligence model and prompts. The server explicitly indicates the generation goal, tone requirements, and security constraints in the prompts. Upon receiving the prompts, the model generates text containing multiple suggested items. The server performs rule-based checks and sensitive content filtering on the generated results, removing segments that do not comply with security policies or exceed a predetermined length, and retaining paragraphs with clear structure and complete semantics. The server returns the final suggested information and combined candidate suggestions as output to the terminal.

[0088] In one implementation, the terminal is responsible for displaying the recommendation results and suggestions. After receiving data from the server, the terminal displays combined candidates in a list or card format on the user interface and displays interpersonal relationship suggestions in a text area. The terminal receives user actions such as selecting, liking, or ignoring combined candidates, as well as reading the suggestions, such as whether to expand the reading or send a new message based on the suggestions. The terminal uploads these actions as behavioral history information to the server, which incorporates this behavioral history information into subsequent feature updates and model inference, thus achieving a feedback loop.

[0089] In one implementation, the server updates the prompts and feature values ​​based on new behavioral history information. When the server detects a user's sustained positive response to a certain type of recommended object, it increases the descriptive weight of that interest dimension in the prompt and adds the corresponding dimension's value to the feature vector; conversely, when the server detects a user's prolonged indifference or repeated negative feedback regarding a certain type of recommended object, it appropriately decreases the weight of that dimension. This method of automatically adjusting the prompt content and feature values ​​based on user behavior ensures that the generative AI model consistently receives input that more closely reflects the user's current state during the inference phase, thereby improving recommendation accuracy and reducing the proportion of irrelevant results.

[0090] In one implementation, the server employs a multi-task learning strategy when training the generative AI model. The server uses a unified neural network architecture to simultaneously handle tasks such as interest inference, combinatorial recommendation, and interpersonal suggestion generation. The server sets different output heads for different tasks and designs corresponding loss functions for each task. For example, cross-entropy loss is used for interest classification, mean squared error loss for matching score regression, and sequence log-likelihood-based loss for text generation. The server combines the losses of each task through a weighted summation, updating the parameters of the shared layer and each output head for all tasks in a single backpropagation. Through this multi-task joint training, the server enables the model to learn abstract features useful for multiple tasks in the shared representation layer, thereby reducing redundant computations during inference, improving overall computational efficiency, and reducing model size.

[0091] In one implementation, the server employs a specific design for its data flow and module structure to improve computational efficiency and reduce communication load. When receiving data uploaded by the terminal, the server caches the raw data in layers, storing frequently updated historical behavioral information in a high-speed cache and infrequently updated attribute information in a backend database. When performing recommendations, the server only needs to read recent behavioral data from the cache and load pre-defined interest features from the database, avoiding the need to repeatedly parse all the raw data. When communicating with the terminal, the server retains complex internal feature vectors and intermediate calculation results internally, sending only the ranking results, simplified scoring information, and natural language suggestion text to the terminal, thereby significantly reducing the amount of data transmitted over the network and minimizing latency.

[0092] In one embodiment, the server improves upon traditional computer technology through the aforementioned technical features. By employing specific data structure design, feature fusion methods, and multi-task training mechanisms, the server enhances the inference efficiency and accuracy of generative artificial intelligence models on multi-source heterogeneous data. By automatically generating and dynamically adjusting prompts, the server ensures the model receives more targeted contextual instructions in each inference iteration, thereby shortening convergence time and reducing the model's reliance on manual rules. Furthermore, by incorporating semantic and sentimental states of communication content into recommendation ranking calculations, the server moves beyond the traditional matching model based solely on static attributes and simple statistical behavior, achieving more refined and dynamic matching decisions. These improvements go beyond simply automating manual matching; they implement new data representation methods, new model invocation processes, and new feedback learning mechanisms within the computer, resulting in significant improvements in processing speed, recommendation accuracy, resource utilization, and communication load.

[0093] In another implementation, the server can use different neural network architectures to implement generative artificial intelligence models. For example, the server can use gated recurrent units or long short-term memory networks as text encoding modules, or it can use graph neural networks to model the interaction between users and objects, and fuse the graph structure embedding results with the output of the transformed structure. The server can also employ different optimization algorithms, such as adaptive moment estimation or other adaptive learning rate methods, to accelerate the model training convergence speed.

[0094] In another implementation, the server can modify its feature selection strategy based on the application scenario. For scenarios prioritizing real-time performance, the server can place greater emphasis on recent behavior vectors and the latest communication content, using a decay factor to weight historical long-term data. For scenarios emphasizing long-term stable preferences, the server can increase the weights of long-term interest statistical features and cumulative interaction features to improve prediction stability. Through these flexible adjustments to parameters and feature strategies, the server can maintain high processing efficiency and recommendation quality under varying hardware resources and business requirements.

[0095] Through the aforementioned implementation methods, the server, terminal, and user form a closed-loop technical structure within the system, centered around a generative artificial intelligence model and prompts. The server not only acquires, analyzes, and stores user data, but also, through specific feature construction, model architecture, and prompt design, achieves comprehensive optimization of recommendation accuracy, processing speed, resource utilization, and communication load. This significantly enhances the computer system's ability to handle complex matching and interpersonal relationship assistance tasks from a technical perspective.

[0096] use Figure 11 The processing flow is explained.

[0097] Step 1: Users input basic information using the terminal.

[0098] Users open the application on their devices, enter their age, gender, occupation, location, interest tags, and other attribute and preference information, and then click to confirm and submit.

[0099] Input: Text and option data (age, gender, occupation, city, interest tags, etc.) entered by the user in the terminal interface.

[0100] Terminal processing and data processing: The terminal converts each field into a predefined key-value pair format, performs type checks on numeric fields, converts multi-select labels into a label list, filters special characters, and then encapsulates all data into structured data (such as key-value mapping).

[0101] Output: The terminal generates a structured set of user attribute and preference information and prepares to send it to the server via the network interface.

[0102] Step 2: The terminal sends user attribute and preference data to the server.

[0103] After the terminal establishes a connection with the server through a secure communication channel, it sends the structured data encapsulated in step 1 as the request content to the server.

[0104] Input: Structured user attribute information and preference information stored internally in the terminal.

[0105] Terminal processing and data processing: The terminal adds metadata such as user identifier and timestamp to the request, serializes the data into a unified transmission format, adds necessary authentication tokens, and then sends it to the specified interface of the server through the communication module.

[0106] Output: The terminal output is a network request message containing user attribute information, preference information, and metadata. This message is received by the server and enters the subsequent processing flow.

[0107] Step 3: The server receives and stores user attribute and preference data.

[0108] After receiving a request message from the terminal, the server parses and verifies the data and writes it into the database.

[0109] Input: Structured user attribute information and preference information messages from the terminal.

[0110] Server processing and data manipulation: The server uses the network service module to parse the message content, extracting user identifiers, attribute fields, and interest tags. The server performs validity checks on the fields (age range checks, gender enumeration checks, tag existence in the tag table, etc.) and performs encoding standardization processing on the text content. The server maps the validated data to database table fields, converts interest tags into an internal tag ID list, and generates standardized structured data records.

[0111] Output: The server generates or updates a record of user attribute information and preference information corresponding to the user in the database, and returns status information for later use.

[0112] Step 4: The terminal records the user's behavior history (browsing, liking, etc.).

[0113] Users browse recommendation lists or search results on the terminal and perform actions such as browsing, liking, and saving certain objects.

[0114] Input: User clicks, swipes, and page dwells on the terminal interface.

[0115] Terminal processing and data processing: The terminal converts each browsing or liking behavior into a behavior event record, which includes user identifier, target object identifier, behavior type and timestamp; the terminal merges consecutive identical behaviors, filters out invalid browsing events with very short stays, and packages the behavior events into behavior history data.

[0116] Output: The terminal generates a list of historical behavioral events and sends it to the server periodically or when a certain number are reached as input.

[0117] Step 5: The server receives behavioral history and updates the behavioral history information database.

[0118] After receiving the list of historical behavior events uploaded by the terminal, the server parses it and writes it into the behavior history information database.

[0119] Input: Historical event data from the terminal (browsing operation information, evaluation operation information, and time information, etc.).

[0120] Server processing and data manipulation: The server parses and standardizes each event record, mapping behavior types to internal codes (such as browsing, liking, and saving), converting timestamps to a unified time format, and inserting new records or updating aggregated statistical data (such as the cumulative number of views for the same object) in the behavior history table based on user and object identifiers. The server can also perform sliding window statistics on recent behavior and update recent user behavior characteristics.

[0121] Output: The server generates standardized behavior records in the behavior history information database, and at the same time generates updated behavior statistical feature data to provide input for subsequent analysis.

[0122] Step 6: Users communicate and send messages on the terminal.

[0123] Users open a dialog box in the terminal to communicate with other objects or services, enter text messages, and send them.

[0124] Input: Natural language message text entered by the user in the terminal text input box.

[0125] Terminal processing and data processing: After the user clicks send, the terminal encapsulates the message content along with the sender identifier, receiver identifier, session identifier, and timestamp into a message record; to ensure data integrity, the terminal performs unified encoding and length limitation processing on the message text.

[0126] Output: The terminal outputs a structured message record and sends it to the server for storage and subsequent analysis.

[0127] Step 7: The server receives communication content and stores message data.

[0128] The server receives communication message records from the terminal and stores them in a message database for analysis.

[0129] Input: Structured message records uploaded by the terminal (text message, sender identifier, receiver identifier, timestamp, etc.).

[0130] Server processing and data processing: The server parses message data, pre-cleans the text content (removes control characters and standardizes emoticon text representations), and adds session context identifiers for subsequent natural language processing; the server stores the cleaned messages in a message table and assigns a unique message ID to each message.

[0131] Output: The server generates a series of cleaned communication content records in the message database, which constitute the raw data source for subsequent semantic and sentiment analysis.

[0132] Step 8: The server constructs comprehensive user feature input.

[0133] The server aggregates information from multiple data sources, including user attributes, preferences, behavioral history, and communication content, to generate a feature set suitable for input generative artificial intelligence models.

[0134] Input: Summary data of user attribute information records, preference information records, behavior history information records, and recent communication content in the database.

[0135] Server Processing and Data Manipulation: The server uses a data processing module to perform join queries on multiple data sources, merging records from different tables according to user identifiers. The server performs category encoding on attribute information, generates multi-dimensional binary vectors for interest tags, and calculates statistical features for behavioral data (such as the number of likes for a specific tag in the last 30 days). It also performs word segmentation or sub-word splitting on communication text, converting important keywords into word vectors. Finally, the server assembles this information into a unified set of feature vectors and text fragments to drive inference in subsequent generative artificial intelligence models.

[0136] Output: The server generates a comprehensive feature set containing numerical feature vectors and text context information, which serves as the basis for subsequent prompt generation and model input.

[0137] Step 9: The server generates prompts for the generative artificial intelligence model.

[0138] The server converts comprehensive features into prompts in natural language based on the current task (such as interest analysis, combined candidate recommendations, and interpersonal relationship suggestions).

[0139] Input: The set of user features (including attributes, interest tags, behavioral statistics, and communication summaries) that the server constructs in step 8.

[0140] Server processing and data manipulation: The server uses a template system or rule engine to insert numerical features and labels into predefined natural language sentences to form clear and semantically complete prompts. The server specifies the type and style requirements of the output to be generated in the prompts according to different tasks (e.g., "list 3 to 5 keywords of interest" or "give specific and actionable suggestions").

[0141] Output: The server generates prompt text for generative artificial intelligence models, for example: "Based on the user's attribute information, interest tags, and behavior records over the past 30 days, please analyze the user's interests and hobbies, and provide 3 to 5 interest keywords. Also, recommend several highly relevant candidate combinations for the user:..." as well as "The following is a description of a problem a user encountered in work communication: 'Recently, I have been having trouble communicating with my colleagues. They always misunderstand me. I am confused and a little frustrated.' Please analyze the main problems this user may have in interpersonal communication and provide 3 to 5 specific and actionable communication suggestions. Use gentle and empathetic language and avoid a commanding tone." Step 10: The server calls a generative artificial intelligence model to perform inference.

[0142] The server inputs the prompts and comprehensive features into the generative artificial intelligence model, and performs inference calculations to obtain interest feature quantities, sentiment feature quantities, and candidate suggestion content.

[0143] Input: The prompt text generated by the server and the numerical feature vector from step 8.

[0144] Server processing and data manipulation: The server first tokenizes the prompt statements and maps them to word vector sequences through an embedding layer. Simultaneously, it inputs the numerical feature vectors into a numerical feature encoding layer, which performs a linear transformation to obtain a representation compatible with the word vector dimensions. In the feature fusion layer, the server concatenates or weights the representations from the two branches, and then performs multi-head attention operations and feedforward network operations on the merged vector through a multi-layer transformation structure. In the output layer, the server sets different headers for different tasks: the interest prediction header outputs the probability distribution and feature vectors of interest categories, the sentiment analysis header outputs the probability distribution of sentiment categories, and the text generation header outputs a text sequence of recommendation reasons or suggestions.

[0145] Output: The server obtains the user's interest and preference features, sentiment state features, matching scores related to the recommended candidates, and candidate descriptions or suggestions in natural language.

[0146] Step 11: The server calculates the similarity between the user and the candidate and generates combined candidates.

[0147] The server uses the features output by the generative artificial intelligence model to calculate the similarity between the user and each candidate object and generate a sorted list of combined candidate objects.

[0148] Input: User interest features obtained from step 10, and feature vectors of candidate objects or service objects pre-stored in the database.

