Information processing system

Abstract

The application provides an information processing system. An information processing system, characterized by comprising: a processor, wherein the processor is configured to: generate prompt information indicating analysis of user interest and skills; generate prompt information indicating clarification of user needs; and generate prompt information indicating analysis of user emotional state and making suggestions accordingly.

Description

Technical Field

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

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

[0003] In existing technologies, systems designed for business communication and collaboration often suffer from the following problems: First, servers struggle to acquire and analyze user interests and skills information in a timely and systematic manner, resulting in incomplete user profiling and hindering the provision of accurate collaboration suggestions or matching results. Second, users often express their business needs in vague and abstract ways, lacking clear descriptions of the target audience, skill range, and industry sector. Existing systems typically lack effective interaction mechanisms to gradually clarify and structure user needs, preventing the system from fully utilizing users' true requirements in subsequent recommendations or matching. Third, users' needs and emotional states change dynamically during platform use, but traditional systems often rely solely on static profile information and simple behavioral statistics, lacking automatic analysis and feedback mechanisms for user emotional states and psychological tendencies. Consequently, they cannot provide appropriate communication suggestions or interaction methods based on the user's current emotional state. Therefore, existing technologies struggle to achieve comprehensive intelligent analysis of user interests, skills, needs, and emotional states, failing to provide personalized, contextualized, and high-quality business communication support. The objective of this invention is to provide a system that can automatically generate multiple types of prompts and improve the quality of user needs expression and enhance matching and suggestion effectiveness through comprehensive analysis of user interests, skills, needs, and emotional states. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides an information processing system comprising a processor configured to perform various generation and analysis processes. Specifically, the processor is configured to generate prompts instructing the analysis of user interests and skills. These prompts guide a generative artificial intelligence model or other analysis modules to analyze and model the user's interest and skill data, thereby forming a relatively complete and structured user capability profile. The processor is also configured to generate prompts instructing the clarification of user needs. Through multiple rounds of interaction with the user, the system gradually breaks down the user's vague business needs into structured requirement elements that can be processed by the system. Furthermore, the processor is further configured to generate prompts instructing the analysis of the user's emotional state and making suggestions accordingly. By analyzing the emotional features in the user's text, voice, or other interactive content, the system infers the user's current emotional state and generates corresponding communication strategies or action suggestions based on this emotional state. According to one embodiment, the processor invokes a generative artificial intelligence model to perform semantic understanding and feature extraction on the user's descriptions of interests and skills, thereby automatically identifying the user's professional direction, skill proficiency, and potential areas of interest. According to another embodiment, the processor analyzes users' behavioral history and user profile information to uncover potential connection needs exhibited by users in different time periods and scenarios, including a sustained focus on a certain type of partner, a specific industry theme, or a particular regional market. This provides a reliable basis for subsequent automatic matching and personalized recommendations. Through the coordination of the above structure and processing flow, the system of this invention can uniformly realize the analysis of user interests and skills, the clarification of user needs, and the analysis and suggestion generation of user emotional states within a single platform. This effectively improves the system's depth of understanding of user needs and the quality of its response, thereby improving the overall effectiveness of business communication and cooperation matching.

[0005] "System" refers to a computer implementation device consisting of at least one processor and optional memory, network interface and input / output modules, used as an overall technical solution carrier for performing the various data processing, prompt generation and analysis functions described in this invention.

[0006] A "processor" is a computing unit that can execute program instructions to complete data processing and control flow. It can be a single physical processor, multiple physical processors, a central processing unit (CPU), a graphics processing unit (GPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or any combination thereof.

[0007] "Prompt information" refers to textual information, tagged information, or structured instructions generated by the processor and provided to generative artificial intelligence models, other analysis modules, or users to indicate the analysis tasks to be performed, interaction goals, or processing content. These include, but are not limited to, prompts for analyzing interests and skills, prompts for clarifying user needs, and prompts for sentiment analysis and suggestion generation.

[0008] "User interests" refer to a user's sustained attention or preference in a specific field, topic, industry, or activity. These interests can be filled in by the user, inferred by the system based on behavioral records, or identified by a generative artificial intelligence model from the user's expressed content.

[0009] "User skills" refers to the abilities and experience a user possesses in one or more professional, technical, or business fields, including but not limited to programming skills, industry knowledge, and analytical abilities, which can be represented through user self-reporting, resume information, or system analysis results.

[0010] "User needs" refers to the goals or problems that users hope to achieve through the system, including but not limited to finding business partners, obtaining industry information, and expanding their network of contacts. Their initial expression can be a vague or abstract text description, which can be transformed into structured demand elements after being guided by the system.

[0011] "User emotional state" refers to the emotional tendencies and psychological states of a user at a specific point in time or in a specific interactive scenario, including but not limited to positive, negative, neutral, anxious, excited, and tired states, which can be obtained by the system through sentiment analysis of the user's text, voice or other interactive data.

[0012] "Generative artificial intelligence models" refer to machine learning models that can generate text, vector representations, or other forms of output based on input prompts. These include, but are not limited to, deep learning-based language models and multimodal models. They are used to analyze user interests and skills, understand and expand needs, or provide features and intermediate results for subsequent processing.

[0013] "Behavioral history" refers to the collection of records related to a user's usage behavior in the system, including but not limited to login behavior, browsing pages, participating in activities, searching keywords, clicking content, sending messages, etc., which are used to reflect the user's actual operation preferences and interest trajectory.

[0014] "User profile information" refers to the collection of static or semi-static information related to identity and characteristics maintained by the system for each user, including but not limited to basic personal information, professional background, industry affiliation, interest tags, skill list, geographical region, etc.

[0015] "Potential connection needs" refer to needs that have not been explicitly stated by users or recorded by the system, but can be inferred through comprehensive analysis of user behavior history, user profile information, and trends in interest changes. These needs are related to establishing business partnerships, networking, or information exchange relationships. Attached Figure Description

[0016] Figure 1 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the first embodiment.

[0017] 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.

[0018] Figure 3 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the second embodiment.

[0019] 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.

[0020] Figure 5 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the third embodiment.

[0021] Figure 6 This is a conceptual diagram illustrating an example of the main functions of the data processing device and head-mounted terminal according to the third embodiment.

[0022] Figure 7 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the fourth embodiment.

[0023] 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.

[0024] Figure 9 This represents an emotion map that maps multiple emotions.

[0025] Figure 10 This represents an emotion map that maps multiple emotions.

[0026] Figure 11 This is a sequence diagram illustrating the processing flow of the data processing system of the first embodiment.

[0027] Figure 12 This is a sequence diagram illustrating the processing flow of the data processing system in Application Example 1.

[0028] Figure 13 This is a sequence diagram illustrating the processing flow of the data processing system of the second embodiment.

[0029] Figure 14 This is a sequence diagram illustrating the processing flow of the data processing system in Application Example 2. Detailed Implementation

[0030] Hereinafter, an example of an implementation of the system according to the present disclosure will be described with reference to the accompanying drawings.

[0031] First, let me explain the terminology used in the following instructions.

[0032] 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.

[0033] 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.

[0034] 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.

[0035] 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.

[0036] 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.

[0037] First Implementation Method Figure 1 An example of the configuration of the data processing system 10 according to the first embodiment is shown.

[0038] 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.

[0039] 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).

[0040] 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.

[0041] 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.

[0042] 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.

[0043] 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.

[0044] Figure 2 The diagram shows an example of the main functions of the data processing device 12 and the smart device 14.

[0045] 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.

[0046] 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).

[0047] 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.

[0048] 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.

[0049] 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."

[0050] The following technical problems exist in existing information exchange platforms and user matching systems: (1) On the user side, interest information and skill information are usually input in the form of unstructured text. The server usually only performs simple storage and keyword retrieval, which makes it impossible for computing devices to efficiently extract features and perform similarity calculations on a large amount of user data, resulting in low utilization of computing resources and unstable matching results.

[0051] (2) On the server side, most of the user matching logic is based on pre-fixed rules or simple similarity threshold judgment. It lacks dynamic adaptive capability based on feedback information during operation and cannot automatically adjust the matching algorithm according to user behavior history, attribute characteristics and interaction feedback, making it difficult to optimize the calculation process for different data distributions and usage scenarios.

[0052] (3) In terms of human-machine collaboration, existing systems often use generative artificial intelligence models as independent modules for dialogue with users, without tightly coupling them with underlying data structure management, feature generation, similarity calculation and notification strategy control. This results in the output of generative artificial intelligence models being unable to directly participate in the automatic optimization of the computing process, and the scalability and self-learning ability of the computing architecture are limited.