[0149] Server processing and data manipulation: For each candidate object, the server calls the similarity calculation module to calculate the distance or similarity value between the user's feature vector and the candidate's feature vector using cosine similarity or other metrics. The server combines interest similarity, behavioral similarity, and sentiment fit to generate a comprehensive score. To improve efficiency, the server uses vectorization and batch processing to calculate the similarity of all candidate objects in parallel on the processor or graphics processing unit.

[0150] Output: The server generates a candidate list of combinations ranked from highest to lowest based on the overall score. Each candidate includes a matching score, a summary of the recommendation reason, and basic descriptive information.

[0151] Step 12: The server analyzes the communication content and generates interpersonal relationship suggestions.

[0152] Based on the emotional feature data and interpersonal relationship-related topic information obtained in step 10, the server constructs specific suggestion text for the problem.

[0153] Input: Emotional state features, topic category labels, and intermediate semantic representations output by the generative artificial intelligence model.

[0154] Server processing and data manipulation: The server re-integrates these features into new prompts, requiring the model to generate suggestions with a specified tone and structure; the server calls the text generation header to instruct the model to output multiple suggestions. The server then uses rule filtering and security policies to check and remove non-compliant content, and segments the suggestion text by item.

[0155] Output: The server generates structured interpersonal relationship suggestions, including suggestion titles and several executable suggestion texts.

[0156] Step 13: The server returns recommendation results and suggestions to the terminal.

[0157] The server combines the candidate list and suggestion information into a single response data package and sends it to the terminal.

[0158] Input: The list of candidate combinations from step 11 and the suggestion information from step 12.

[0159] Server processing and data manipulation: The server simplifies the list content, retaining key information (such as avatar address, nickname, match percentage, and main recommendation reasons), and organizes the suggestion information into paragraph structures that can be directly displayed on the terminal interface. The server serializes and compresses the response data to reduce network transmission volume.

[0160] Output: The server outputs a response message containing recommended combination candidates and interpersonal suggestions, which is received by the terminal for front-end display.

[0161] Step 14: The terminal displays the recommendation results and suggestions.

[0162] After receiving the server's response, the terminal displays the combined candidate and suggested content on the user interface.

[0163] Input: A list of candidate combinations and relationship suggestions returned by the server.

[0164] Terminal processing and data processing: The terminal parses the response message, maps the candidate object information to interface components (such as list items and cards), and presents the matching degree in numerical or graphical form; the terminal displays the suggestion title and multiple suggestion contents in the text area, and automatically wraps and paginates according to the screen size and language length.

[0165] Output: The terminal output is a visual interface, where users can see the matched recommended objects and corresponding suggestions on the screen, thus enabling them to perform subsequent interactive operations.

[0166] Step 15: Users provide feedback on recommendations and suggestions.

[0167] Users can perform new actions based on the recommendations and suggestions on the terminal interface, such as liking a candidate again, sending a message, or ignoring the recommendation. They may also adjust their reply content based on the suggestions.

[0168] Input: Recommended candidate information and suggestions displayed on the terminal, as well as user clicks, inputs, and dwell time on the interface.

[0169] Terminal processing and data processing: The terminal captures the user's specific actions, encodes them into new behavioral events (such as "clicking on a recommended candidate", "sending a message according to the suggestion", "ignoring a candidate of a certain category", etc.), and encapsulates them together with timestamps and object identifiers into new behavioral history records.

[0170] Output: The terminal generates new historical behavioral event data and uploads it to the server as input for the next round of feature updates and model inference.

[0171] Step 16: The server updates features and prompts based on feedback behavior.

[0172] After receiving new user behavior history, the server updates the user's interest features, behavioral statistics, and subsequent prompts.

[0173] Input: New behavioral historical event data uploaded by the terminal and existing user characteristic records in the server.

[0174] Server processing and data manipulation: The server performs time-weighted accumulation of new behaviors, giving higher weight to recent behaviors; the server recalculates the weights of interest dimensions and adjusts the intensity of interest in certain tags; when generating the next round of prompts, the server re-describes the user's interests and behavioral tendencies based on the updated statistical results, such as increasing the number of mentions of high-frequency tags or highlighting the latest behavioral changes.

[0175] Output: The server receives updated user characteristics, behavior statistics, and new prompt templates. These outputs are used in subsequent steps to call the generative artificial intelligence model again, thus forming a continuously optimized closed-loop processing flow.

[0176] 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".

[0177] As generative AI models are increasingly used in recommendation and shopping guide scenarios, the common approach in existing technologies is for the server to directly input basic user information or simple queries into the generative AI model, which then generates recommended content or response text. This approach presents several technical problems. First, the server does not perform structured integration of user attribute information, historical behavior records, and location information in the physical space. This results in the generative AI model's input lacking refined contextual features, leading to a low degree of matching between the generated results and the current scenario and the user's actual preferences. Second, the server typically treats the generative AI model as a "black box text generator," lacking a collaborative control mechanism between the recommendation model and the generative AI model. It cannot accurately instruct the generative AI model's learning or inference goals through programmatic prompts, making it difficult for the system to stably reproduce and optimize for specific business logic. Third, the server only records user interactions on the terminal without linking these interactions to regional information in the current physical space for dynamically updating the learning data of the recommendation and generative AI models. This makes it difficult to reflect changes in user interests in a timely manner, hindering long-term adaptive optimization of recommendation effectiveness and communication experience. Fourth, the common server-side architecture separates the recommendation logic and natural language generation logic, lacking a unified mechanism for generating and managing prompts. This results in complex interfaces between different modules and fragmented data flow, which is not conducive to steadily improving processing efficiency and recommendation accuracy in high-concurrency environments.

[0178] Therefore, there is a need for a system that improves the data processing flow and model calling method on the server side, achieves efficient integration of user attribute information, historical behavior records and entity spatial location information within the computer, and precisely controls the collaborative work of generative artificial intelligence models and recommendation models by automatically generating structured prompts. This system aims to improve recommendation accuracy and user experience while optimizing the server-side computing process and storage structure, thereby improving the overall technical performance of the computer system.

[0179] 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.

[0180] In this invention, the server includes a computing unit communicatively connected to an information storage unit. The computing unit is configured to: retrieve user attribute information and past behavior records from the information storage unit based on a user identifier; process the information into feature data; and generate prompts to instruct a generative artificial intelligence model to perform learning or inference. It also receives location information calculated by a terminal device based on signals from a location information acquisition device in physical space; maps the location information to region information; and generates prompts to instruct the region information to be included as input in the generative artificial intelligence model. Furthermore, based on user attribute information, past behavior records, and region information, it performs judgment processing on a set of candidate items or services to generate a recommendation model. The system processes user attributes, historical behavior, and entity spatial location information into structured context features within the server using a unified data structure and control logic. The programmatically generated prompts precisely drive the collaborative reasoning and adaptive learning of the recommendation model and the generative AI model. This improves the relevance and personalization of the recommendation results with the natural language proposal content, reduces invalid data interaction between the server and model services, optimizes memory access and computational load distribution, and ultimately improves the overall processing performance and stability of the computer system in complex recommendation scenarios. The system also acquires user selection and attribute information from the candidate items or services selected by the recommendation model as explanatory data, inputs these information into the generative AI model to generate prompts in natural language to inform the user to generate user-oriented proposal content, and sends the generated proposal content as notification information to the terminal device. Furthermore, the system obtains user selection and operation information on the terminal device in response to the notification information, appends this selection and operation information as user behavior records to the information storage unit, and generates prompts instructing the user to utilize this selection and operation information as learning data for both the recommendation model and the generative AI model.

[0181] A "system" refers to a combination of devices that include one or more information processing components and are used to perform data storage, communication, computation and output processing. It can be distributed across multiple computing devices.

[0182] A "server" refers to an information processing device in a system that undertakes centralized data processing, model invocation, and result distribution functions. It can be a single computing device or a cluster of multiple computing devices.

[0183] "Computational unit" refers to the collection of processors and the program logic running on them in a server that perform data reading and writing, feature extraction, model calling, prompt statement generation, and result post-processing.

[0184] "Information storage unit" refers to a storage device used to store user attribute information, user behavior records, product or service information, location information, and model-related data, which can be accessed by the server through read and write operations.

[0185] "User identification information" refers to data used to uniquely identify a user in the system, such as user ID, account identifier, or other identification data that can be uniquely mapped to a user.

[0186] "User attribute information" refers to structured data related to a user's basic characteristics, including but not limited to age, gender, occupation, hobbies, budget preferences, etc.

[0187] "Past behavior records" refer to data records generated by users in the system in the past related to browsing, clicking, collecting, purchasing, rating or other interactions, which are used to reflect the user's behavior patterns.

[0188] "Feature data" refers to numerical or structured data that has been processed by cleaning, transformation, and encoding of raw data and is suitable as input to a model.

[0189] "Physical space" refers to the physical environment in which users conduct on-site activities, such as offline stores, showrooms, service venues, and other areas with actual spatial layouts.

[0190] "Location information acquisition device" refers to a hardware device arranged in a physical space to provide location-related information to terminal devices via radio wave signals or wireless signals, including but not limited to beacon devices and wireless access devices.

[0191] "Terminal device" refers to a computing device carried or used by a user for data interaction with a server and for displaying content to the user, including but not limited to mobile communication terminals, tablet terminals, etc.

[0192] "Location information" refers to data calculated based on radio or wireless signals received by a terminal device, used to indicate the user's location in physical space.

[0193] "Regional information" refers to the identification data used to represent the spatial region to which a user belongs, obtained by mapping location information to predefined regions in physical space.

[0194] "Candidate items or services" refers to a collection of goods or services that are considered available for recommendation or proposal at a specific time.

[0195] "Recommendation model" refers to an algorithm model that scores, ranks, or filters candidate items or services based on user attribute information, past behavior records, and regional information to determine the recommendation target.

[0196] "Processing conditions" refers to the parameters, weights, thresholds, rules, or other control information used by the recommendation model when performing candidate selection processing.

[0197] "Identification information" refers to identification data used to uniquely identify an item or service, such as an object number or code.

[0198] "Attribute information" refers to descriptive data related to an item or service, including but not limited to category, price, brand, and functional characteristics.

[0199] "Explanatory data" refers to a set of structured or semi-structured data used to provide generative artificial intelligence models with information about items or services that are candidates.

[0200] "Proposal content" refers to the user-oriented natural language recommendation text and its structured representation, generated by a generative artificial intelligence model based on the items or services of the candidate.

[0201] "Notification information" refers to message data sent by the server to the terminal device to prompt or display the content of the proposal, including but not limited to title, summary, link and related parameters.

[0202] "Selection operation information" refers to the interactive data generated when users perform actions such as clicking, confirming, rejecting, adding to favorites, or purchasing on notification information or proposal content on their terminal devices.

[0203] "Generative AI models" refer to AI models that can automatically generate text or other content based on input prompts. They typically employ deep learning structures and have generative capabilities.

[0204] "Prompt statements" refer to text or instruction data generated by the server and input into generative artificial intelligence models or other models to instruct the models to perform specific learning, reasoning, or generation tasks.

[0205] "Learning data" refers to the data set used to train, fine-tune, or update the parameters of recommendation models and generative artificial intelligence models, including user behavior records, attribute information, and region information.

[0206] "Weighted conditions" refers to the weighting of different features or scores to indicate their relative importance during the selection or ranking of candidates.

[0207] "Priority conditions" refer to the sorting rules or priority settings that determine the execution or recommendation order among multiple candidate objects or multiple rules.

[0208] "Learning conditions" refers to the settings of the data range, sample weights, objective function, or other training parameters used during the training process of a generative artificial intelligence model or recommendation model.

[0209] "User-input text" refers to natural language text content entered by users through terminal devices, including but not limited to questions, descriptions of needs, and feedback information.

[0210] "Response text" refers to the natural language response content generated by generative artificial intelligence models based on user input text and related context, used to reply to users.

[0211] In one embodiment of the present invention, the server, as the core information processing device, is deployed in a computing environment with centralized management of computing resources. The server can be constructed using general-purpose computer hardware, including, for example, a multi-core central processing unit, an optional graphics processing unit, main memory, and external memory. The server runs middleware and applications on an operating system, and the applications implement functions such as user data management, feature processing, recommendation model inference, generative artificial intelligence model invocation, and prompt statement generation.

[0212] The server's software architecture includes: a data access module, a feature processing module, a recommendation model inference module, a generative artificial intelligence model interface module, a prompt generation module, a result post-processing module, and a log and learning data management module. The server communicates bidirectionally with the terminal and the information storage unit via a network interface. The information storage unit can be implemented using a relational data management system.

[0213] At the data storage level, the server stores user attribute information, past user behavior records, product or service information, and physical spatial region information in multiple logical data tables. User attribute information may include, but is not limited to, fields such as user ID, age group, gender category, interest tag set, and budget range code; past behavior records may include fields such as timestamp, product category ID, operation type code, terminal identifier, and the region identifier at the time. The server uses an index structure when accessing these tables to accelerate queries, thereby reducing disk access latency during high concurrency of recommendation requests.

[0214] The terminal, acting as a user-side device, can be a mobile communication terminal or a portable terminal. The terminal runs a client application on an operating system. This client application is configured to collect location-related signals, present proposal content sent by the server, and collect user interaction actions. The terminal receives wireless signals from location information acquisition devices deployed in the physical space via a Bluetooth module. It calculates or participates in the calculation of the user's location information in the physical space through the operating system's location service interface and sends this location information, along with the user's identifier, to the server. The terminal also displays the proposal content generated by the server through a graphical user interface and reports user clicks, selections, and other actions to the server as event data.