[0053] (4) At the user interaction level, the generation of notification strategies and communication support information usually relies on fixed templates or simple logic, making it difficult to personalize the generation based on the user's emotional state, changes in needs, and the discovery of potential connection needs. This prevents the server from making full use of the calculation results to improve the quality of interaction, thereby limiting the contribution of the overall computing system to communication efficiency and platform activity.

[0054] Therefore, a computer system architecture is needed that can implement the following within the server: structured management of user interest and skill information, efficient calculation of similarity between users, dynamic optimization and selection of connection candidates, and direct use of the response feedback generated by the generative artificial intelligence model to adjust feature generation and notification content. This would improve the adaptability, scalability and overall performance of the computing process, thereby improving the processing capabilities of computer technology itself in user matching and communication support tasks.

[0055] 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.

[0056] In this invention, the server includes a processing device control unit for receiving interest information and skill information input via a terminal and acquiring them as text information; a data storage management unit for managing a storage area for storing the text information in structured information form and storing interest information and skill information therein; a feature calculation unit for generating feature quantities for calculating similarity between users and calculating similarity indices based on the stored interest information and skill information; a candidate selection unit for extracting user combinations that meet predetermined conditions based on similarity indices and selecting them as connection candidates; a notification control unit for generating and presenting notification messages related to connection candidates via a terminal; a prompt generation unit for generating prompt statements that instruct the generative artificial intelligence processing unit to perform interest and skill analysis, demand clarification, and emotional state analysis based on interest information, skill information, similarity indices, and connection candidate information; and a strategy update unit for updating connection candidate selection conditions and notification message content based on response information output by the generative artificial intelligence processing unit and generating communication support information to promote user communication. This allows for the formation of a closed-loop dynamic optimization mechanism within the computer. When performing operations such as data structure management, feature generation, similarity calculation, and notification control, the server can adaptively adjust the calculation process and parameter configuration using feedback from the generative artificial intelligence model. This improves the accuracy and efficiency of user matching calculations, reduces computing resource consumption, and enhances the overall technical performance of the computer system used for user matching and communication support tasks.

[0057] A "system" refers to a collection of devices and programs consisting of multiple information processing components that can electronically perform data reception, storage, calculation, control, and output processing to achieve user matching and communication support functions.

[0058] The "processing unit control unit" refers to the processing unit used to uniformly control and coordinate the various functional modules in the system. It controls the computational logic of data reception, storage management, feature calculation, candidate selection, notification generation, and interaction with the generative artificial intelligence processing unit by executing program instructions.

[0059] "Terminal" refers to an electronic device operated by a user to access a system, including but not limited to computer devices, mobile communication devices, or tablet devices, used to present a user interface and send and receive data with the server.

[0060] "Users" refers to entities that register or use services in the system, including individuals or organizations, who are participants who provide interest and skill information and connect and communicate as matching partners.

[0061] "Interest information" refers to descriptive data that characterizes the user's interest in a specific field, topic, or activity. It can be represented in the form of text or tags and is used to reflect the user's focus.

[0062] "Skills information" refers to descriptive data that characterizes the user's ability, experience, or knowledge level in a certain field. It can be represented in the form of text or labels and is used to reflect the ability that the user can provide or hopes to improve.

[0063] "Text information" refers to data consisting of character sequences that can represent interest information, skill information, or other relevant descriptive content, and is used as input for subsequent structured processing and feature extraction.

[0064] "Structured information" refers to information that is organized and stored according to a predetermined data structure or format, with field divisions and data type definitions, and is used to represent interest information and skill information in a searchable and computable manner in the storage area.

[0065] "Storage area" refers to the data space provided by storage media for persistently storing text and structured information, including databases, data tables, or other data storage units.

[0066] The "Data Storage Management Department" refers to the functional component used to manage the storage area. It is responsible for controlling the writing, updating, querying, and associated storage of data in order to achieve the structured preservation of interest information and skill information.

[0067] "Feature quantity" refers to numerical data generated by the system based on interest information and skill information, used to represent the features of the user subject in mathematical space, and used as input for similarity calculation.

[0068] "Similarity index" refers to a numerical value or score calculated based on feature quantities, used to quantify the degree of similarity between two or more users in terms of interests or skills, and used as the basis for selecting connection candidates.

[0069] The “feature calculation unit” refers to the functional component used to generate feature quantities from stored interest and skill information and calculate similarity indicators based on the feature quantities. It processes the data by performing numerical calculations or statistical analysis.

[0070] "Selection criteria" refers to a set of rules, thresholds, or logic used to determine whether a combination of user entities is identified as a candidate for connection, and is used to filter out matching objects that meet the expected criteria based on similarity indicators.

[0071] "Connection candidates" refers to combinations of users selected from multiple users based on similarity indicators and selection criteria, who are considered suitable as partners for mutual connection and communication.

[0072] "Candidate Selection Unit" refers to a functional component used to extract and determine connection candidates from all users based on similarity indicators and selection criteria.

[0073] "Notification message" refers to message content used to inform users of the existence of connection candidates and related information, including text, links, or other information units that can be displayed on the terminal.

[0074] The "Notification Control Unit" refers to the functional component used to generate, manage, and control the sending and presentation of notification messages. It constructs notification messages based on connection candidate information and provides them to the user through the terminal.

[0075] The “Generative Artificial Intelligence Processing Unit” refers to an artificial intelligence processing unit that can automatically generate output content such as text based on input prompts. It analyzes input information and outputs response information through machine learning models.

[0076] "Prompt statements" refer to instructional text sent to the generative AI processing unit, which describes the analysis tasks, generation goals, or output forms that the generative AI processing unit is expected to perform in natural language or structured statements.

[0077] The "prompt generation department" refers to a functional component that automatically constructs prompt statements based on interest information, skill information, similarity indicators, and connection candidate information, and provides them to the generative artificial intelligence processing department.

[0078] "Response information" refers to the result data output by the generative artificial intelligence processing unit based on the prompt statements, including analysis conclusions, suggestions, explanatory text, or other text information used to optimize the system processing flow.

[0079] The "Strategy Update Department" refers to a functional component used to adjust and update selected conditions, feature generation methods, similarity calculation methods, and notification message content based on the response information output by the generative artificial intelligence processing department.

[0080] "Communication support information" refers to auxiliary information generated by the system to facilitate communication and collaboration among users, including dialogue suggestions, introductory texts, personalized guidance content, or other outputs to improve communication efficiency.

[0081] "Behavioral history information" refers to data that reflects the operation records or usage trajectory of the user in the system, including access records, interaction records, response records, etc., which are used to assist in calculating similarity indicators and mining potential connection needs.

[0082] "Attribute information" refers to static or semi-static data that describes the basic characteristics of the subject, including category, domain, experience level, etc., and is used to conduct comprehensive analysis in combination with interest information and skill information.

[0083] "Potential connection needs" refer to potential needs that have not yet been explicitly expressed by the subject, but can be inferred through a comprehensive analysis of interest information, skill information, behavioral history information, and attribute information, and are suitable for establishing new connections or communication relationships.

[0084] In one embodiment of the invention, the server is configured to run on one or more computing devices, including a general-purpose computer with a central processing unit, semiconductor storage devices, and a network interface. The server runs application server software on top of an operating system (e.g., a Unix-like operating system) and deploys software components that implement the functions of the invention thereon. The server uses a relational database management system (e.g., a database server employing a relational query language) as its data storage infrastructure and uses data processing libraries, numerical computation libraries, and machine learning libraries (e.g., data table processing libraries, vector computation libraries, machine learning frameworks) as the software foundation for feature generation and similarity calculation. The server can also communicate with an external generative artificial intelligence server via a network interface, which performs inference processing on the generative artificial intelligence model.

[0085] In one embodiment, the server connects to multiple terminals via a network. The terminals include computing devices running general-purpose operating systems, such as mobile communication terminals or desktop computing devices. The terminal's browser or local application is responsible for presenting the user interface and exchanging data with the server via a secure version of Hypertext Transfer Protocol (HTTP). Users operate input and display devices through the terminals to provide the server with information of interest and skills, and receive notification messages and communication support information from the server.

[0086] In one embodiment, the server loads the software program of this invention upon startup. This program is divided into multiple functional modules, including: a request receiving module, a data verification module, a data storage management module, a feature calculation module, a similarity evaluation module, a candidate selection module, a notification control module, a prompt generation module, a policy update module, and an external generative artificial intelligence communication module. These modules can collaborate within the same process via function calls, or they can communicate as independent services in a distributed architecture via a remote procedure call protocol.

[0087] In one implementation, the server receives interest and skill information sent by the terminal through a request receiving module. This information is encapsulated in the form of text data within an application-layer request. The server uses a data verification module to perform length checks, character set checks, and basic semantic verification on the received string data to prevent meaningless or malicious input. After successful verification, the server transfers the interest and skill information to the data storage management module.