[0215] When using this system, users input their personal attributes and preferences through the terminal. Users also move naturally within the physical space, browsing, selecting, and purchasing. The terminal provides this behavioral data to the server as one source of past behavior records. Users do not need to understand the internal model structure to receive personalized proposals, but their behavioral data is continuously transformed into input for model training and inference within the server.

[0216] In terms of feature processing, the server first uses the data access module to read the original user attribute fields, behavior log fields, and regional information fields from the information storage unit. In the feature processing module, the server uses a data analysis library to clean, encode, and aggregate the raw data. The server can convert categorical fields into discrete indexes and then map them into low-dimensional dense vectors through an embedding matrix; the server can normalize or bucket encode numerical fields to concatenate them with other features into feature vectors of uniform length. The server can also construct derived features based on the time dimension, such as the interaction frequency and average unit price of a certain type of item in a recent period. This structured feature processing enables the recommendation model to efficiently compute in a vectorized manner internally, thereby reducing unnecessary data format conversions and memory copies at the computer level and improving overall processing speed.

[0217] The server uses a machine learning framework to load a pre-trained recommendation model in the recommendation model inference module. This model can employ a deep neural network structure containing embedding layers, fully connected layers, and activation function layers. During forward inference, the server concatenates user feature vectors, item or service feature vectors, and region context feature vectors, inputs them into the deep neural network, and calculates a one-dimensional score through multiple linear transformations and nonlinear functions. The server performs this calculation in batches for multiple candidate objects and sorts or filters them based on the scores. Through this vectorized batch processing method, the server can fully utilize the parallel computing power of the processor, thereby achieving high-throughput recommendation computation, which is superior to the traditional method of judging each rule list one by one.

[0218] In the generative AI model interface module, the server interacts with the generative AI model using text encoding and decoding mechanisms. The generative AI model can employ a multi-layer neural network architecture based on a self-attention mechanism. Instead of directly inputting fragmented fields into the generative AI model, the server combines user profile summaries, candidate object key points, and regional context into structured natural language prompts in the prompt generation module. These prompts explicitly indicate the task boundaries and output style of the generative AI model, allowing the model to assign higher weights to key fields during internal attention calculations, thus improving the stability and controllability of the generated results.

[0219] In generating prompts, the server uses programmatic rules or template mechanisms to convert internal structured data into text. The server can select different templates based on different scenarios, such as for single-item recommendations, set pairings, or budget constraints. When filling in the templates, the server uses deterministic string concatenation and formatting operations to ensure the prompts are syntactically stable, facilitating learning and generalization by generative AI models. This prompt generation process is essentially the server's constraint and reconstruction of the model's input space. From a computer system perspective, this can be viewed as engineering control of the model's input distribution at the application layer to reduce the variance and anomaly probability of the generated results.

[0220] In a specific example, the server can construct the following prompt statement: User profile: Male, 32 years old, mainly purchased running shoes and running shorts in the past three months, with an average price of about 600 yuan per item.

[0221] The current user is located in the athletic shoe section of the physical store.

[0222] Candidate Products: 1. Brand A cushioned running shoes, priced at 650 yuan, suitable for long-distance jogging; 2. Brand B lightweight racing running shoes, priced at 580 yuan, suitable for speed training; 3. C brand stable support running shoes, priced at 620 yuan, suitable for people with low arches.

[0223] Please use Simplified Chinese to generate a personalized recommendation message for this user, explaining the order of priority and why each shoe is suitable for them. Use a natural and concise tone. The server sends the aforementioned prompt as a single text input to the generative AI model. Internally, the model performs word segmentation, embedding mapping, and multi-layer attention operations on the text, outputting a natural language proposal that matches the candidate products and user profile. In the post-processing module, the server can truncate the generated text, filter sensitive content, and verify the price field to ensure the compliance and consistency of the output content.

[0224] In the log and learning data management module, the server records user responses to proposal content as new behavior logs. These logs include user actions on the terminal, such as clicking, ignoring, purchasing, or rejecting proposals, along with the corresponding regional and time information. During subsequent model updates, the server merges these behavior logs with existing training data for fine-tuning of the recommendation model and the generative AI model. During recommendation model training, the server can use cross-entropy loss or mean squared error as the error function, updating the model weights through stochastic gradient descent or its variants. The generative AI model can be updated through supervised fine-tuning or preference-feedback-based training, updating local parameters or adaptation layers. By continuously introducing new behavior data closely related to regional information, the server enables the model to gradually strengthen its representation of spatial context within its internal weight space, thereby achieving rapid adaptation to new layouts or product structures.

[0225] In terms of location calculation, the terminal uses the system's signal strength measurement interface to sample signals from the location information acquisition device. The terminal can employ a simple signal strength comparison algorithm or perform region estimation based on a pre-configured fingerprint database. The terminal transmits the estimation result to the server in the form of a region identifier. The server internally maps and verifies this against a predefined table of physical space regions. If necessary, it can smooth the location trajectory based on multiple uploaded location information to reduce jitter. Through this multi-terminal collaborative approach, the system establishes a stable mapping between physical and logical spaces, thereby supporting the accurate triggering of proposals for adapting regions in real-world scenarios.

[0226] When users view proposals, they can see natural language descriptions generated based on their current location and preferences on the terminal interface. Users can select a candidate product to enter its detailed page or express feedback such as "not interested" or "look later" through operation buttons. The terminal encodes these actions as event types and sends them to the server with timestamps. After receiving this feedback, the server not only updates the behavior log but also dynamically adjusts the weights or filtering rules of the recommendation model at runtime. For example, it can temporarily reduce the exposure frequency of such products to lower the proportion of invalid recommendations, thereby reducing unnecessary data transmission and model calls, achieving a synergistic reduction in communication and computational load.

[0227] In terms of technical effectiveness, the server improves the input quality of the generative AI model by internally managing structured features and natural language prompts, enabling the generated results to better align with the recommendation model's output, thereby enhancing the overall system's recommendation accuracy. Through feature vectorization and batch inference processing, the server significantly reduces the average computation time per recommendation. An incremental training mechanism based on behavior logs and region information allows the model to quickly adapt to changes in user interests and reduce invalid inferences. Through engineered control of prompt templates and output results, the server avoids the overhead of extensive manual configuration in traditional manual rule systems and the uncontrollability caused by relying entirely on model-generated content, achieving deep integration of the recommendation system and the natural language generation system at the computer technology level.

[0228] In another implementation, the server can replace the specific structure of the recommendation model, for example, using a model combining matrix factorization and deep neural networks, or using graph-structured data to model user-item relationships. The server can also add an additional content constraint layer before the generative AI model to score and filter candidate outputs, sending only text that meets specific confidence levels or rules to the terminal. The terminal can employ different operating systems or hardware platforms in different implementations, as long as it can achieve network communication with the server and capture and report location signals. Users can utilize this system to obtain dynamic proposal content highly relevant to their current environment in various application scenarios, such as retail spaces, exhibition spaces, or service halls.

[0229] In this invention, the server does not simply automate human sales pitches. Instead, it employs a series of techniques at the underlying level, including feature engineering, model structure design, prompt generation, and log-driven learning, to fundamentally alter the way the computer collaboratively processes recommendation and text generation tasks. Through meticulous design of data structures and processing sequences, this system achieves technological improvements in recommendation accuracy, response speed, data management, and resource utilization, thus representing a substantial advancement in the field of computer technology.

[0230] use Figure 12 The processing flow is explained.

[0231] Step 1: The server receives and stores user attribute information.

[0232] The server receives attribute information input by the user through the interface from the terminal. This input includes fields such as user ID, age group, gender category, interest tags, and budget range. The server uses its communication module to parse the messages sent by the terminal, performing format validation and field integrity checks on the received structured or semi-structured data. Based on the user ID, the server performs write or update operations in the information storage unit, writing the attribute fields to the user attribute data table and setting default values ​​for missing fields. During data processing, the server maps string-type interest tags to internal tag numbers and converts budget ranges into numerical range codes, thus obtaining standardized user attribute records for subsequent model processing. The output of this step is the user attribute information record and its internal encoded representation stored in the information storage unit.

[0233] Step 2: The terminal collects and reports the spatial location information of the entity.

[0234] The terminal receives wireless signals emitted by location information acquisition devices in the physical space via Bluetooth, inputting identifiers and signal strength values ​​from multiple signal sources. The terminal uses an interface provided by the operating system to sample and filter the signals, estimating the user's location in the physical space using predefined fingerprint data or a simple signal strength comparison algorithm. The estimated result is then mapped to an area identifier, such as a shelf area or functional area number. The terminal encapsulates the area identifier, the original signal list, and a timestamp along with the user identifier into reporting data, which is then sent to the server via the network interface. The output of this step is a location information reporting message containing the user's current area identifier and the original signal information.

[0235] Step 3: The server receives location information and generates region information.

[0236] The server receives location information reports from the terminal. Inputs include the user identifier, candidate region identifiers, a list of signal source identifiers and their corresponding signal strengths, and timestamps. After parsing the message, the server matches the candidate region identifiers with the physical space region mapping table. If necessary, it recalculates or corrects the region determination based on the signal strength of multiple signal sources. During data processing, the server performs time-series smoothing on continuously reported location information to remove unreasonable jumps within short periods, thus obtaining stable region information. The server writes the final determined region identifier into the user location log table. The output of this step is the current region information record associated with the user identifier.

[0237] Step 4: The server retrieves users' historical behavior records and constructs behavioral characteristics.

[0238] The server reads historical behavior records corresponding to the user's identifier from the information storage unit. Inputs include browsing history, click history, purchase history, favorites history and their timestamps, related item or service identifiers, and the current location identifier. The server uses the data processing module to aggregate and perform calculations on this behavioral data, such as accumulating the number of interactions with different categories of items, calculating the average spending amount over a recent period, and calculating the frequency of visits to different regions. The server segments the time dimension, giving higher weight to recent behaviors and smoothing occasional behaviors to construct a behavioral feature vector that reflects the user's long-term preferences and recent changes in interests. The output of this step is a set of behavioral feature data corresponding to the user's identifier.

[0239] Step 5: The server integrates user features and region features to generate recommended input features.

[0240] The server takes the user attribute encoding obtained in step 1, the region information obtained in step 3, and the behavioral features obtained in step 4 as input. In the feature processing module, the server standardizes features from different sources, maps categorical features to embedding indices, and normalizes or bins numerical features. The server then concatenates the user attribute vector, behavioral feature vector, and region context vector in a predetermined order to generate a uniform-length recommendation input feature vector. During this process, the server sets specific positions or weighting coefficients based on feature importance to ensure that the recommendation model can effectively utilize key features internally. The output of this step is a set of multi-dimensional feature vectors suitable for inputting into the recommendation model.

[0241] Step 6: The server filters candidate sets of items or services.

[0242] The server reads candidate objects related to the current region from the product or service information data table, with input fields including current region identifier, inventory status, promotion status, and listing time. Based on the region identifier, the server filters out items or services that can be displayed or sold in this region and applies business filtering rules, such as minimum inventory thresholds, prohibited recommendation lists, and new product priority rules. The server constructs item or service characteristics for the filtered candidate objects, including category codes, price range codes, brand identifiers, and attribute tags, forming a candidate set. The output of this step is candidate set data containing candidate object identifiers and their feature vectors.

[0243] Step 7: The server calls the recommendation model to calculate the scores of candidate objects.

[0244] The server combines the user feature vector obtained in step 5 with the candidate object feature vector obtained in step 6 as input to the recommendation model. The input is a combination of multiple features (user features, candidate object features, and region features). The server invokes the recommendation model deployed on a machine learning framework, performing forward inference for each combination to calculate the corresponding preference score or purchase probability. The server uses batch processing during computation, packaging multiple combinations into a single inference request to fully utilize vector computation capabilities. After obtaining the scores of all candidate objects, the server sorts the scores and selects the highest-scoring objects as recommendation targets. The output of this step is an ordered list of recommended objects and their score information.

[0245] Step 8: The server provides prompts for building generative artificial intelligence models.

[0246] The server extracts key attribute information for each object from the recommended object list obtained in step 7. Inputs include object name, category, price, and main features. Simultaneously, the server obtains the user profile summary from step 5 and the region description information from step 3. In the prompt generation module, the server selects a text template suitable for the scenario, embedding the user profile, current region description, and recommended object summary into the template according to a predefined format to generate prompts in natural language. During text processing, the server formats numerical fields and performs readability conversion on tag fields, resulting in semantically complete and structurally clear prompt text. The output of this step is one or more prompts used to drive the generative AI model to generate proposal content.

[0247] Step 9: The server invokes a generative artificial intelligence model to generate proposal content.

[0248] The server sends the prompt generated in step 8 as input text to the generative AI model. The input includes the complete prompt and optional control parameters such as generation length and diversity. The server receives the output text from the generative AI model, which is the natural language proposal content generated for the given prompt. In the result post-processing module, the server performs semantic and format checks on the output text, including sensitive content filtering, length truncation, and comparison with actual product attributes. If content inconsistent with real data is detected, the server can correct it by replacement or regeneration. The output of this step is the validated and displayable proposal content text.

[0249] Step 10: The server sends a notification message to the terminal along with the proposal content.