[0088] In one implementation, the server uses a data storage management module to convert unstructured text into structured information for interest and skill data. The server defines a database table to store user information, interest entries, skill entries, and matching results between users. The server associates the identifier of the corresponding user with the interest tag field and skill tag field. This structured organization allows the server to perform efficient block-based read and write operations between memory and secondary storage, reducing the amount of irrelevant data scanning and thus improving data management performance.

[0089] In one implementation, the server generates feature quantities from the interest and skill information stored in the database using a feature calculation module. First, the server batch-reads the interest and skill descriptions of all or part of the users from the database, converting multiple text entries into standardized text through string concatenation, stop word removal, case neutralization, or word segmentation. Then, the server uses a text vectorization algorithm (e.g., a term frequency-inverse document frequency vectorization method) to map the text representation of each user to a high-dimensional real-valued vector space. The server constructs a sparse matrix structure in memory using a numerical computation library to represent the feature set of all users, facilitating subsequent batch similarity calculations using matrix operation instructions.

[0090] In one implementation, the server performs similarity calculations on a set of features using a similarity evaluation module. The server utilizes vector similarity metrics (such as cosine similarity) to perform inner product operations and norm normalization operations on the feature vectors of any two users in a high-dimensional space, thereby generating a similarity index matrix. The server can leverage highly optimized matrix multiplication and vector operation instructions through a linear algebra library, fully utilizing the instruction set extensions and caching mechanisms of the underlying hardware to achieve high-speed batch computation. This batch matrix operation method significantly reduces computation time and increases throughput when processing large-scale user data compared to traditional iterative computation methods.

[0091] In one implementation, the server filters the generated similarity index matrix through a candidate selection module. The server can set different selection criteria for different users, such as minimum similarity threshold, maximum number of candidates, and rules for excluding already connected users. The server uses a sorting or heap-based algorithm in memory to select several connection candidates from the similarity matrix for each user that rank highly in similarity and meet the threshold criteria. The server stores these connection candidates in a database as structured records for subsequent notification and statistical analysis. Because the filtering process is performed on a numerical matrix, the server can utilize vectorized comparison operations and batch sorting algorithms, reducing a large number of string comparisons compared to traditional methods based on plain text matching, thus improving overall computational efficiency.

[0092] In one implementation, the server generates notification messages based on connection candidate records through a notification control module. The server combines basic information about the candidate objects (such as summaries of interest keywords and skill keywords) with a fixed template to generate notification content adapted to different users. The notification control module can generate messages in different formats for different terminal types, such as short text or rich text formats. The server sends these notification messages to the terminals via the network, and the terminals display these messages in their user interface, indicating to the user that the system has found new potential communication partners.

[0093] In one implementation, the server generates multiple types of prompt statements through a prompt generation module and sends them to the generative AI processing unit where the generative AI model resides. Based on interest information, skill information, similarity metrics, and connection candidate information, the server constructs the following types of prompt statement examples: "Based on the following description of the user's interests and skills, analyze which types of communication or collaboration this user is best suited for: interests include 'data analysis' and 'machine learning,' and skills include 'Python programming' and 'statistical modeling.' Please provide a concise analysis result." "Based on the following similarity scores and interest / skill information between users, explain why we recommend them to communicate with each other, and write a recommendation explanation in plain language: User A's interests are 'data analysis' and 'Python programming', and User B's interests are 'data analysis' and 'visualization'. The similarity score is 0.92." "Based on the following dialogue and operation records, analyze the user's current needs and confusion, and clearly express their needs in one or two sentences." "Based on the user's self-description and recent interaction records, please determine their emotional state (e.g., positive, hesitant, under a lot of pressure), and provide appropriate communication suggestions." The server takes the aforementioned prompt as the body of the input request and sends it to an external generative AI service via a network interface. This service internally runs a generative AI model. This model employs a neural network structure with multi-layered attention mechanisms, using multi-head self-attention layers, feedforward network layers, residual connections, and layer normalization to construct a deep model, and has been pre-trained on a large-scale corpus. Based on the response information returned by the interface, the server parses and persists the analysis text, explanatory text, or suggested text output by the generative AI model as response information.

[0094] In one implementation, the server updates the system's internal control parameters using response information returned by the generative artificial intelligence model through a policy update module. The server can adjust the construction of feature vectors based on the model's analysis conclusions, such as assigning higher weights to specific interest words or skill words, reflecting this in the parameter settings of the text vectorization algorithm. The server can also change the similarity threshold or the ranking priority of connection candidates based on the recommendation reasons provided by the model. The server stores these policy parameters in the database as configuration records. The feature calculation module and candidate selection module read these updated parameters during the next execution, thus forming a closed-loop optimization mechanism. Because this update process directly affects matrix calculations and the database index structure through numerical parameter adjustments and data structure changes, the server can reduce invalid comparisons and redundant candidates in subsequent matching operations, achieving the technical effect of improving accuracy and computational efficiency.

[0095] In one embodiment, the server can also fine-tune the generative AI model using historical data provided by the server. The server records the difference between the suggested text output by the generative AI model and the actual user behavior, and constructs a feedback signal using an error function (such as cross-entropy loss or mean squared error) to update some of the model's weight parameters. In this case, the generative AI model uses a gradient descent-type optimization algorithm (such as an adaptive learning rate algorithm) to adjust the network weights through backpropagation, enabling the model to produce outputs that better conform to the matching logic of the target system when processing similar prompts in subsequent iterations. This learning process gradually aligns the model's behavior with the specific matching pattern of this invention, thereby further improving the overall system performance.

[0096] In one implementation, the server, by introducing a collaborative mechanism between the aforementioned generative artificial intelligence model and prompt statements, transforms the computation process from a simple automated execution of manually generated rules into an adaptively evolving computational structure based on both numerical computation and neural network inference. Internally, the server employs explicit data structures (such as sparse matrices and index tables), specific algorithmic processes (such as batch vector operations, heap sort filtering, and dynamic parameter updates), and neural network learning mechanisms to achieve higher matching accuracy and lower response time under the same hardware conditions. Simultaneously, by reducing unnecessary data transmission and redundant comparison operations, the server also optimizes load in terms of network communication and storage access.

[0097] In one implementation, the terminal acts as the interface between the user and the server, responsible for rendering the user interface and performing basic input validation. The terminal parses the hypertext and style documents obtained from the server using a browser engine and executes scripting languages ​​to construct the interactive interface. The terminal locally formats the user's input of interest information, skill information, and selection results for recommended objects, and then sends these locally to the server. The terminal is also responsible for presenting the communication support information returned by the server to the user in a human-computer interface format, such as displaying recommendation reasons and communication suggestions generated by a generative artificial intelligence model, helping the user to more efficiently understand and utilize the system's matching results.

[0098] In one implementation, users actively input descriptions of their interests and skills via a terminal, browse a list of connection candidates automatically provided by the system, and choose whether to accept the recommendation or begin communication. While this behavior appears to be part of a business process from an application perspective, in this invention, the user's input is strictly mapped to the server's internal data structure updates and parameter tuning processes. The user's choices and feedback are incorporated into the feature calculation and strategy update modules through behavioral history and attribute information, thus directly driving the computer's internal algorithms. Through this feedback loop, this invention enables the computer system to continuously adapt and optimize for specific application scenarios, rather than simply automating manual decision-making, thereby improving data management methods, similarity calculation efficiency, and the quality of matching results from a technical perspective.

[0099] In other implementations, the server can use a vector database or graph database to replace or supplement the relational database, representing the connections and similarity edge weights between subjects using a graph structure. This allows for the use of graph algorithms (such as graph-based clustering or community detection algorithms) to further uncover potential connection needs. The server can also employ different feature representation methods, such as encoding interests and skills as multidimensional discrete features, inputting them into a feedforward neural network or graph neural network for embedding learning, and directly outputting similarity indices or connection probabilities through an end-to-end representation learning mechanism. As long as the server internally achieves numerical processing and matching optimization of interest information, skill information, and other relevant information through explicit data structures, algorithmic flows, and parameter control, it can be considered a modified embodiment of the present invention.

[0100] Through the various implementation forms described above, the server, terminal, and user collaborate within the system constituted by this invention, enabling the acquisition, storage, analysis, and matching of interest and skill information to form a highly efficient data flow and control flow structure. Relying on a dynamic strategy update mechanism combining generative artificial intelligence models and prompt statements, this invention achieves improved accuracy, faster processing speed, reduced errors, and optimized resource utilization in user matching tasks at the computer technology level, rather than merely mechanically replacing the manual matching process.