[0250] The server encapsulates the proposal content obtained in step 9, along with the structured information of the recommended objects, into a notification message. Inputs include user identifiers, a list of recommended object identifiers, basic attributes of each object, and the proposal content text. The server sends the notification message to the terminal via a network interface, using either push service or a direct request-response method. Before sending, the server adjusts the message structure according to the terminal type to ensure correct parsing and display. The output of this step is a notification message stream indicating successful delivery to the target terminal.

[0251] Step 11: The terminal displays the proposal content and collects user interaction data.

[0252] The terminal receives a notification message from the server, inputting the proposal content text and display information for recommended objects. After parsing the message, the terminal generates a list or card-style display component in the graphical user interface, displaying the proposal content in natural language text on the screen, and generating clickable action elements next to the corresponding objects, such as "View Details" and "Add to List." Users interact via touch or other input methods, and the terminal records these action events as interaction data, including event type, target object identifier, timestamp, and current area identifier. The output of this step is the interaction event record cached locally on the terminal.

[0253] Step 12: The terminal reports user interaction events.

[0254] The terminal packages the interaction event records generated in step 11 into reporting data. Inputs include user identifier, object identifier, event type, timestamp, current region identifier, and session identifier. The terminal sends this reporting data to the server via the network interface. Before sending, the terminal can merge or compress multiple events to reduce network transmission volume. The output of this step is one or more reporting messages containing user interaction events.

[0255] Step 13: The server updates the behavior records and builds learning data.

[0256] The server receives interaction event reports from terminals, with input including user identifier, object identifier, event type, time information, and region information. The server writes these events to the user behavior log table, appending them as new behavior records. Simultaneously, the server constructs training sample labels for recommendation and generative AI models based on event type; for example, click events are considered positive feedback, and prolonged periods of no response are considered weak negative feedback. During data processing, the server merges these new records with the existing training dataset to generate a learning dataset containing feature vectors and feedback labels. The output of this step is the updated behavior log records and learning data that can be used for model training or fine-tuning.

[0257] Step 14: The server adjusts model parameters or recommendation strategies based on new data.

[0258] The server uses the learning data generated in step 13 to update the recommendation model or generative AI model online or train it offline. Inputs include feature data, label data, and current model parameters. During recommendation model training, the server calculates the error between the predicted output and the actual feedback, and updates the weight parameters using an optimization algorithm based on the selected loss function. During generative AI model adaptation, the server can evaluate the output quality under specific prompt patterns and adjust prompt templates or output filtering rules. The server can also update the weighting or priority conditions in candidate selection based on statistical results. The output of this step is the updated model parameters, the adjusted recommendation strategy, and the improved prompt template configuration.

[0259] 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.

[0260] 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."

[0261] With the widespread adoption of communication networks and terminal devices, users generate a large amount of natural language messages on social platforms, instant messaging tools, and various online services. While existing technologies utilize generative artificial intelligence models to automatically generate response content, the following technical challenges remain: (1) In traditional solutions, servers often pass a single message directly as input to the generative artificial intelligence model. There is a lack of systematic modeling and utilization of user attribute information and user behavior information, which leads to the generated response content being inconsistent with the user's personal style, interests and historical behavior, thus affecting the interactive experience and personalization.

[0262] (2) Existing systems often simply return the model output results directly to the terminal, lacking structured processing and screening mechanisms for generated content in terms of politeness, security, and adaptability. This can easily lead to inappropriate or even illegal content, increasing the risk of the dialogue system and hindering fine-grained control in different application scenarios.

[0263] (3) Traditional generative artificial intelligence invocation methods usually use fixed or coarse-grained prompt statements. The server side does not dynamically construct the prompt statements in a fine-grained manner, and cannot explicitly embed various constraints such as tone, length, content type, and response strategy in the prompt statements. Therefore, it is difficult to implement controllable and configurable technical constraints on the output behavior of generative artificial intelligence models.

[0264] (4) Existing technologies generally lack a technical mechanism for closed-loop feedback and adaptive updates based on user selection results. The server cannot use the user's actual selection behavior on the response candidates to dynamically update user behavior information, nor can it automatically optimize the construction of subsequent prompt statements and the generation conditions of response candidates. This results in the system being in a static policy configuration state for a long time, affecting the generation quality and system efficiency.

[0265] Therefore, in the field of computer technology, there is an urgent need for an improved server-side processing mechanism and system architecture that enables the server to: integrate user attribute information, user behavior information, and communication information in a unified data structure; perform language parsing and constraint extraction on this information; automatically construct high-quality prompts for generative artificial intelligence models based on the parsing results; perform content judgment and structuring processing on the model output; and utilize user selection behavior to achieve online learning and adaptive adjustment, thereby improving the personalization, security, controllability, and overall system performance of the generated responses.

[0266] 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.

[0267] In this invention, the server includes means for recording user attribute information and user behavior information based on communication information and user identification information from a user terminal; means for constructing a prompt statement for inputting into a generative artificial intelligence model using the user attribute information, the user behavior information, and the communication information; means for inputting the prompt statement into the generative artificial intelligence model to obtain multiple response candidates for the communication information; means for associating the multiple response candidates with the user terminal to generate structured data and sending it to the user terminal; and means for updating the user behavior information and adjusting the construction of the prompt statement or the response candidate generation conditions based on user selection information from the user terminal. This allows for unified modeling and parsing of multidimensional user information and communication context on the server side in a computer-executable manner. Response constraints are explicitly encoded in the prompt statement to finely control the generation behavior of the generative artificial intelligence model. Content judgment and structured encapsulation improve the security and adaptability of response candidates, and user selection feedback forms a closed-loop optimization process, thereby improving the personalization, robustness, and overall processing efficiency of the automatic response system at the computer technology level.

[0268] "User terminal" refers to a computing device operated by a user to send and receive communication information, including but not limited to mobile communication devices, portable terminal devices, fixed terminal devices, and computing devices with communication functions.

[0269] "Communication information" refers to data information representing content transmitted over a network between users or between a user and a service, including but not limited to text data, symbolic data, and metadata related to the dialogue context.

[0270] "User identification information" refers to the identification information used to uniquely or quasi-uniquely distinguish different users in the system, including but not limited to user identifiers, account information, and identity-related tag data.

[0271] "User attribute information" refers to descriptive information related to a user's basic characteristics, including but not limited to age, gender, occupation, language preference, interest preference, and various classification information used to characterize user traits.

[0272] "User behavior information" refers to behavioral characteristic information obtained based on users' historical operation records and interaction patterns in the system, including but not limited to message sending behavior, response selection behavior, usage frequency, content preferences, and behavioral pattern data derived from them.

[0273] "Generative artificial intelligence models" refer to models that are based on machine learning and deep learning technologies and can automatically generate text content based on input data, including but not limited to large-scale language models and models for natural language generation tasks.

[0274] "Prompt statements" refer to the instructional text contained in the data structure used as input to a generative artificial intelligence model. The instructional text explicitly describes the generation goal, constraints, and contextual information to guide the generative artificial intelligence model to generate response content in the expected manner.

[0275] "Response candidate" refers to one or more reply text units that can be selected by the user after being generated by a generative artificial intelligence model for a given communication message and processed by the system.

[0276] "Structured data" refers to a data set that is organized according to a predetermined data format and contains candidate response text and related attribute information, including but not limited to response text, tags, evaluation values, and information associated with user terminals or message identifiers.

[0277] "Language parsing and processing" refers to natural language processing operations performed on communication information and related text data, including but not limited to word segmentation, syntactic analysis, semantic analysis, sentiment analysis, and intent recognition, in order to extract semantic features and constraints for constructing prompt statements.

[0278] "Content type" refers to the classification of response content in terms of function or semantics, including but not limited to acceptance, rejection, inquiry, suggestion, confirmation and other dialogue function categories.

[0279] "Response strategy" refers to the pre-set generation guidelines or dialogue goals when generating response candidates, including but not limited to strategies such as favoring positive responses, politely declining, requesting supplementary information, providing suggestions, or remaining neutral.

[0280] "Content determination processing" refers to the analysis and screening of the response text output by the generative artificial intelligence model, in order to determine whether the text meets predetermined standards in terms of politeness, security, suitability, and compliance.

[0281] "Labels" refer to the marking information attached to response candidates to characterize their attributes or categories, including but not limited to classification tags indicating tone category, content type, risk level, and applicable scenarios.

[0282] "Evaluation value" refers to a numerical indicator used to quantify the quality or priority of response candidates, including but not limited to relevance score, politeness score, security score, and comprehensive score.

[0283] In one embodiment of the present invention, the server comprises at least one electronic computing device, which includes a multi-core central processing unit, memory, non-volatile storage, and a network interface. The server can be deployed in a data center environment and runs a general-purpose operating system, such as a Unix-like operating system or a similar server operating system. At the application layer, the server runs a backend application framework, such as a network framework based on an interpreted language or a network framework based on a compiled language, to perform functions such as receiving requests, processing data, and forwarding requests to generative artificial intelligence models.

[0284] In one embodiment of the present invention, the terminal is a communication device operated by a user. The terminal may be a mobile terminal, tablet terminal, desktop terminal, or other computing device equipped with a display module, an input module, and a network communication module. The terminal runs a client application that executes on top of the operating system, displays message content and response candidates to the user in the form of a graphical user interface, and receives touch input or keyboard input from the user.

[0285] In one embodiment of the present invention, the server uses a relational data management system as the basic software for storing user attribute information and user behavior information. The relational data management system can be a general-purpose relational database management system. The server creates multiple data tables in a non-volatile storage device, such as a user attribute table and a user behavior log table, to store user identifier fields, age fields, gender fields, interest tag fields, common reply type fields, message interaction count fields, and other feature fields used to construct user profiles. The server maintains a corresponding data cache structure in memory to accelerate the reading operation of user profile data in key-value format, thereby reducing the number of database accesses, improving overall processing speed, and reducing communication load.

[0286] In one embodiment of the present invention, the server converts communication information and user identification information from the terminal into an internal unified data structure. The server can represent a message as a record containing a message identifier, session identifier, sender identifier, receiver identifier, timestamp, and message text. The server allocates a storage area of ​​structure or object type for this record in memory for subsequent use by the natural language processing module.

[0287] In one embodiment of the present invention, the server utilizes a natural language processing (NLP) software library to perform language parsing processing on the communication information. The server can invoke a word segmentation component to segment Chinese text, invoke a syntactic analysis component to construct a dependency syntax tree, invoke a semantic analysis component to obtain the part-of-speech and semantic roles of words, and invoke a sentiment analysis model to evaluate the sentiment tendency of the sentence. The server extracts the analysis results into a set of feature values, including word sequences, sentiment scores, sentence intent labels (e.g., invitation, inquiry, rejection), and context relevance scores. The server stores these features in an intermediate feature vector for use in constructing subsequent prompt statements.

[0288] In one embodiment of the invention, the server uses a generative artificial intelligence model as the primary natural language generation module. The server can deploy this model within a separate inference service process. The model structure employs a multi-layer self-attention network neural network architecture, including word embedding layers, multi-layer encoder-decoder structures, feedforward network layers, and normalization layers. The server configures several hyperparameters for the generative artificial intelligence model, such as hidden layer dimension, number of attention heads, maximum sequence length, and sampling temperature, to achieve a balance between accuracy and inference speed.

[0289] In one embodiment of the present invention, the server drives the behavior of a generative artificial intelligence model through prompt statements. Based on the parsing results of user attribute information, user behavior information, and communication information, the server calls a prompt construction module to generate a prompt statement in natural language. This prompt construction module uses a combination of template rules and adaptive rules: on the one hand, it uses fixed templates to describe task objectives and format requirements; on the other hand, it dynamically inserts information such as age, gender, interests, and desired tone, length, and content type based on user profiles, thereby explicitly including multiple constraints in the prompt statement.

[0290] For example, if a user is a male of a certain age group with interests including "science fiction" and "action," and receives a message asking "Want to go to the movies this weekend?", the server could generate the following message: "You are a generative AI model that excels at Chinese conversations and need to help users generate appropriate response suggestions based on the chat content."

[0291] User basic information: - Age: 26 - Gender: Male - Interests: Sci-fi movies, action movies The message the user received reads: "Want to go to the movies together this weekend?" Task requirements: 1. Please provide three brief suggestions for replying in Chinese; 2. It must contain at least one positive response and one polite rejection; 3. Each reply should not exceed 25 characters, and the tone should be natural and conversational; 4. If appropriate, you can mention science fiction or action movies.

[0292] Please simply output the list of replies; do not add explanations. In another embodiment of the invention, the server constructs a prompt statement for communication information containing budget and activity plans. For example, if the user is a woman of a certain age group with interests such as "relaxation and leisure" and "short city trips," and receives a communication message such as "Want to go on a trip together during the May Day holiday? The budget is roughly within 3000 yuan," the prompt statement generated by the server could be: "Below is a chat message. Please provide three Chinese replies that include both the content of the reply and simple activity suggestions, based on the user's budget and preferences."

[0293] The user received the message: "Want to go on a trip together during the May Day holiday? The budget is around 3000 yuan." User information: 25-year-old female, enjoys relaxed and casual travel, and short city trips.

[0294] Require: 1. Each reply should not exceed 40 characters; 2. At least one suggestion should include a specific city or route; 3. Use a natural and friendly tone.

[0295] Please only output a list of the replies; do not add any other descriptions. In one embodiment of the invention, the server does not simply invoke a generative AI model to generate text. Instead, it achieves technical control over the model's output behavior through the refined construction and feature embedding of the aforementioned prompt statements. The server explicitly injects constraint information such as tone, length, content type, and response strategy into the prompt statements, enabling the generative AI model to simultaneously receive task requirements and user profile information within a unified context sequence during internal self-attention computation. This allows the model to automatically favor word selection that meets the constraints during the decoding stage through attention weight allocation. Unlike manually predefined fixed template responses, this method reduces the search space, accelerates the decoding process, and improves the matching accuracy and stability of the generated results through instruction and constraint enhancement at the input layer.

[0296] In one embodiment of the present invention, the server uses a supervised learning method during the training phase of the generative artificial intelligence model. The server constructs a training dataset, which includes sample pairs consisting of historical communication information, user attribute information, user behavior information, and corresponding high-quality responses. The server fine-tunes the pre-trained language model, using cross-entropy as the loss function to measure the difference between the model's output sequence and the target response sequence, and updates the model parameters through a backpropagation algorithm. During training, the server may employ gradient pruning, learning rate scheduling, and regularization techniques to prevent overfitting and improve the model's generalization ability.

[0297] In one embodiment of the present invention, the server performs content judgment processing on the output of the generative artificial intelligence model during the inference phase. The server can be configured with an independent content judgment module, which can use a combination of rule-based detection and classification models to determine the politeness, security, and suitability of the generated text. The server can pre-set a sensitive word list, tone detection rules, and domain adaptation rules to quickly filter candidate texts; simultaneously, it can call a dedicated classification model to score whether the text output contains offensive, illegal, or inappropriate content. The server removes candidate texts that do not meet the minimum scoring threshold, retaining only the filtered response candidates.

[0298] In one embodiment of the invention, the server attaches structured attributes, including tags and evaluation values, to each reserved response candidate. Based on semantic and tone features appearing in the candidate text, the server uses rule-based classification or a lightweight classification model to assign category tags such as "accept," "politely refuse," "ask for supplementary information," and "suggested solution." It then calculates a comprehensive evaluation value based on indicators such as semantic similarity to the communication information, user interest matching, and politeness score. The server combines the response text, tags, and evaluation values ​​into a structured data record and sends it to the terminal via a network interface.

[0299] In one embodiment of the present invention, after receiving structured data, the terminal displays multiple response candidates in a list or pop-up manner in a graphical user interface. The terminal can display different colors or icons based on labels to help the user quickly identify the response type. The terminal maintains a mapping relationship between the current message and response candidates in its local memory and marks the response finally selected by the user as the selected candidate.

[0300] In one embodiment of the present invention, a user clicks on a candidate response via a terminal interface, and the terminal fills the candidate text into an input area, which the user can then edit. After the user confirms the send, the terminal transmits the edited text to the communication server via the network, thus realizing the actual message sending. The terminal can also report the user's selected candidate identifier, selection time, and whether editing was performed to the server as part of the user's behavior information.

[0301] In one embodiment of the present invention, the server updates user behavior information using selection information from the terminal. The server selects and edits flags for each candidate response record in the user behavior log table, and statistically analyzes the selection frequency and editing ratio of a user under different response strategies, thereby gradually forming a preference vector. When constructing subsequent prompt statements, the server can dynamically adjust constraints based on this preference vector, for example, increasing the weight of user preference strategies and reducing strategies that the user chooses less frequently, thereby achieving adaptive optimization in prompt statement construction.

[0302] In one embodiment of the invention, the server continuously optimizes system performance based on updates to user behavior information. Since the server only conveys a small amount of high-weight strategy information to the generative artificial intelligence model in the prompt statements, the search space required for actual generation can be reduced, decreasing the number of invalid candidates generated. This feedback-based prompt optimization mechanism improves the utilization efficiency of computing resources, reduces the average inference time, and decreases the number of redundant candidates in network transmission, thereby achieving the technical effects of reduced communication load and increased processing speed.

[0303] In one embodiment of the invention, the entire data processing flow of the server adopts a modular structure, with all modules communicating with each other through memory queues or message queues. The server divides communication information parsing, prompt statement construction, generative artificial intelligence model invocation, content judgment, and structured encapsulation into different functional modules, each connected through a clearly defined data structure interface. This modular structure facilitates parallel processing and horizontal scaling. When the number of user requests increases, the server can independently expand the number of instances of specific bottleneck modules, thereby achieving scalability and high availability at the computer system level.

[0304] In another embodiment of the invention, the server can employ different types of generative artificial intelligence model structures, such as autoregressive generative models or sequence-to-sequence models using encoder-decoder structures. The server can also employ a multi-task learning approach, simultaneously learning response generation and intent classification tasks within the same model to better utilize intent features when generating responses. The server can also use knowledge distillation methods to distill the behavior of a large teacher model into smaller student models, enabling the inference phase to use models with lower computational complexity, thereby reducing hardware resource overhead and improving response speed.

[0305] In another embodiment of the invention, data augmentation techniques can be employed during the training of the generative artificial intelligence model. For example, the server can perform synonym substitution, word order adjustment, or add appropriate noise to the message text in the training data to improve the model's robustness to different expressions. During training, the server can set different weights, assigning higher weights to samples closely related to user attributes and behaviors, enabling the model to place greater emphasis on personalized features during the learning process.

[0306] In one embodiment of the invention, the cooperation between the server and the terminal is not limited to text-based responses. The server can combine response candidates with the terminal's notification mechanism, speech synthesis module, or other control modules. For example, the server can instruct the terminal to create a schedule reminder or navigation request locally based on structured information such as time and location contained in the response candidates. This combines the natural language response result with the actual hardware function control of the terminal, realizing an integrated technical process from dialogue generation to device behavior triggering. This combination means that the invention goes beyond simply generating dialogue content; it further maps the generated results into terminal behavior control instructions, technically enhancing the system's practical application capabilities.

[0307] In various embodiments of this invention, the server forms a specific data flow and algorithm processing flow within the computer by jointly modeling user attribute information, user behavior information, and communication information; by providing structured control over the input to the generative artificial intelligence model through prompt statements; by standardizing the output through a content judgment module; and by employing a user feedback-driven continuous optimization mechanism. This flow combines the sequence modeling capabilities of neural networks with rule constraints and achieves efficient computation through specific data structures and module divisions, thereby technically achieving multiple benefits such as improved generation accuracy, shortened inference time, reduced error output, and reduced network load.

[0308] use Figure 13 The processing flow is explained.

[0309] Step 1: The terminal receives communication information and generates request data.

[0310] The terminal obtains a raw data packet containing message text, sender identifier, receiver identifier, and timestamp from the communication network as input. The terminal parses this data packet, converting it into an internal message object, and displays the message content on the interface. When the user selects a message and triggers the "Request Reply Suggestion" operation, the terminal reads the message text, message identifier, session identifier, and current user identifier from the internal message object, encapsulates these fields into a request data structure, and sends it as output to the server.

[0311] Step 2: The server receives the request and extracts user attribute information and user behavior information.

[0312] The server takes the request data sent by the terminal (including user identifier, message identifier, message text, session identifier, etc.) as input and performs format validation and authentication on the request. Based on the user identifier, the server performs a query operation in the relational database, retrieving attribute information such as age, gender, and interest tags from the user attribute table, and statistically analyzing behavioral information such as historical response choices and preference strategies from the user behavior log table. The server encapsulates the retrieved attribute and behavioral information along with the original communication information into a unified user context data structure, which is then provided as output to subsequent processing modules.

[0313] Step 3: The server performs language parsing processing on the communication information and generates semantic features.

[0314] The server takes the message text and conversation-related text from the user's context data structure as input, and calls natural language processing components to perform data processing operations such as word segmentation, part-of-speech tagging, syntactic analysis, sentiment analysis, and intent recognition. Based on the word segmentation results, the server constructs word sequences; based on syntactic analysis, it constructs dependency relations; based on sentiment analysis, it obtains sentiment scores; and based on the intent recognition model, it calculates the probability that the message belongs to an invitation, inquiry, or other category. The server combines these results into a set of semantic feature vectors, and outputs this feature vector along with user attribute information and user behavior information to the prompt statement construction module.

[0315] Step 4: The server constructs prompts for generative artificial intelligence models.

[0316] The server takes user attribute information, user behavior information, and semantic feature vectors as input and performs template filling and rule matching operations. First, the server selects a prompt template corresponding to the current intent category. Then, based on the user's age, gender, interests, and behavioral preferences, it inserts constraints into the template, such as the expected number of responses, response tone, length limits, content type, and response strategy. The server integrates the task description, basic user information, original message text, and constraints into a prompt statement in natural language form through string concatenation and variable substitution. The server outputs this prompt statement text as input instructions for the generative artificial intelligence model.

[0317] Step 5: The server invokes a generative artificial intelligence model to generate multiple response candidates.

[0318] The server takes a pre-constructed prompt as input, encodes it into a labeled sequence that the model can process, and passes it to the generative AI model deployed in the inference service. During model inference, the server sets parameters such as maximum generation length, sampling temperature, and decoding strategy (e.g., greedy search or bundle search), and performs forward inference operations. The generative AI model, based on its internal neural network weights, performs multi-layered self-attention and feedforward operations on the input sequence, progressively generating the output labeled sequence. The server decodes the model's output labeled sequence back into natural language text, splits one or more segments of text into several original response texts by line or delimiter, and passes these as output to the content determination module.

[0319] Step 6: The server performs content judgment and filtering on the original response text.

[0320] The server takes the list of raw response texts output by the generative artificial intelligence model as input and performs content judgment processing on each text. The server calls the rule detection module to match sensitive words, prohibited phrases, and non-compliant expressions, marking texts that do not meet security rules as invalid. The server calls the classification model to score the politeness, security, and contextual suitability of each text. The server filters out texts with scores below a preset threshold, retaining only response texts that meet the security and suitability criteria. The server provides the filtered list of response texts as output to the tagging and scoring modules.

[0321] Step 7: The server generates tags and evaluation values ​​for the response text and constructs structured data.

[0322] The server takes a filtered list of response texts and a user context data structure as input, and calculates a tag and evaluation value for each response text. The server determines whether the response belongs to a category such as "accept," "politely decline," "asking for supplementary information," or "suggested solution" using rules or a lightweight classification model, and writes this category as a tag into a structure field. The server comprehensively calculates the evaluation value for each response based on indicators such as semantic similarity, user interest matching, and politeness score. Subsequently, the server combines the response text, tag, evaluation value, and corresponding message identifier and user identifier into a structured data record, and sends the set containing multiple records as output to the terminal.

[0323] Step 8: The terminal receives structured data and displays response candidates.

[0324] The terminal takes structured data returned by the server as input and uses a parsing module to read the response text, tags, and evaluation values ​​from each record. The terminal establishes a mapping relationship in its local memory based on message identifiers, binding response candidates to their corresponding messages. The terminal generates multiple clickable response options in the graphical user interface, which can be displayed with different visual styles based on tags, such as using different colors or icons to distinguish between "accept" and "reject" responses. The terminal presents this list of response candidates as output to the user.

[0325] Step 9: Users select or edit response candidates and send the final reply.

[0326] The user inputs a list of response candidates displayed on the terminal, selects one by clicking, or edits the text in the input area after selection. After confirming the response, the user triggers a send operation. The user's selection and any possible text modifications are recorded by the terminal and used to generate a data packet containing the final response text, the selected candidate identifier, and an edit flag. The user ultimately causes the terminal to output a formal response message for the communication service.

[0327] Step 10: The terminal reports the user's selection information, and the server updates the user's behavior information and optimizes subsequent prompts.

[0328] The terminal takes the user-selected candidate identifier, whether to edit, message identifier, and user identifier as input to generate behavioral feedback data. After sending a formal response, it asynchronously reports this feedback data to the server. Upon receiving the feedback data, the server merges and updates it with existing user behavior information. This is done by adding fields such as selection count, edit count, and strategy preference weight to the behavior log table. Based on the updated behavioral characteristics, the server adjusts the rules for constructing subsequent prompt statements. For example, it increases the weight of commonly used response strategies in the template or reduces the generation ratio of less frequently used strategies. The server stores the new parameters and rules in the configuration storage for use as input when constructing subsequent prompt statements, thereby outputting prompt statements that better match user preferences and further improving the overall generation effect and system efficiency.

[0329] 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".

[0330] In existing technologies, AI-based matching and recommendation systems, dialogue support systems, and decision support systems mostly employ fixed rules or static model parameters for information processing. Specifically, regarding the utilization of user attribute information, preference information, and behavioral history information, matching results are typically output directly through pre-trained prediction models, lacking fine-grained control over the input side of generative AI models. Regarding the parsing of conversations or messages, most systems only perform simple text classification or keyword extraction, failing to abstract and arrange different tasks at a unified prompt level. Regarding the recognition and utilization of user emotions, existing systems often treat emotion tags as simple filtering conditions or weighting coefficients, failing to systematically couple emotional states with the generation process of generative AI models, thus making it difficult to fine-grained adjust the tone, structure, and strategies of the output content.

[0331] Furthermore, in existing technologies, user interaction feedback (such as selection information and evaluation information) is mostly used for offline retraining, lacking a mechanism to dynamically generate prompts and adjust subsequent generation logic in real time during runtime. This results in a slow response of the system to long-term user preferences and short-term emotional changes, making it difficult to update matching ranking, response content, or plan generation strategies in a timely manner within the computer. From a computer technology perspective, this leads to: rigid processing flow, poor configurability, low resource utilization efficiency, and an inability to fully leverage the expressive and reasoning capabilities of generative artificial intelligence models.