[0101] use Figure 11 The processing flow is explained.

[0102] Step 1: Users access the platform and load the interface using a terminal.

[0103] Users access the Uniform Resource Locator (URL) provided by the server through a browser or application on their terminal, requesting the homepage of the information exchange platform. The input is the user's access action in the address bar or application, and the output is the Hypertext Markup Language (HTML) document, style sheets, and script resources received by the terminal. The terminal parses the received document, executes the script, and renders a page containing graphical interface elements such as interest input boxes, skill input boxes, and submit buttons.

[0104] Step 2: Users input their interests and skills on the terminal.

[0105] Users type descriptions of their interests and skills in the text input boxes on the terminal interface, or select preset interests and skills by clicking on the label components. Input consists of keyboard input events and touch selection events, while output is a set of text strings stored in the terminal's memory. The terminal performs basic format validation on this string set locally, such as checking for empty strings and whether the length exceeds a set limit, and then displays an error message or allows the user to continue on the interface.

[0106] Step 3: The terminal sends interest and skill data to the server.

[0107] When the user clicks the submit button, the terminal packages the locally cached interest text and skill text along with the user identifier into request data. The input is the text string and user identifier generated by the user in step 2, and the output is the application-layer request message sent to the server via the network interface. The terminal calls the network request interface in the script, sends the request using a secure transport protocol, and enters a state of waiting for the server's response locally.

[0108] Step 4: The server receives and validates user input data.

[0109] The server receives request messages from the terminal on the network listening port, and the application processing module parses them to extract the user identifier, interest text, and skill text. The input is a request message containing text fields, and the output is a structured request object represented in the server's memory. The server performs validation operations on this request object, including checking if required fields exist, if field lengths meet range limits, and if character encodings are valid. Based on the validation results, the server generates a Boolean value. If validation fails, it constructs a response object containing error information and returns it to the terminal; if validation succeeds, it passes the structured data to the subsequent data storage module.

[0110] Step 5: The server converts interests and skills into structured information and stores it.

[0111] The server extracts interest text and skill text from the validated request object and converts them into a predefined field structure in a database table. The input is a user identifier and the original text description; the output is a set of data record insertion or update operations to be performed. Based on the user identifier, the server determines whether to create a new record or update an existing one, and generates corresponding data operation commands. It decomposes the interest text into multiple interest items and the skill text into multiple skill items, associates them with the user identifier, and writes them into a relational data table. The server executes these data write and update operations, and generates a confirmation status and returns it to the terminal upon successful storage, thus completing the conversion from unstructured text to structured records.

[0112] Step 6: The server reads user information in batches from the database for feature calculation.

[0113] When a predetermined time or trigger condition is met, the server reads all or a portion of the users' interest and skill entries from a relational database. The input is a structured set of records in the database, and the output is a collection of user data in memory, such as a list or table object. The server aggregates multiple interest and skill entries corresponding to the same user identifier into a single logical record and prepares these logical records as input for text vectorization and feature calculation.

[0114] Step 7: The server preprocesses the interest and skill text.

[0115] The server processes the text of interests and skills in each user record. The input is the user data set obtained in step 6, and the output is a standardized set of text representations. The specific data processing performed by the server includes: concatenating multiple interest and skill entries into a unified string, removing redundant whitespace and punctuation, segmenting or cutting words according to preset rules, filtering stop words, and standardizing case and character format. Through these text preprocessing operations, the server normalizes the original description to reduce noise features and provide a unified format for subsequent numerical processing.

[0116] Step 8: The server maps the preprocessed text into numerical feature vectors.

[0117] The server invokes a text vectorization algorithm to convert each user's standardized text into a high-dimensional numerical vector. The input is the set of text representations output from step 7, and the output is a feature vector matrix. The server counts the frequency of each term in all documents and calculates statistics such as term frequency and inverse document frequency. Then, it calculates the weights corresponding to each term in each user's text and fills them into the corresponding positions in the sparse matrix. Through this data processing, the server encodes discrete text information into real-number vectors that can participate in linear algebra operations, enabling similarity calculations to be performed efficiently through matrix operations.

[0118] Step 9: The server calculates similarity metrics between users.

[0119] The server uses a numerical computation library to perform batch similarity calculations on the feature vector matrix. The input is the feature vector matrix generated in step 8, and the output is a similarity matrix or a list of similarities between users. The server performs an inner product calculation on any two rows of feature vectors, calculates the norm for each row, and normalizes the inner product result to obtain the cosine similarity value. The server stores these values ​​in appropriate data structures, such as two-dimensional arrays or association lists, to represent the proximity of each pair of users in the interest and skill spaces.

[0120] Step 10: The server filters connection candidates based on similarity metrics.

[0121] The server reads the similarity matrix and filters it according to predefined selection criteria. The input is the similarity data structure and user set obtained in step 9, and the output is a set of candidate connection pairs. For each user, the server finds other users whose similarity exceeds a threshold, sorts them by similarity from high to low, and selects the top few as candidates. During the filtering process, the server can also exclude user pairs with established connections or blacklisted users. For each candidate user pair, it generates a record containing the identifiers of both parties and the similarity value, and stores these records in the matching result data table.

[0122] Step 11: The server generates and sends connection candidate notification messages.

[0123] The server reads newly generated connection candidate records from the matching result data table and constructs notification content for each user. The input consists of the connection candidate pairs and their basic information records; the output is a set of notification message objects to be sent. The server integrates the candidate users' interest keywords, skill keywords, and similarity information into a text template to form a personalized notification message, which is then sent to the corresponding user's terminal via the message transmission module. Upon receiving the notification, the terminal displays it on its interface, informing the user that the system has recommended new communication partners.

[0124] Step 12: The server generates prompts for analysis and explanation.

[0125] The server constructs various prompts based on current interest information, skill information, similarity metrics, and connection candidate information. The input includes a set of user profiles, matching results, and relevant contextual data; the output is a set of natural language prompts. The server generates texts for different purposes, including prompts for interest and skill analysis, prompts explaining recommendation reasons, prompts clarifying needs, and prompts for sentiment analysis. For example, "Based on the following description of a user's interests and skills, analyze which types of communication or collaboration this user is suitable for." The server saves the generated prompts in a structured format and prepares to send them to the generative artificial intelligence model.

[0126] Step 13: The server sends prompts to the generative artificial intelligence model and receives response information.

[0127] The server, through its external communication module, packages the prompt statement generated in step 12 into a request and sends it to the processing end running the generative AI model. The input consists of the prompt statement text and necessary context parameters, while the output is the response text returned by the generative AI model. Upon receiving the response, the server parses it, extracting the analysis conclusions, recommendation reasons, demand summaries, and sentiment judgments to form a structured response object, which serves as the basis for subsequent strategy updates and notification content optimization.

[0128] Step 14: The server updates the matching strategy and feature calculation parameters based on the response information.

[0129] The server reads suggestions related to weight adjustment, threshold adjustment, or ranking preferences from the structured response object generated in step 13. The input is the response information of the generative AI model and the current configuration parameters; the output is the updated set of policy parameters. Based on the suggestions, the server adjusts the weight allocation rules in the text vectorization stage, such as increasing the weight of keywords in certain domains and decreasing the weight of frequent but low-discrimination terms. The server can also adjust the similarity threshold or the maximum number of candidates per user. The updated parameters are written to configuration storage or cache for use in subsequent feature calculation and candidate selection processes, thereby achieving feedback-based adaptive optimization.

[0130] Step 15: The server generates communication support information based on the response information and sends it to the terminal.

[0131] The server uses analysis results and suggested text provided by a generative artificial intelligence model to construct communication support information. The input consists of structured response objects and connection candidate records; the output is user-oriented explanatory text and communication suggestions. The server combines information such as recommendation reasons, descriptions of shared interests, and suggested opening remarks into one or more segments of natural language text, which are then sent to the corresponding user's terminal via a notification control module. The terminal presents this communication support information in the matching interface or chat interface, assisting users in initiating conversations and establishing collaborations more efficiently.

[0132] Step 16: Users can initiate or conduct communication by referring to support information on the terminal.

[0133] Users view the server-pushed list of connection candidates and communication support information on the terminal interface, and choose to accept or ignore the recommendations based on their own preferences. Input consists of user clicks on terminal interface elements and text input, while output consists of new session requests or chat content. The terminal sends these requests and content to the server, which further creates session records and forwards messages in the session management module. Through this process, the server transforms its complex internal data processing, feature calculations, and generative artificial intelligence model inference results into concrete communication session establishment and message routing behaviors, enabling the entire system to provide technical support and optimization for the communication process in the real world.

[0134] 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".

[0135] Existing computer-based information exchange platforms typically recommend other users or information resources to users only through simple keyword matching or rule-based recommendations. These systems suffer from the following technical shortcomings: (1) On the user side, users’ interest and ability information are often stored as static tags. The server side does not perform structured modeling and similarity calculation optimization on these attribute information, resulting in low matching accuracy between users, low efficiency of computing resources utilization, and difficulty in efficiently retrieving suitable collaborating partners in a large user set.

[0136] (2) During the dialogue, when the user’s needs are expressed vaguely, incompletely or semantically unclearly, the server usually only passively forwards the message and lacks the ability to automatically analyze and refine the vague needs in combination with the conversation context. As a result, the backend processing flow cannot make full use of the existing dialogue history data, and the user often needs to manually try to express himself multiple times, resulting in low interaction efficiency.

[0137] (3) When introducing generative artificial intelligence models, most existing systems simply send the user's single input directly to the model to generate the result, without dynamically constructing prompts on the server side based on user attribute information and conversation history information. This results in the generative artificial intelligence model's reasoning ability not being fully utilized, and the generated content lacking relevance to the user's real needs and collaborative scenarios, easily leading to redundant or incompatible outputs.

[0138] (4) Many existing technologies only use the large amount of attribute data and dialogue history data accumulated during system operation for statistical analysis or offline recommendation, and do not use them to infer potential relationship needs online or expand the collaborative candidate set in real time. Therefore, they cannot continuously explore new high-value connection opportunities in the underlying computing process, and the system lacks adaptability in matching strategies and resource scheduling.

[0139] Therefore, a technical solution is needed to uniformly manage user attribute information on the server side, efficiently calculate user similarity, automatically extract and decompose fuzzy needs in the dialogue history, and dynamically generate prompts for generative artificial intelligence models. This solution aims to improve the accuracy and efficiency of user matching at the computer system level, enhance the relevance and usability of dialogue generation based on generative artificial intelligence models, and automatically infer potential relationship needs through the analysis of historical data, thereby improving the overall computing architecture and processing performance of the information exchange platform.

[0140] 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.

[0141] In this invention, the server includes: a device for controlling the acquisition and storage of user attribute information; a device for calculating the similarity between users and extracting multiple collaboration candidates based on the user attribute information; a device for sending relevant information of the multiple collaboration candidates to a communication terminal and presenting it to the user through the communication terminal; a device for managing bidirectional communication between the user and the collaboration candidates and establishing synchronous dialogue in a virtual space; a device for generating prompt statements input to a generative artificial intelligence model based on the user attribute information and the content of the bidirectional communication, and using the prompt statements to indicate the concretization of the user's needs and the construction of dialogue content; a device for sending dialogue text candidates contained in the generation results obtained from the generative artificial intelligence model to the communication terminal and enabling the user to use the dialogue text candidates as communication content sent to the collaboration candidates by selecting or editing them; and a device for parsing the user attribute information and the historical information of the bidirectional communication to infer potential relationship needs and determine new collaboration candidates or information resources based on the potential relationship needs. This allows for the formation of an integrated processing flow within the server, encompassing user attribute modeling, similarity calculation, session management, and generative artificial intelligence model invocation. On one hand, structured storage and similarity calculation enhance the accuracy and scalability of user matching operations. On the other hand, the automatic generation of targeted prompts significantly improves the matching degree between the output content of the generative artificial intelligence model and the user scenario. Furthermore, it utilizes dialogue history and attribute data to infer potential relationship needs and dynamically updates collaboration candidates and information resources, thereby improving the processing efficiency, intelligence level, and resource utilization of information exchange services at the computer system level.

[0142] A "system" refers to a collection of information processing devices consisting of one or more processing devices, storage devices, and communication devices, which work together to execute processing procedures such as user matching, session management, and generative artificial intelligence model invocation.

[0143] A "server" refers to a computing device or node in a system that is responsible for centrally storing user attribute information, performing similarity calculations, managing two-way communication sessions, and calling generative artificial intelligence models.

[0144] "Communication terminal" refers to an electronic device operated by a user for inputting and displaying information and interacting with a server through a communication network, including but not limited to mobile terminals, fixed terminals, or wearable terminals.

[0145] "User attribute information" refers to a set of data used to characterize user features, including interest information, ability information, and various information related to user features extracted from user behavior.

[0146] "Interest information" refers to data in user attribute information that indicates a user's level of interest or preference in a certain field, topic, or activity.

[0147] "Capability information" refers to data in user attribute information that represents a user's proficiency or ability in a certain field, skill, or knowledge.

[0148] "Similarity" refers to a numerical value or indicator that quantifies the degree of similarity between different users in terms of interests, abilities, or other characteristics based on user attribute information.

[0149] "Collaboration candidates" refer to at least a portion of the set of target users selected from multiple users based on similarity calculation results, who have a high relevance to the target user and are considered suitable for information exchange or collaborative activities.

[0150] "Virtual space" refers to a digital interactive environment built through communication networks and information processing devices that is independent of physical location, in which multiple users can communicate or collaborate in the form of data.

[0151] "Two-way communication" refers to a communication method in virtual space in which two or more communication entities send and receive information to each other, where each party can act as both a sender and a receiver.

[0152] "Synchronous dialogue" refers to a dialogue format in which participants send and receive messages and interact continuously in near real-time during two-way communication.

[0153] "Generative artificial intelligence models" refer to artificial intelligence models trained using machine learning or deep learning methods that can automatically generate text, speech, or other forms of output based on input data.

[0154] "Prompt statements" refer to textual information provided as input to generative artificial intelligence models to limit the content, form, and contextual conditions of the generation task.

[0155] "Dialogue text candidates" refer to one or more selectable text contents output by a generative artificial intelligence model based on prompts, which are suitable as messages or statements sent by the user in a conversation.

[0156] "Historical information" refers to time-series data accumulated and stored during system operation that is related to user behavior, session content, or attribute changes.

[0157] "Vague demand expression" refers to user demand information that is described in a vague and abstract way in the dialogue or input content without clearly stating specific goals, conditions or scope.

[0158] "Potential relationship needs" refer to needs that users may need to establish in the future for relationships, cooperation, or information acquisition, which have not yet been explicitly stated by users but can be inferred by analyzing user attribute information and historical information.

[0159] "Information resources" refer to structured or unstructured data, documents, service interfaces, or functional modules that are available for users to access or utilize, in order to meet users' needs in the process of information exchange or collaboration.

[0160] The embodiments of this invention will be described in detail, combining specific hardware and software structures, data structures, and algorithm flows, to provide a user matching and dialogue assistance system based on a generative artificial intelligence model and prompt statements. It should be noted that the following embodiments are not intended to limit the scope of the claims and can be appropriately modified according to specific application scenarios.

[0161] I. System Overall Structure In a typical implementation, a server consists of a general-purpose computer or a cloud computing node, including a multi-core central processing unit, optional graphics processing unit, large-capacity random access memory, and non-volatile storage media. The server runs a general-purpose operating system, such as Unix-like operating systems, and runs middleware and service programs, including: - Web application frameworks (such as Python-based service frameworks); - Database management system (e.g., relational database management system); - Generative artificial intelligence model inference services (such as deep learning inference engines deployed locally on servers).

[0162] In one embodiment, the terminal is a mobile terminal or a head-mounted display terminal, comprising a processor, memory, display module, input module, and network communication module. The terminal runs a mobile operating system and has client applications installed, which communicate with the server via a network.

[0163] Users interact with the server through a terminal, inputting user attribute information, browsing collaboration candidates, initiating dialogues, and selecting or editing dialogue text candidates provided by the server. The server returns the processing results to the terminal via the network, enabling users to conduct synchronous dialogues in virtual space.

[0164] II. User Attribute Information and Data Structure In one implementation, the server categorizes user attribute information into at least the following categories: - Basic identification information: Identifiers used to identify the user; - Interest information: The topics that users are interested in are represented by tags, short texts, or domain codes; - Ability Information: User ability characteristics are represented by fields such as skill tags, proficiency level, and years of experience; - Behavioral information: including browsing history, project participation history, historical matching results, etc.

[0165] The server stores the aforementioned data in a relational database. In one example architecture, the server stores interest and ability information in a normalized form across multiple related tables. This structured storage allows the server to directly vectorize and index discrete labels when calculating similarity, thereby reducing storage redundancy and improving query efficiency.

[0166] In one variant implementation, the server can store some attribute information in the form of a document database or key-value store to store unstructured data such as long text self-introductions and project descriptions, and convert it into vector features through a text feature extraction module during similarity calculation.

[0167] III. Similarity Calculation and Collaborative Candidate Extraction In order to efficiently select collaboration candidates from a large set of users, the server uses specific feature vector encoding and similarity calculation methods.