[0332] In summary, a new information processing system and its computer implementation are needed. This system should manage and dynamically generate prompts for generative artificial intelligence models on the server side, organically combining user attributes, behavioral history, emotional state, real-time interactive feedback, and different types of tasks (matching, dialogue support, plan generation, decision guidance, etc.). This would create a programmable, schedulable, and continuously optimizable processing pipeline within the computer, thereby improving the flexibility and intelligence of the information processing flow, enhancing the relevance and personalization of the generated results, and improving the overall utilization efficiency of computing resources.

[0333] 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.

[0334] In this invention, the server includes a processing unit for acquiring attribute information, preference information, and behavioral history information of the user and generating prompt statements to instruct the generative artificial intelligence model to parse and process the information; a processing unit for converting dialogue information or transmitted information obtained from the communication terminal into structured data through natural language processing and generating prompt statements based on the structured data to instruct the generative artificial intelligence model to generate response text or suggestion text; and a processing unit for using an emotion-specific model or emotion analysis function to determine the user's emotional state and generating prompt statements based on the determined emotional state to instruct the generative artificial intelligence model to adjust matching candidates, suggestion content, or plan schemes. The system includes a processing unit for prompt statements, a processing unit for generating prompt statements containing query content, stylistic conditions, and output constraints based on the user's query or question information to instruct the generative AI model to automatically generate solution information or guidance information, a processing unit for generating prompt statements based on constraint information including budget information, time conditions, and preference conditions to instruct the generative AI model to automatically generate activity plans, transaction processes, or interpersonal communication plans, and a processing unit for regenerating prompt statements containing additional condition information based on selection or evaluation information from the communication terminal to update subsequent matching candidates, response texts, suggestion texts, or plan generation processes. This allows for unified modeling and dynamic generation of prompt statements within the server, decoupling and programmablening of different types of input data from the generative AI model's generation behavior. This enables fine-grained control based on context and emotional state, thereby improving the relevance and real-time performance of matching and recommendation results, enhancing the personalization of dialogue responses and plan generation, and improving the flexibility and scalability of information processing flows by reducing reliance on fixed rules and static model structures. From a computer technology perspective, this optimizes the overall system's resource utilization efficiency and processing performance.

[0335] "System" refers to an overall set of devices consisting of at least one server, at least one communication terminal, and a communication network for data interaction between the server and the communication terminal, used to perform the information processing functions described in this invention.

[0336] A "server" is an electronic device that includes a processor, memory, and communication interface. It is a computing device used to execute program code to perform data acquisition, parsing, prompt generation, and calling generative artificial intelligence models.

[0337] "Communication terminal" refers to an electronic device operated by a user and capable of sending and receiving data with a server through a communication network, including but not limited to mobile terminals, fixed terminals, or wearable terminals, used to send input information to the server and receive output results from the server.

[0338] "Processing unit" refers to a software module, hardware module, or combination of software and hardware module that runs on the processor of a server and performs specific functions by executing predetermined program instructions, and is used to complete the various data processing steps defined in this invention.

[0339] "User" refers to the object that uses the system to input information, receive system output, and interact with the system, including natural persons, legal persons, or other organizational entities.

[0340] "Attribute information" refers to a set of data used to represent the basic characteristics of the user, including but not limited to static or relatively stable information such as age, gender, occupation, geographical location, and language preference.

[0341] "Preference information" refers to data used to indicate the user's interests or preferences, including but not limited to hobbies, content preferences, style preferences, and consumption preferences.

[0342] "Behavioral history information" refers to a data set that records the past operations, choices, and interactions of the user in the system or related environment, including but not limited to browsing records, click records, information records sent or received, historical matching results, or historical plan adoption status.

[0343] "Dialogue information" refers to communication content, primarily in natural language, generated during the interaction between the user and the system or between the user and the system. This includes text messages, transcribed speech content, and metadata related to the dialogue context.

[0344] "Information transmission" refers to various types of information data sent or received in the system through communication terminals, including but not limited to notification messages, request messages, response messages, and service-related parameter information.

[0345] Natural Language Processing (NLP) refers to the technical process of computer processing of natural language text or speech-to-text text, including operations such as word segmentation, part-of-speech tagging, syntactic analysis, semantic analysis, intent recognition, entity recognition, and sentiment analysis.

[0346] "Structured data" refers to data obtained from natural language processing or other data processing processes, organized in a predetermined format or data structure (such as key-value pairs, tables, tree structures, or graph structures), which is convenient for further processing by computer programs.

[0347] "Generative artificial intelligence models" refer to artificial intelligence models trained based on machine learning or deep learning methods that can generate natural language text or other forms of output based on input information, including but not limited to language models and multimodal generative models.

[0348] "Prompt statements" refer to control text or instructions input into a generative artificial intelligence model, used to indicate the target task, input elements, output requirements and constraints to guide the model to generate expected results.

[0349] "Response text" refers to the natural language text output by a generative artificial intelligence model after receiving a prompt statement, which processes the input information according to the task indicated therein and is used to answer questions or respond to requests.

[0350] "Suggestion text" refers to natural language text generated by generative artificial intelligence models based on prompts and related input data, used to provide suggestions, opinions, or guidance to users.

[0351] "Emotion-specific models" refer to models used to analyze input data to identify or classify emotional states, including emotion recognition models based on statistical learning, machine learning, or deep learning.

[0352] "Sentiment analysis function" refers to the function of processing text, voice or other behavioral data using sentiment-specific models or rule systems to obtain results such as sentiment labels, sentiment intensity or sentiment tendency.

[0353] "Emotional state" refers to the result that indicates the psychological state or emotional tendency of the subject by identifying relevant data through emotion-specific models or emotion analysis functions, including but not limited to categories such as happiness, sadness, anger, anxiety, loneliness, or combinations thereof.

[0354] "Matching candidates" refers to a set of objects or solutions that can be selected or referenced by the user based on the user's attribute information, hobby information, behavioral history information, and emotional state, and is used to achieve matching between people, between people and content, between people and services, or between people and plans.

[0355] "Suggested content" refers to the specific set of information provided by the system to users based on the output of generative artificial intelligence models and related data analysis results, which is used to assist in decision-making or behavioral choices. This includes behavioral suggestions, communication suggestions, consumption suggestions, etc.

[0356] "Plan" refers to a structured or semi-structured plan generated by the system for the user, which includes elements such as time arrangement, location arrangement, step arrangement and resource allocation, including but not limited to activity plans, transaction processes, interpersonal communication plans, etc.

[0357] "Inquiry information" refers to input information sent by a user to a system for the purpose of obtaining explanations, answers, or guidance, including questions, inquiries, or requests in natural language.

[0358] "Problem information" refers to input information from a user describing the difficulties, malfunctions, conflicts, or needs they face, in order to request the system to provide solutions or processing suggestions.

[0359] "Solution information" refers to a set of step-by-step or strategic information generated by the system based on the output of a generative artificial intelligence model, used to solve the problems raised by the user.

[0360] "Guidance information" refers to the prompting information output by the system to the user to indicate the next operation, select a path, or choose a direction of action. This includes text guidance, step prompts, or strategy descriptions.

[0361] "Budget information" refers to data related to the acceptable expenditure range of the user, including the upper limit of the amount, the type of currency, payment method preferences, and related constraints.

[0362] "Time constraints" refers to time constraints related to the execution of a plan or the conduct of an activity, including elements such as date, time period, duration, and time flexibility.

[0363] "Preference conditions" refer to the constraints on the use of the subject's interests, preferences, or dislikes that need to be considered when generating plans or matching candidates.

[0364] "Constraint information" refers to a set of parameters, consisting of budget information, time conditions, preference conditions, and other limiting factors, used to limit the search space for generated results or the optimization objective.

[0365] An "event plan" refers to an execution plan generated for offline or online activities, which includes the event content, schedule, location selection, and resource allocation.

[0366] A "transaction process" refers to a process structure designed to complete a transaction, which includes various operational steps, their sequence, and condition checks.

[0367] "Interpersonal communication plan" refers to a plan or scheme that is generated to achieve specific interpersonal communication goals, including communication methods, communication frequency, topic arrangement and coping strategies.

[0368] "Selection information" refers to the record of the selection results or adoption status of the user subject in response to the matching candidates, response texts, suggestion texts or plan schemes provided by the system.

[0369] "Evaluation information" refers to data formed by users' subjective scoring, comments, or labeling feedback on the system's output results, which is used to reflect the users' satisfaction or preference for the relevant results.

[0370] "Additional conditional information" refers to supplementary conditional data added during the regeneration of prompt statements, based on the user's selection information, evaluation information, or other newly acquired information, to further constrain or guide the output of the generative artificial intelligence model.

[0371] In the following embodiments, the server, terminal, and user participate as subjects in the operation of the system of the present invention. The server is implemented in a data center or cloud computing environment, the terminal is the electronic device actually operated by the user, and the user is the subject that interacts with the system. All embodiments revolve around the unified management and dynamic generation of generative artificial intelligence models and prompt statements. By constructing specific data structures, algorithmic processes, and neural network configurations within the computer, efficient control over matching, dialogue response, plan generation, and decision guidance is achieved, thereby improving the processing efficiency and accuracy of the computer technology itself.

[0372] I. Example of Overall Hardware and Software Composition Servers use computing devices that include multi-core general-purpose processors and large-capacity memory, such as rack-mount servers employing multi-core central processing units (CPUs) and / or graphics processing units (GPUs). Servers include: - Processor: Multi-core CPU (e.g., x86 architecture processor) and optional GPU accelerator card for large-scale matrix operations and neural network inference; - Main memory: such as 128GB or more of random access memory, used to store runtime data structures and model parameter caches; - Non-volatile memory: solid-state drives or disk arrays, used to store user data, behavior logs, training datasets and model files; - Communication interface: Ethernet adapter or wireless communication module, used for data interaction with the terminal via the Internet; - Operating system and runtime environment: such as general server operating system and application runtime environment; - Software components: Natural language processing libraries (e.g., text parsing modules based on spaCy), sentiment analysis libraries (e.g., sentiment classification models based on deep learning), generative artificial intelligence model interface modules (e.g., calling interfaces compatible with GPT series models), etc.

[0373] The terminal is a user-side electronic device, which can be a smartphone, tablet computer, personal computer, or head-mounted display device. The terminal includes: - Input modules: touch screen, physical keyboard, microphone, etc., used to collect text or voice; - Display module: LCD screen, head-mounted display, etc., used to display text, icons or interface components output by the server; - Local processor and memory: used for lightweight natural language preprocessing, interface rendering, and cache management; - Communication modules: Cellular network modules, wireless LAN modules, etc., used for data communication with servers via networks.

[0374] Users input natural language text or voice through a terminal, which sends the data to a server. The server performs the multi-stage data processing and calculation as described in this invention and sends the generated results back to the terminal for display to the user.

[0375] II. Server-side modules and data structures The server comprises multiple logical modules, which are implemented as programs on the processor and exchange data via shared memory or message queues, including but not limited to: - User data management module; - Natural Language Processing Module; - Sentiment analysis module; - Prompt statement generation module; - Generative artificial intelligence model interface module; - Result post-processing and sorting module; - Log and feedback management module.

[0376] Internally, the server uses a unified data structure to manage information. For example, the server organizes all data related to a particular request into a session object, which can use a key-value mapping structure and contain: - Basic information fields: user identifier, timestamp, device type, etc.; - Input text field: User's raw natural language input; - Parsing result fields: including intent tags, entity list, syntax tree structure index, etc.; - Sentiment status field: sentiment label and confidence score; - Constraint fields: budget limit, time range, preference tags, etc.; - Model configuration fields: target task type, generation length limit, temperature parameters, etc.; - Output fields: matching candidate list, response text, suggestion text, or list of plans, etc.

[0377] With the unified data structure described above, the server can pass data between different modules by reference, reducing repeated parsing and redundant calculations, thereby improving the overall processing speed and resource utilization.

[0378] III. Implementation Forms of User Attribute and Behavior Analysis In one implementation, the server maintains user profiles through a user data management module. The server stores user attribute information, preference information, and behavioral history information as standardized feature vectors, which are used as input conditional information when generating prompt statements later.

[0379] In this implementation, the server uses feature engineering methods to transform behavioral history information into high-dimensional sparse vectors or dense embedding vectors. For example, the server statistically encodes the user's browsing history, click history, previously selected matching objects, and adopted plans to obtain the following features: - Frequently appearing topic tag vectors; - The number of interactions with different types of objects and the distribution of those interactions over time; - The probability of adopting different price range plans; - The probability of behavioral shifts under different emotional states, etc.

[0380] When generating prompts, the server embeds these features into the text, guiding the generative AI model to construct more targeted output internally in a textual way. For example, the server constructs the following prompt: "User profile: Male, 28 years old, usually prefers outdoor activities and low-budget plans, and has a positive feedback on activities involving natural scenery. Please prioritize outdoor, natural, and low-cost options when generating date plans." In this way, the server does not need to implement a complex rule system on the client or in the business logic layer. Instead, it injects the rules into the prompt statements with high-level semantics, enabling the generative artificial intelligence model to adaptively generate results based on rich context. This reduces the maintenance cost of traditional rule engines and improves system scalability.

[0381] IV. Implementation Forms of Session Parsing and Response Generation In another implementation, the server uses a natural language processing module to perform multi-layered parsing of dialogue or transmitted information from the terminal. The server first performs word segmentation and part-of-speech tagging on the input text, then constructs a dependency syntax tree to identify subject-verb-object relationships, conditional clauses, and time adverbs, among other information. The server stores the parsing results in structured data fields for use by the prompt generation module.