[0168] In one implementation, the server maps interest information and capability information to a one-dimensional or multi-dimensional feature space: - The server pre-builds a tag dictionary and assigns a unique index to each interest tag and ability tag; - The server constructs a sparse vector for each user, with the vector dimension corresponding to the length of the label dictionary. The value is 1 for a label that appears and 0 for a label that does not appear. The server can further assign weights to different tags, such as increasing the weight of tags that users frequently use or that the system identifies as high-value.

[0169] The server can use at least one of the following two methods to measure similarity: 1. The server uses cosine similarity to measure the angle between user vectors, thereby reflecting the consistency of interests and abilities. 2. The server uses set similarity to measure the ratio of the intersection and union of labels.

[0170] In a preferred embodiment, the server uses a vector index structure to index user feature vectors, reducing the overhead of calculating similarity on large user sets. By constructing this type of index structure, the server can retrieve user sets with high similarity to the target vector in sublinear time, thereby significantly reducing server processing time and CPU utilization, and accelerating matching in high-concurrency scenarios.

[0171] By using the aforementioned feature modeling and indexing structure, the server enables the matching algorithm to move beyond simple keyword comparison and instead quantify the similarity between users through numerical vector operations, thereby achieving a combined improvement in matching accuracy and computational efficiency at the technical level.

[0172] IV. Two-way Communication and Session Management In one implementation, the server runs a real-time communication service to manage bidirectional communication between users and collaboration candidates.

[0173] The server assigns a unique session identifier to each session and records metadata such as participant identifiers, session creation time, and status flags in the storage medium. The server maintains a message list structure for each session to store fields such as message content, timestamp, sender identifier, and read status.

[0174] When managing bidirectional communication, the server maintains a connection mapping table for each online terminal. When the server receives a message from a terminal, it looks up the mapping table to determine the target receiving terminal connection and forwards the message with low latency. By maintaining long-term connections, the server reduces the overhead of repeatedly establishing connections, thereby lowering round-trip latency and increasing message delivery rate.

[0175] Through the aforementioned session management mechanism, the server can continuously accumulate historical information from two-way communication, providing a reliable data foundation for extracting fuzzy demand expressions and inferring potential relationship demands from history.

[0176] V. Generative Artificial Intelligence Models and Prompt Statement Construction In a core embodiment of this invention, the server deploys a generative artificial intelligence model. In a preferred embodiment, this model is a language model based on a deep neural network, comprising: - The server uses multiple embedding layers to encode the input natural language segments or sub-words into fixed-dimensional vectors; - The server uses a multi-layer self-attention network to encode the input sequence, capturing long-range dependencies and contextual information; - On the output side, the server predicts the next word sequentially using an autoregressive decoding mechanism, forming a variable-length natural language output.

[0177] When training this type of generative artificial intelligence model, the server can use a large-scale text corpus and adopt a supervised learning approach: - The server defines a cross-entropy loss function to measure the difference between the model's predicted distribution and the true distribution; - The server uses gradient descent-type optimization algorithms for parameter updates; - During training, the server can employ data augmentation strategies, such as randomly masking parts of the input or shuffling non-critical word order, to improve the robustness of the model.

[0178] During the inference phase, the server does not directly send the user's raw input to the generative artificial intelligence model, but instead constructs structured prompts based on user attribute information and conversation history.

[0179] In a typical implementation, the prompt statement constructed by the server includes the following elements: - User's interests and abilities; - Some attribute information of the collaboration candidates; - Vague expressions of needs extracted from conversation history; - Explicit instructions for the generation task (e.g., number of generation, tone, language type, etc.).

[0180] The following is an example of a prompt message constructed by the server: - Example 1: "You are a conversational assistant. There is now a user whose interests are 'data analysis' and 'machine learning', and whose skills are Python and SQL."

[0181] This user has just started a conversation with a collaborator who is experienced in the field of data analytics.

[0182] Please generate 5 Chinese prompts for this user to use in initiating communication with collaborators.

[0183] Each prompt should be concise, polite, and convey a willingness to cooperate. - Example 2: "You are a conversational assistant. Below is the user's basic information:" Interests: Data analysis, machine learning Skills: Python, SQL In a recent chat, a user said: 'I only know I want to do analysis related to financial data, but I don't know what specific project I want to work on.' Based on this information, please generate 5 Chinese prompts to help users further clarify project requirements with their collaborators.

[0184] Each prompt should be a question or explanation that the user can directly send to their collaborator. By constructing prompts in this way, the server enables generative AI models to comprehensively consider user attributes and dialogue context when generating candidate dialogue texts, rather than relying solely on one-time input. This structured prompting method technically improves the relevance and specificity of the model's output, reduces redundant text generation, and thus increases the proportion of effective output with the same computing resources, indirectly reducing the average latency of inference services.

[0185] VI. Extraction and Gradual Concretization of Fuzzy Requirement Expressions In one implementation, the server parses historical information from bidirectional communication and automatically detects and extracts fuzzy requirement expressions.

[0186] The server can process the dialogue text using a combination of rule-based and statistical methods: - The server performs sentence-level segmentation on historical text; - The server uses a pre-trained sequence labeling model or classification model to label each sentence and identify whether it contains ambiguous requirements; - The server can combine specific keywords and negative expressions as auxiliary features to improve detection accuracy, such as recognizing expressions like "not sure yet", "don't know exactly what to do", and "probably want to do".

[0187] After detecting a vague expression of demand, the server embeds it as specific tagged content into the prompt statement, guiding the generative artificial intelligence model to generate clarifying or detailed questions and suggestions for the vague part, thereby technically achieving the gradual convergence of demand abstraction to demand specificity.

[0188] By organically combining fuzzy expression extraction and prompt statement construction on the server side, the system of this invention can automatically convert unstructured dialogues into structured targets that can be further processed without increasing the user's operational burden. This processing flow is different from the traditional process of repeated manual questioning. Instead, it is an automatic semantic decomposition achieved within the computer through a combination of specific rules and models, thereby improving the intelligence and processing efficiency of the dialogue management module.

[0189] VII. Potential Relationship Needs Inference and Collaboration Candidate Expansion In one implementation, the server uses long-accumulated user attribute information and bidirectional communication history to infer potential relationship needs.

[0190] The server can perform cluster analysis or pattern mining on historical data: - The server clusters users in the attribute space to identify frequently occurring interest-ability combinations; The server analyzes which combinations are frequently associated with successful collaboration records; The server identifies skills or interests that are in high demand but low supply within the existing user base.

[0191] Based on the above analysis results, the server maps the inferred potential relationship needs back to the individual user level. For example, when a user's attributes are highly similar to a combination of high collaboration success rates, the server can prioritize recommending specific types of collaboration candidates or information resources to that user.

[0192] Because the server internally employs numerical statistics and clustering algorithms to perform offline or online calculations on large-scale historical data, the inference of potential relationship needs becomes an automatic pattern recognition process at the computer level, rather than solely based on manual rules. This technically enables adaptive reconstruction of the user relationship network, improving the dynamic adjustment capabilities of the matching engine and overall resource utilization.

[0193] VIII. Terminal Interaction and Rendering In one implementation, the terminal is responsible for displaying collaborative candidate information and dialogue text candidates provided by the server, and providing editing and sending interfaces.

[0194] After receiving the dialogue text candidates from the server, the terminal displays them as a clickable list near the dialogue input area. The user selects a prompt by tapping, and the terminal fills the text into the input box, which the user can then edit as needed. After the user confirms the submission, the terminal transmits the text to the server via the network, which then forwards it to the collaborating candidate terminals.

[0195] The terminal reduces repeated requests to the server and lowers network bandwidth usage through local rendering and caching mechanisms. At the same time, it refreshes only newly added or changed content through a differentiated update strategy, thereby further improving the user interface response speed and reducing the overall communication load on the client side.

[0196] IX. Technical Effects and Improvements in Computer Technology The server achieves the following technical effects through the aforementioned specific data structures, vectorization methods, similarity calculation strategies, prompt statement construction mechanisms, and generative artificial intelligence model invocation processes: - By uniformly encoding interest and ability information into sparse vectors and building an index, the server maintains low retrieval latency in high-concurrency scenarios, thereby accelerating the matching process and improving scalability. - By actively analyzing dialogue history and automatically extracting ambiguous demand expressions on the server side, the system transforms unstructured text that could not be directly used in numerical calculations into structured targets that can be further processed by the model, thereby improving the automation level and computational efficiency of the dialogue management module. - By constructing prompts that include user attributes, historical dialogues, and task instructions, the server can effectively limit the model's inference space to a highly relevant region when calling the generative artificial intelligence model, thereby reducing the probability of irrelevant outputs and improving the adaptability and usability of the generated results to the user's specific scenario. - By performing pattern mining on historical attributes and session data, the server can internally form an adaptive mechanism for inferring potential relationship needs, thereby achieving dynamic optimization at the matching strategy level, rather than just static rule matching.