[0382] The server interacts with externally or internally deployed generative AI models using a generative AI model interface module. This model can be a multi-layer neural network based on a transformer architecture, incorporating multi-head attention mechanisms and feedforward networks. Model training employs an autoregressive language modeling approach, using a loss function such as cross-entropy loss. Training data includes large-scale corpora and filtered human-computer dialogue data. During training, the server or training environment updates the weights of each layer using backpropagation and iterates the parameters using optimization algorithms (such as the Adam optimization algorithm) until a predetermined accuracy is achieved on the validation set.

[0383] The server does not retrain the generative AI model during runtime; instead, it provides fine-grained control over the model's behavior through prompts. For example, when a user asks a payment question, the server generates the following prompt: "You are a customer service assistant on an electronic trading platform. A user asks: 'My payment keeps failing to complete, what should I do?' Please use simplified Chinese to explain step-by-step what you can check, such as network connection, account status, payment method settings, etc., and remind the user at the end that if the problem persists, they can contact customer service. The total word count should not exceed 200 characters." The server provides the prompt and user question as input to the generative AI model. Internally, the model uses an attention mechanism to jointly encode the prompt and question, generating a response text that meets the constraints at the output layer. The server then uses a post-processing module to format the response and filter out sensitive content before sending the processed response text to the terminal.

[0384] Because the prompt statement limits the structure and content range of the response, the server can flexibly adjust the output style and information density without modifying the model weights. Compared with traditional fixed templates or rule systems, it can maintain a high degree of consistency and professionalism while ensuring diversity, thus improving the accuracy and controllability of response generation.

[0385] V. Implementation Forms of Sentiment Analysis and Dynamic Adjustment In another implementation, the server uses a sentiment analysis module to perform sentiment recognition on user input. The server can employ a text classification neural network based on bidirectional encoding representation. This network uses corpora with sentiment labels during the training phase, with a loss function of multi-class cross-entropy, and updates the network weights through an optimization algorithm.

[0386] During runtime, the server converts the user-input text into a sequence of word embedding vectors, which are then fed into a sentiment analysis network to obtain a sentiment probability distribution, such as loneliness, sadness, anger, and anticipation. The server writes the sentiment tag corresponding to the highest probability and its confidence level into the sentiment state field, and uses it as a conditional embedding when generating prompts.

[0387] For example, when a user enters, "I've been feeling very lonely lately," the server recognizes the emotional state as "lonely" and generates the following prompt: "You are an emotional counselor. The user's current primary emotion is loneliness. The user expresses: 'I've been feeling very lonely lately.' Please use simplified Chinese to first briefly express empathy, and then provide 3 specific and actionable suggestions to help the user alleviate loneliness. Each suggestion should not exceed 40 characters, and your tone should be gentle and encouraging." The server inputs the prompt into a generative AI model, allowing the model to consider emotional state during generation and thus output suggestions more suitable for the current psychological state. This method of modulating prompts by emotional state enables the system to dynamically adjust its generation strategy internally, rather than simply making post-processing adjustments to the result. This reflects emotional conditions within the model's internal reasoning path, improving the accuracy of emotional matching and user satisfaction.

[0388] VI. Implementation Forms of Plan Generation and Decision Guidance In the implementation planning stage, the server stores budget information, time conditions, and preference conditions as constraint information fields. When generating prompt statements, the server reads these fields from the session object to construct a rich text description of the constraints.

[0389] For example, when a user enters through the terminal: "With a budget of 5000 yen, I'd like to arrange a date that allows me to get close to nature," the server generates the following prompt: "You are a dating and romance planning expert. The user has a budget of approximately 5,000 yen and wants to plan a date with their partner that involves getting close to nature. Please design 2-3 specific date plans in and around the city. Each plan should include: the location of the activity, the main activities, the estimated cost breakdown (transportation, food, entrance fees, etc.), and a suitable time of day. Please answer in simplified Chinese and keep the total word count under 300 characters." The server inputs the prompt into a generative artificial intelligence model. Internally, the model automatically searches and combines suitable activity elements based on constraints to generate a specific solution. The server parses the generated solution into structured data, dividing it into location, activity list, and cost details, and returns it to the terminal. The terminal can display these solutions in card format, allowing users to directly select, adopt, modify, or regenerate them.

[0390] Through this structured generation and parsing, the server not only generates natural language text but also data structures that can be further manipulated by the terminal. Compared to traditional static recommendation systems, this implementation uses a generative artificial intelligence model to jointly optimize the content and text representation of the solutions, enabling the system to quickly search for solutions that meet complex constraints within a large-scale solution space, thereby improving computational efficiency and the quality of generated results.

[0391] VII. User Feedback-Driven Dynamic Update Implementation Model In a feedback-driven implementation, the server collects user selection and evaluation information through a log and feedback management module. The server stores this information, along with the original prompts, in a history for later use in generating subsequent prompts.

[0392] When the server generates matching candidates or plans for the same user again in a subsequent session, it adds the previous selection and evaluation information as additional conditions to the new prompt statement. For example: "The user has repeatedly chosen outdoor dating options and rated the previous park picnic plan as 'very satisfied'. When generating this dating plan, please prioritize similar options and avoid recommending expensive indoor activities." The server embeds this information into new prompts, guiding the generative AI model to prioritize outputs consistent with historical preferences. By accumulating long-term preference information in the prompts, the server can achieve personalized long-term adaptation without changing model parameters, thereby reducing uncertainty during model inference and improving the stability and persuasiveness of the results.

[0393] Furthermore, the server can rank candidate results based on feedback information, employing a weighted scoring algorithm that linearly or non-linearly combines the relevance score of the generated content with historical adoption rates and evaluation scores, thus prioritizing the presentation of high-scoring solutions on the terminal. This ranking method relies only on computable features and functions internally, possessing a well-defined algorithmic structure, which helps improve recommendation accuracy and reduce errors.

[0394] VIII. Technical Effects and Improvements in Computer Technology The server achieves the following technical improvements through the prompt statement generation and generative artificial intelligence model control mechanism of this invention: 1. Improved processing speed: The server unifies multiple tasks (matching, responding, and plan generation) into a prompt statement-driven framework, avoiding the need to design complex rule engines for each task, reducing branch logic judgments and multiple database queries, thereby shortening the overall response time.

[0395] 2. Improved accuracy: By incorporating emotional states, historical behavioral features, and constraints into the prompt statements, the server enables the generative AI model to obtain richer context during reasoning, thereby improving the consistency between matching results, response content, and planned solutions and user needs.

[0396] 3. Data Management Optimization: The server adopts a unified data structure to manage session objects and feature vectors, reducing duplicate storage and redundant conversion, improving memory utilization efficiency, and reducing serialization and deserialization overhead when transferring data between different modules.

[0397] 4. Computational efficiency: The server uses GPUs to accelerate the inference of generative artificial intelligence models and controls the generation length and level of detail in the prompts to avoid generating excessively long and useless text, thereby reducing computational load and network transmission volume.

[0398] 5. Communication load reduction: The server performs lightweight preprocessing on the terminal side to reduce the amount of raw data. The results generated on the server side are compressed and structured before being sent to the terminal, reducing bandwidth usage.

[0399] Unlike traditional systems that simply automate human operations, this invention improves computer processing at the algorithmic level by introducing specific prompt generation algorithms, feature injection strategies, and emotion modulation mechanisms within the server. This allows for controllable changes to the internal reasoning process of generative artificial intelligence models. This improvement not only alters how the system processes input data but also introduces new control variables into the model's reasoning path, enabling explicit, programmable control over the neural network's output behavior. This represents a structural enhancement of computer technology itself.

[0400] IX. Optional Variations and Other Implementation Forms In its optional variants, the server can employ generative AI models of varying sizes and structures. For example, smaller models can be used for offline, local execution on the terminal, while larger models can be used for centralized inference in the cloud. The server can also select different specialized models based on the task category, such as models specifically designed for customer service Q&A or models specifically designed for generating plans.

[0401] In another embodiment, the terminal can be a wearable device with a camera and microphone. When a user moves within a physical space, the terminal collects environmental information and voice input. Based on this information, the server generates suggestions or prompts in real time, which can be used for scenarios such as on-site guidance and assisting with customer service in physical stores. In this scenario, the server output is not only used for display but can also be used to control the terminal's reminder mode, interface layout, or notification frequency, further demonstrating the integration of this invention with real-world device control.

[0402] Users can exist as individual users, groups of users, or organizations in different implementation forms. The server can achieve group-level plan generation and matching optimization by introducing group characteristics and collaborative preferences into the prompt statements. For example, for multi-person activity plans, the server can add "the common available time and common preferences of multiple participants" to the prompt statements, so that the generative artificial intelligence model can automatically generate a solution that meets the constraints of multiple people, thereby reducing the burden of complex coordination logic on the terminal side.

[0403] Through the above-mentioned implementation forms and their variations, the server, terminal and user work collaboratively under a unified prompt statement-driven framework to achieve efficient control and utilization of generative artificial intelligence models, thereby achieving a comprehensive improvement in speed, accuracy and resource utilization at the computer technology level.

[0404] use Figure 14 The processing flow is explained.

[0405] Step 1: Users input information and send requests using the terminal.

[0406] Users can input natural language content through the terminal interface using a keyboard, touch screen, or voice input module, such as: "With a budget of 5,000 yen, I want to arrange a date that allows me to get close to nature." or "I can't complete the payment, what should I do?"

[0407] The terminal takes the raw text input by the user as input data, along with metadata such as user identifier, device type, timestamp, and session identifier, and organizes it into a request object. The terminal performs basic data processing on the text, such as removing leading and trailing spaces and limiting the maximum length, and encodes the result as a UTF-8 string. The terminal then sends this request object as output to the server via network protocols.

[0408] Step 2: The server receives the request and performs preprocessing.

[0409] The server receives request objects from the terminal via the communication interface as input data and performs format validation on the fields (e.g., checking if the text is empty, if the length exceeds the limit, and if required fields exist). The server then processes the data, including standardizing character encoding, removing control characters, logging access, and writing the cleaned request data into the session cache structure. The server's output is a "session data structure" maintained in memory, containing the original text, basic user information, and preliminary validation results.

[0410] Step 3: The server performs natural language parsing to generate structured data.

[0411] The server takes the user text from the session data structure as input and calls a natural language processing module (such as a word segmentation and syntactic analysis component built on spaCy) to process the text, including word segmentation, part-of-speech tagging, named entity recognition, and dependency parsing. Based on the parsing results, the server extracts intents (e.g., "date plan generation," "payment inquiry") and key entities (e.g., amount "5000 yen," topic "natural," keywords "payment," "unable to complete"). The server then constructs a structured data object (e.g., JSON-formatted intent tags, entity lists, and syntactic tree indexes) based on these parsing results. This structured data object is written as output into the session data structure for use in subsequent steps.

[0412] Step 4: The server performs sentiment analysis to determine the user's emotional state.

[0413] The server takes user text and optional historical dialogue fragments as input, calls the sentiment analysis module (e.g., a sentiment classification model based on bidirectional coding networks), transforms the text into a sequence of word embedding vectors, and performs forward propagation to obtain the probability distribution of each sentiment category. This process involves data computation, including matrix multiplication, nonlinear activation, and normalization. The server selects the primary sentiment tag (e.g., "loneliness," "anxiety," "expectation") based on the highest probability and generates a sentiment state object containing the sentiment tag and confidence value. This object is then written as the output into the sentiment state field of the session data structure.

[0414] Step 5: The server integrates user characteristics and constraints.

[0415] The server reads attribute information, preference information, and behavioral history information related to the current user from the storage system, and uses this historical feature data, along with the parsing results and sentiment state in the current session, as input. The server processes the behavioral history data, such as counting the number of times the user adopted different types of solutions, historical evaluation scores, and common activity themes, to generate feature vectors or preference label sets. The server combines these features with current budget information, time constraints, and preference conditions to form a unified constraint information object. This constraint information object is written as output into the session data structure, providing conditional input for generating prompt statements.

[0416] Step 6: The server generates a prompt statement based on the task type.

[0417] The server takes structured data objects (intent, entity, sentiment state, constraint information) as input, and the prompt generation module selects a matching template or strategy. The server transforms the task description, user characteristics, sentiment state, and output requirements into natural language text, forming prompts to drive the generative artificial intelligence model.

[0418] For example, when the intent is to generate a date plan, the server generates the following prompt as output: "You are a dating and romance planning expert. The user has a budget of approximately 5,000 yen and wants to plan a date with their partner that involves being close to nature. The user usually prefers outdoor and low-cost activities. Please design 2-3 specific date plans in and around the city. Each plan should include: the location of the activity, the main activities, the estimated cost breakdown (transportation, food, entrance fees, etc.), and a suitable time of day. Please answer in simplified Chinese and keep the total word count under 300 characters." Step 7: The server inputs the prompts and necessary context into the generative AI model and retrieves the generated results.

[0419] The server takes the prompt from step 6, along with the user's original question or context text if necessary, as input data, encapsulates it into a model input request, and calls the generative AI model interface module. The server sends the prompt to the generative AI model via network or local process calls. This model is typically a multi-layer neural network based on a transformer structure, internally performing data operations such as attention calculation and positional encoding. The model calculates the probability distribution of the next word for each output position and generates an autoregressive response text or plan description text. The server receives the complete generated text returned by the model as output and writes it to the "Model Original Output" field of the session data structure.

[0420] Step 8: The server performs post-processing and structure transformation on the generated results.

[0421] The server takes the natural language text output by the generative artificial intelligence model as input and processes it, including filtering sensitive words, formatting, and segmenting paragraphs. For planning-related results, the server further uses regular expressions, keyword matching, or lightweight parsing algorithms to split the text into structured fields, such as location names, activity lists, and cost details. For question-and-answer-related results, the server can extract key sentences and operational steps. The server outputs the cleaned and potentially structured results, including both "displayable text" and "machine-processable structure," and stores them in the session data structure.

[0422] Step 9: The server adjusts the sorting and selection of results based on feedback or historical information.

[0423] The server takes the candidate result list obtained in step 8, along with the user's historical selection and evaluation information, as input. The server performs data processing, such as calculating a comprehensive score for each candidate result, taking into account: the degree of matching between the generated text and the current intent, the degree to which constraints are met, consistency with the user's historical preferences, and historical evaluation scores. The server sorts the candidate results according to the scores and may remove low-scoring results. The sorted list of preferred and alternative results, along with relevant metadata information, is prepared as output and sent to the terminal.

[0424] Step 10: The server sends the final result to the terminal.

[0425] The server takes the result object obtained in step 9 as input, including the main text content, structured fields, and sorting information. The server encapsulates the result data, constructs a response object, and sends it to the terminal via a network protocol. The output includes: the response text or plan text displayed to the user, structured data usable by terminal components (e.g., fields for generating buttons and labels), and possible next-step suggestions.

[0426] Step 11: The terminal receives and displays the results.

[0427] The terminal receives the server's response object as input from the communication module, parsing its text content and structured data. Through the interface rendering module, the terminal displays the response text or plan as speech bubbles, cards, or lists on the screen. For example, the terminal might display the appointment plan's title, location, activity details, and estimated cost in separate columns, or present customer service responses as selectable steps. Simultaneously, the terminal generates actionable components based on the structured data, such as buttons for "Accept this plan," "Regenerate," and "View more details." The terminal's output is an updated user interface and interactive controls that the user can click.

[0428] Step 12: Users select or rate based on the displayed results.

[0429] Users use the results displayed on the terminal as input, read the response or plan, and make selections by clicking buttons or entering text, such as "Adopt the plan," "Regenerate the budget to 8000 yen," or "This suggestion is helpful." Users can rate the results, select "like" or "dislike," or enter evaluation text. User actions are recorded in the terminal as selection or evaluation information, which is then packaged as output data and sent back to the server. This data is used in subsequent sessions as additional conditions to add new prompts, thereby enabling continuous optimization and personalization of system behavior.

[0430] 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.

[0431] 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.

[0432] 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.

[0433] 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.

[0434] 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.

[0435] Second Implementation Method Figure 3 An example of the configuration of the data processing system 210 according to the second embodiment is shown.

[0436] 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.

[0437] 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).

[0438] 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.

[0439] 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.

[0440] 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).

[0441] 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.

[0442] 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.

[0443] 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.

[0444] 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).

[0445] 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.

[0446] 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".

[0447] 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.

[0448] 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.

[0449] 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.

[0450] 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.

[0451] 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.

[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 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.

[0454] 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.

[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 glasses 214.

[0456] Third Implementation Method Figure 5 An example of the configuration of the data processing system 310 according to the third embodiment is shown.

[0457] 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.

[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 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.

[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 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.

[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 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.

[0466] 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.

[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 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".

[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 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.

[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 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.

[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 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.

[0475] 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.

[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 head-mounted terminal 314.

[0477] Fourth Implementation Method Figure 7 An example of the configuration of the data processing system 410 according to the fourth embodiment is shown.

[0478] 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.

[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] 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.

[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, 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).

[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] 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.

[0485] 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.

[0486] 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.

[0487] 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.

[0488] 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.

[0489] 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".

[0490] 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.

[0491] 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.

[0492] 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.

[0493] 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.

[0494] 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.

[0495] 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.

[0496] 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.

[0497] 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.

[0498] 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.

[0499] 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.

[0500] 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.

[0501] 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.

[0502] 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).

[0503] 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.

[0504] 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."

[0505] 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.

[0506] 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).

[0507] 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.

[0508] 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.

[0509] 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.

[0510] 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.

[0511] 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.

[0512] 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.

[0513] 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.

[0514] 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.

[0515] 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.

[0516] 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.

[0517] In addition, the following notes are provided in response to the above explanation.

[0518] Example 1 (Note 1) An information processing system, characterized in that it comprises: A device for acquiring user attribute information, preference information, and behavioral history information, and storing the information as standardized structured data; An apparatus for taking the stored user attribute information, preference information, and behavioral history information as input information, generating prompt statements for a generative artificial intelligence model, and performing inference processing of the generative artificial intelligence model based on the prompt statements and the input information to calculate the feature quantities of user interests and preferences. An apparatus for performing similarity calculations on the feature quantities of other objects or service objects based on the calculated feature quantities, selecting combined candidates according to the similarity, and sorting the combined candidates according to the user to generate output information; An apparatus for acquiring communication content sent and received between users or between a user and a service object, and inputting the communication content along with prompts for the generative artificial intelligence model for parsing, and extracting interpersonal relationship-related topics and emotional states from the communication content. An apparatus for dynamically adjusting the content or order of the combined candidates based on the relationship between the extracted topics and emotional states and the combined candidates, generating prompts for the generative artificial intelligence model to generate suggestions related to interpersonal relationships, and providing the suggestions to the user. An apparatus for re-acquiring the user's response to the combined candidates or the suggestion information as behavioral history information, and using the behavioral history information as input information for learning or reasoning of the generative artificial intelligence model, thereby updating at least one of the feature quantities of the interests and preferences, the combined candidates, and the suggestion information.

[0519] (Note 2) The information processing system according to Appendix 1 is characterized in that, The device for generating prompt statements is configured to automatically generate the prompt statements using at least one of user attribute information, preference information, behavioral history information, and communication content as elements, and output the prompt statements as natural language instructions to the generative artificial intelligence model to comprehensively indicate multiple processing purposes such as the estimation of interests and preferences, the generation of combined candidates, and the generation of suggestion information related to interpersonal relationships.

[0520] (Note 3) The information processing system according to Appendix 1 is characterized in that, The device for acquiring behavioral history information is configured to acquire evaluation operation information and browsing operation information for other objects or service objects as behavioral history information, and update the prompt statements for the generative artificial intelligence model based on this behavioral history information, thereby automatically optimizing the generation of the combined candidates and the generation of the suggestion information over time.

[0521] Application Example 1 (Note 1) An information processing system, characterized in that it comprises: A computing unit is used to communicate with an information storage unit that stores user attribute information and user past behavior records. The computing unit is configured to obtain the user's attribute information and past behavior records from the information storage unit based on user identification information, and use the attribute information and past behavior records as feature data for a generative artificial intelligence model to perform learning or reasoning, and generate a prompt statement to instruct the generative artificial intelligence model to perform the learning or reasoning. A unit for receiving location information calculated by the terminal device based on radio or wireless signals sent by a location information acquisition device arranged in a physical space, corresponding the location information with area information in the physical space, and generating a prompt statement for indicating that the area information is included as input in the generative artificial intelligence model. The processing conditions of the recommendation model are used to perform judgment processing on the set of items or services of candidate objects based on the user's attribute information, past behavior records and the area information to select the target object, and the processing conditions are specified to the generative artificial intelligence model to generate a unit for generating a prompt statement to instruct the recommendation model to perform the selection processing of the items or services of the candidate objects; A unit is used to input the identification information and attribute information of the candidate objects or services selected by the recommendation model as explanatory data into the generative artificial intelligence model, and to generate a prompt statement for instructing the generative artificial intelligence model to generate a proposal content to be provided to the user in natural language. A unit for sending the proposal content generated by the generative artificial intelligence model as notification information to the terminal device; A unit for acquiring selection operation information made by the user in relation to the display of the notification information from the terminal device, appending the selection operation information as the user's behavior record to the information storage unit, and generating a prompt statement to indicate that the selection operation information is used as learning data for the recommendation model and the generative artificial intelligence model.

[0522] (Note 2) The information processing system according to Appendix 1 is characterized in that, The computing unit is configured to update the weighting or priority conditions in the selection process of the candidate items or services based on the user's behavior records and the region information, and generate a prompt statement to indicate that the updated weighting or priority conditions are provided as learning conditions to the generative artificial intelligence model.

[0523] (Note 3) The information processing system according to Appendix 1 is characterized in that, The computing unit is configured to input user input text obtained from the terminal device as parsing object data into the generative artificial intelligence model, generate a prompt statement to automatically generate a response text for the user's inquiry based on the user's attribute information, past behavior records and the regional information, and send the automatically generated response text to the terminal device.

[0524] Example 2 (Note 1) An information processing system, characterized in that it comprises: A device for recording user attribute information and user behavior information based on communication information from user terminals and user identification information; A device for constructing prompt statements for input to a generative artificial intelligence model using the user attribute information, the user behavior information, and the communication information; A device for inputting the prompt statement into the generative artificial intelligence model to instruct the generative artificial intelligence model to generate multiple response candidates for the communication information, and for acquiring the multiple response candidates; A device for associating the acquired plurality of response candidates with the user terminal, generating structured data, and sending the structured data to the user terminal; An apparatus for updating user behavior information based on user selection information of response candidates from the user terminal, and for adjusting the construction of the prompt statement or the generation conditions of the response candidates.

[0525] (Note 2) The information processing system according to Appendix 1 is characterized in that, The device for constructing prompt statements is configured to perform language parsing processing on the user attribute information, the user behavior information, and the communication information, and based on the parsing results, explicitly record the tone, length, content type, and at least one response strategy in the prompt statement as constraints, thereby generating prompt statements for controlling the generative artificial intelligence model to generate the response candidate behavior.

[0526] (Note 3) The information processing system according to Appendix 1 is characterized in that, The apparatus for acquiring multiple response candidates is configured to perform content determination processing on response texts acquired from the generative artificial intelligence model, filter the response texts based on determination results regarding politeness, security, or suitability, and assign at least one label and evaluation value to the filtered response texts to generate the structured data.

[0527] Application Example 2 (Note 1) An information processing system, characterized in that it comprises: This is used to acquire attribute information, preference information, and behavioral history information of the user, use the information as input data, and generate prompt statements to instruct the generative artificial intelligence model to perform parsing processing on the input data, so that by inputting the prompt statements into the generative artificial intelligence model, the generative artificial intelligence model can accurately select the processing unit of the object candidate or plan scheme that is suitable for the user. Used to acquire dialogue information or transmit information from a communication terminal, convert the dialogue information or transmit information into structured data through natural language processing, and generate prompt statements based on the structured data to instruct the generative artificial intelligence model to generate response text or suggestion text, so as to generate the response text or suggestion text containing information exchange support content through the generative artificial intelligence model, and output it to the processing unit of the communication terminal. It is used to specify the emotional state of the user subject using an emotion-specific model or emotion analysis function, and generate prompt statements containing emotional condition information based on the specified emotional state, so as to instruct the generative artificial intelligence model to adjust the content of matching candidates, suggested content or plan schemes, and output the adjusted result information generated based on the prompt statements to the processing unit of the communication terminal. Used to obtain inquiry information or question information from the user, and generate prompt statements containing inquiry content, stylistic conditions and output constraints, so as to instruct the generative artificial intelligence model to automatically generate solution information or guidance information corresponding to the inquiry information or question information, and send the answer information generated based on the prompt statements to the processing unit of the communication terminal. The processing unit is used to acquire constraint information including the user's budget information, time conditions, and hobby conditions, and to generate prompt statements to instruct the generative artificial intelligence model to automatically generate activity plans, transaction processes, or interpersonal communication plans based on the constraint information, so as to generate multiple plan candidates through the generative artificial intelligence model and prompt them to the communication terminal. The processing unit is used to obtain the selection information or evaluation information of the user subject sent by the communication terminal, and to include the selection information or evaluation information as additional condition information in the regenerated prompt statement, so as to instruct the generative artificial intelligence model to update the generation and processing of subsequent matching candidates, response text, suggestion text or plan scheme based on the regenerated prompt statement.

[0528] (Note 2) According to the information processing system described in Appendix 1, the processing unit is configured to generate a prompt statement based on the behavioral history information and emotional state of the user subject, for instructing the generative artificial intelligence model to automatically determine the priority or prompting order of matching candidates to be prompted to the user subject, and to dynamically update the matching candidates by inputting the prompt statement into the generative artificial intelligence model.

[0529] (Note 3) According to the information processing system described in Appendix 1, the processing unit is configured to acquire input data including transaction-related information, settlement condition information, and the emotional state of the user, and generate a prompt statement to instruct the generative artificial intelligence model to propose at least one of electronic settlement methods, preferential measures, or fee reduction methods based on the input data, so that by inputting the prompt statement into the generative artificial intelligence model, the generative artificial intelligence model can provide the user with guidance information on settlement methods or expenditure plans corresponding to their emotional state.

Claims

1. An information processing system, characterized in that, include: processor, The processor is configured to: Generate prompts to instruct generative AI models to analyze a user’s profile, interests, and past behavioral patterns in order to provide more accurate matches; Generate prompts to instruct generative AI models to parse conversations and messages in order to provide more effective communication support; as well as Generate prompts to instruct generative artificial intelligence models to recognize user emotions and adjust matching candidates based on those emotions.

2. The information processing system according to claim 1, characterized in that, The processor is configured to generate prompts that instruct the generative artificial intelligence model to automatically generate matching candidate objects based on the user's behavioral patterns.

3. The information processing system according to claim 1, characterized in that, The processor is configured to generate prompts that instruct generative artificial intelligence models to automatically generate answers to questions from users.

Images