[0197] The aforementioned improvements all occur within the computer system's data structure, algorithm flow, and model invocation method. They are not merely simple replacements of manual business processes, but rather, through new feature representation methods, index structures, and model interaction mechanisms, enable computers to exhibit higher processing speeds, better matching accuracy, and lower resource consumption when performing user matching and dialogue assistance tasks, thus constituting a substantial improvement to computer technology itself.

[0198] use Figure 12 The processing flow is explained.

[0199] Step 1: Users input user attribute information through the terminal. Users open the application interface on the terminal, enter text content in the interest input area and ability input area, and select the corresponding tags.

[0200] Input: Text containing information about interests and abilities that the user enters or selects on the interface.

[0201] The terminal reads the above text from the graphical interface component, performs preprocessing such as removing spaces, word segmentation, and splitting by delimiters on the string, and assembles the user identifier, interest list, and ability list into structured data.

[0202] Output: The terminal generates a structured data object containing user identifiers, a list of interest information, and a list of ability information, and prepares it to be sent to the server.

[0203] Step 2: The terminal sends user attribute information to the server. The terminal encapsulates the structured data generated in step 1 into a request message through the network communication module and sends it to the server through a preset interface.

[0204] Input: Structured data containing user identifiers, lists of interest information, and lists of ability information.

[0205] The terminal serializes the data, adds necessary protocol headers and authentication information, and sends it to the server's receiving interface via the communication link.

[0206] Output: The terminal outputs an attribute upload request message to the server over the network.

[0207] Step 3: The server receives and stores user attribute information. The server receives request messages from the terminal at the receiving interface and deserializes them into an internally processable data structure.

[0208] Input: The attribute upload request message sent by the terminal.

[0209] The server performs format validation, field integrity validation, and content length validation on the message. It then writes the interest and ability information into a data table or dataset in the storage device, including inserting or updating user attribute records. During this process, the server also uniformly encodes interest and ability tags and creates or updates corresponding indexes.

[0210] Output: The server generates updated user attribute data records and index structure, and returns storage result status information.

[0211] Step 4: The terminal receives the stored results and provides feedback to the user. The terminal receives the storage result status information returned by the server.

[0212] Input: Status information data returned by the server.

[0213] The terminal parses the status field and displays "Information saved" or an error message on the user interface based on the success or failure status, while also updating the locally cached user attribute information.

[0214] Output: The terminal outputs feedback information on the display module and maintains user attribute data consistent with the server in local storage.

[0215] Step 5: The user sends a request on the terminal to find collaborating partners. Users can select "Find Collaboration Partners" or a similar function entry on the terminal interface.

[0216] Input: User's operation instructions (button clicks or menu selections).

[0217] The terminal assembles the current user identifier and possible filtering conditions (such as domain preference, time range, etc.) into query request data.

[0218] Output: The terminal generates a query request containing the user identifier and optional filter conditions, and prepares to send it to the server.

[0219] Step 6: The terminal sends a collaborative candidate query request to the server. The terminal sends the query request generated in step 5 to the server's matching interface via the network.

[0220] Input: Query request data containing user identifiers and filter criteria.

[0221] The terminal performs data packaging and transmission control, sending the request message to the interface address specified by the server.

[0222] Output: The terminal outputs a collaborative candidate query request message to the server over the network.

[0223] Step 7: The server calculates similarity based on user attribute information. After receiving the query request, the server reads the attribute data of the target user and other users from the storage device.

[0224] Input: Query request data and a collection of user attribute data stored in the database.

[0225] The server first encodes the interest and ability information of all users, converting each user into a numerical feature vector. Then, using the target user vector as a benchmark, the server calculates the similarity score between the target user and other users using vector operations (such as cosine similarity or set similarity). Finally, the server sorts and filters according to the similarity scores to select multiple collaborative candidate users.

[0226] Output: The server receives a list of collaborative candidate users sorted by similarity and their corresponding similarity scores.

[0227] Step 8: The server sends a list of potential partners to the terminal. The server organizes the list of collaborative candidate users obtained in step 7, along with the necessary display information, into response data.

[0228] Input: List of collaborative candidate users, and attribute summary information for each candidate.

[0229] The server trims (e.g., limits the number of candidates), formats (e.g., adds field descriptions) the candidate list, and encapsulates it into a response message, which is then sent back to the terminal over the network.

[0230] Output: The server outputs a response message containing information about several candidate users for collaboration.

[0231] Step 9: The terminal displays collaboration candidates and receives user selections. The terminal receives the list of collaboration candidates returned by the server and displays it on the interface in the form of a list or cards.

[0232] Input: A response message containing information about potential collaboration users.

[0233] The terminal parses the name, description, interest tags, ability tags, and other information of each candidate to generate corresponding interface elements; after viewing the list, the user can click the "start a conversation" control of a candidate.

[0234] Output: The terminal displays a list of collaboration candidates on the display module and generates a user selection event.

[0235] Step 10: The terminal sends a session creation request to the server. After a user selects a target collaboration candidate on the terminal, the terminal generates a session creation request based on that selection.

[0236] Input: User-selected operation and identification information of the selected collaboration candidate.

[0237] The terminal assembles the current user identifier, the target collaboration candidate identifier, and possible initial description text into a session creation request message, and sends it to the server over the network.

[0238] Output: The terminal outputs a session creation request message directed to the server.

[0239] Step 11: The server creates a session and returns a session identifier. After receiving a session creation request, the server generates a new session record in the session management module.

[0240] Input: Session creation request message.

[0241] The server assigns a unique session identifier to each session, writes the identifiers, creation time, and initial state of both parties to the storage device, and updates the session mapping table in memory. After that, the server constructs response data with the session identifier and the other party's user profile information and returns it to both terminals.

[0242] Output: The server outputs a response message containing the session identifier and session participant information.

[0243] Step 12: The terminal establishes a two-way communication channel and enters the chat interface. The terminal receives the session identifier and the other party's user information returned by the server.

[0244] Input: A response message containing the session identifier and participant information.

[0245] If the terminal has not yet established a long connection, it initiates a long connection with the server (e.g., to maintain the connection of the session) and maintains the mapping of the session identifier to the connection channel locally; at the same time, the terminal switches to the chat interface to prepare to receive and send messages.

[0246] Output: The terminal establishes a network connection that can be used for bidirectional communication and presents a new dialog window in the user interface.

[0247] Step 13: Users send conversation messages via the terminal. Users enter text content in the chat input area of ​​the terminal, and then click send after editing or modifying it as needed.

[0248] Input: The text message entered by the user in the chat input area.

[0249] The terminal assembles the message text, session identifier, sender identifier, and local time information into a message data packet, and sends it to the server through the established communication channel.

[0250] Output: The terminal outputs a dialogue message to the server.

[0251] Step 14: The server forwards conversation messages and records historical information. After receiving the dialogue message data from the terminal, the server processes it in the session management module.

[0252] Input: Dialogue message data sent by the terminal.

[0253] The server writes the message content and related metadata into the session message storage structure to form a complete historical record; then, the server finds the other participating terminal connection in the session and forwards the message to it through the corresponding channel.

[0254] Output: The server generates and outputs a dialogue message forwarded to the other terminal, and appends new historical message records to the storage.

[0255] Step 15: The server extracts fuzzy demand expressions from historical information. When the number of session messages accumulates to a certain amount, the server performs text analysis on the historical message content.

[0256] Input: A sequence of stored session history message texts.

[0257] The server segments and annotates the message text, and uses pre-trained text classification or sequence labeling models and keyword rules to identify sentences containing vague requirements such as "not sure yet", "don't know what to do", and "probably want to do". These sentences are then extracted as vague requirement expressions and associated with the current session and the corresponding user.

[0258] Output: The server receives a set of text fragments expressing vague needs associated with the current user.

[0259] Step 16: The server constructs prompts for generative artificial intelligence models. The server constructs a prompt statement based on user attribute information, collaboration candidate attribute information, and the fuzzy requirement expression in step 15.

[0260] Input: Current user's interest and ability information, relevant attributes of collaboration candidates, and fuzzy requirement expression text.

[0261] The server combines the above information into a structured natural language description, including model role settings, user background descriptions, ambiguous requirements, and generation task descriptions, such as the number of items to be generated and tone requirements, thus forming a complete prompt statement.

[0262] Output: The server generates one or more prompt text statements to drive the generative artificial intelligence model.

[0263] Step 17: The server invokes a generative artificial intelligence model to generate candidate dialogue texts. The server sends the prompt generated in step 16 to the inference interface of the generative artificial intelligence model.

[0264] Input: A prompt text containing user background, session context, and instructions for generating the task.

[0265] The server segments and vectorizes the prompts in the inference engine, encodes them into an input sequence, and then performs forward propagation through a multi-layer neural network. Based on the self-attention mechanism, it calculates the context representation and predicts the output words in sequence until multiple dialogue text candidates that meet the length and format constraints are generated.

[0266] Output: The server obtains several candidate dialogue text strings related to the current scene.

[0267] Step 18: The server sends dialogue text candidates to the terminal. The server filters and organizes the candidate dialogue texts output by the generative artificial intelligence model.

[0268] Input: Multiple candidate texts output by a generative artificial intelligence model.

[0269] The server removes obviously duplicate or non-compliant candidates, retains several high-quality texts based on criteria such as length and semantic relevance, and encapsulates these texts together with the session identifier into a response message, which is then sent to the corresponding user's terminal over the network.

[0270] Output: The server outputs a prompt message containing multiple dialogue text candidates.

[0271] Step 19: The terminal displays dialogue text candidates and receives user selections. The terminal receives a dialog text candidate prompt message sent by the server.

[0272] Input: A prompt message containing multiple dialogue text candidates.

[0273] The terminal parses the candidate text and displays it to the user in the chat interface as a list or button. The user can click on any of the candidates to automatically fill the text into the input area, or edit the text before clicking send.

[0274] Output: The terminal displays selectable dialogue text candidates on the interface and generates new message content after the user selects or edits it.

[0275] Step 20: Users continue to communicate with collaborators using dialogue text suggestions. Users select or modify appropriate expressions based on candidate texts on the terminal and then send them to collaborating partners.

[0276] Input: The dialogue text content selected or edited by the user.

[0277] The terminal then sends the text, along with the session identifier, back to the server as message data. The server forwards and records the message as described in step 14. Through multiple iterations, the vague requirements are gradually specified, and the direction of collaboration becomes clearer.

[0278] Output: The terminal and the server collaboratively generate new dialogue message streams. The server continuously accumulates richer historical information for subsequent demand inference and collaborative candidate optimization.

[0279] 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.

[0280] 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."

[0281] In existing computer-implemented communication and matching platforms, the following technical problems typically exist when automatically processing abstract and vague user requests in natural language. First, the server side often only performs one-time keyword extraction or simple classification of user input, lacking a multi-round iterative clarification mechanism based on dialogue history. This results in the inability to stably converge low-resolution abstract requests into structured, computable specific requests within the machine, thus limiting the accuracy of subsequent retrieval and matching algorithms. Second, even when existing systems introduce generative artificial intelligence models, they often use these models as "direct answerers" rather than engineering the steps of "prompt statement generation," "dialogue state management," and "reuse of model output questions" within the server's internal program structure. This lack of adaptive control mechanisms for the content of prompt statements and the amount of contextual information leads to uncontrollable model calls, high resource consumption, and fluctuating output quality. Third, existing candidate matching is usually based on static user tags or simple rules, roughly mapping the requirement text to a small number of tags before retrieval. It does not implement an integrated data processing pipeline on the server side from "abstract requirements → multi-round question and answer → specific requirements → candidate selection", which makes it difficult for the matching algorithm to make full use of the context information and session state information accumulated in multi-round dialogues, and it is difficult to effectively discover potential collaborators.

[0282] Therefore, there is a need for an improved computer implementation method and system on the server side. This would involve integrating technologies such as prompt generation, generative artificial intelligence model invocation, multi-turn question and answer management, conversation state-driven context control, and collaboration object selection based on specific needs into the information processing device. This would improve the computational efficiency, automation, and accuracy of the end-to-end process from vague user needs to precise matching results, thereby improving the natural language requirement processing and matching technology in the field of computer technology.

[0283] 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.

[0284] In this invention, the server includes: a unit for receiving and storing natural language abstract demand information obtained from a user device in an information processing device; a unit for generating prompt statements for inputting into a generative artificial intelligence model based on the abstract demand information, parsing the abstract demand information using the generative artificial intelligence model, and generating question information for concretizing the abstract demand information based on the parsing result; a unit for sending the question information to the user device, obtaining user response information to the question information through the user device, and storing the response information in association with the abstract demand information; and a unit for storing the response information based on the user response information. The system generates additional prompts for inputting into the generative AI model using historical records of answer information and question information, and utilizes the generative AI model to generate additional question information or specific requirement information to progressively concretize the abstract requirement information; it also has a unit for extracting collaboration object information from a data storage device storing a set of candidate object information with attribute information based on the specific requirement information, and selecting collaboration object information according to the correspondence between the specific requirement information and the collaboration object information; and a unit for sending the specific requirement information and the collaboration object information to the user device and outputting the results. This allows for the formation of a multi-turn natural language processing pipeline driven by session state information within the server. By leveraging the synergy between prompts and the generative AI model, the user's abstract requirements are progressively transformed into structured specific requirements in the computer, and candidate object information is then filtered with high precision. This improves the automation level of the requirement clarification process and the accuracy of matching results, thereby enhancing computer technology for natural language-based human-computer interaction and collaboration object recommendation.

[0285] "Information processing device" refers to a device that has computing capabilities and can execute programs to process data, including a single computing node or a computing system composed of multiple computing nodes, such as server devices, virtual machines, or cloud computing platforms.

[0286] "User device" refers to a terminal device used to interact with information processing devices and operated by a user, including electronic devices with input and display functions, such as smart terminals, computing terminals or communication terminals.

[0287] "Natural language abstract requirement information" refers to requirement information that users input in natural language form, which is relatively vague and lacks specific constraints or structured elements, and is used to express the user's initial intention or general goal.

[0288] "Generative artificial intelligence models" refer to artificial intelligence models based on machine learning algorithms that can receive inputs such as text and automatically generate new text outputs. They are typically trained on large-scale data to model and generate natural language.

[0289] "Prompt statements" refer to control text information provided as input to a generative artificial intelligence model to instruct the model to perform a specific generative task. The prompt statements contain task descriptions, contextual information, or constraints.

[0290] "Question information" refers to natural language text expressed in the form of a question, generated to clarify or refine abstract natural language requirements, and used to request supplementary information or more specific conditions from users.

[0291] "Additional prompts" refer to prompts generated based on existing dialogue history and response information, which are then input into the generative artificial intelligence model again. They are used to guide the model to generate additional questions or specific requirements.

[0292] "Additional question information" refers to subsequent question information generated based on existing answer information and question history, in order to further refine or supplement the information required for natural language abstraction.

[0293] "Specific requirement information" refers to requirement information that contains more explicit constraints and structured elements, obtained after multiple rounds of parsing and clarification of abstract requirement information in natural language, and is used to support subsequent retrieval or matching processes.

[0294] "Data storage device" refers to storage resources used to store various types of data in a read-write manner, including local storage devices, network storage devices, or distributed storage systems.

[0295] "Candidate object information set" refers to a set of data entries corresponding to multiple candidate objects stored in a data storage device, where each data entry contains at least attribute information used to represent the characteristics of the candidate object.

[0296] "Attribute information" refers to one or more fields of information used to describe the characteristics of candidate objects, including but not limited to category information, geographic information, scale information, business type information, or other feature information that can be used for matching.

[0297] "Collaboration object information" refers to information corresponding to one or more candidate objects selected from the candidate object information set based on specific requirement information and adapted to the user's needs, used to indicate objects that the user can collaborate with or connect with.

[0298] "Session state information" refers to the state data used to represent the current dialogue process and context during multi-turn interactions, including generated question information, obtained answer information, generated specific requirement information, and related time sequence information.

[0299] In one embodiment of the present invention, the server serves as the core computing node of the information processing device, and the terminal serves as the user device, through which the user interacts with the server. The server is typically composed of a computer system with a multi-core processor, main memory, persistent storage, and a network interface, such as a rack-mounted computing device equipped with a general-purpose central processing unit (CPU) and a graphics processing unit (GPU); the terminal can be a smart terminal or computing terminal running a mobile operating system or a desktop operating system, and is connected to the server via a wired or wireless communication network.

[0300] When implementing this invention, the server can run a general-purpose server operating system, such as a UNIX-like server operating system, and middleware software, such as a web server or application server. The server can use application frameworks to implement business logic, such as scripting language frameworks or object-oriented language frameworks. The server can also deploy a relational database management system or a document-oriented database system for persistently storing user request data, question information, answer information, session state information, and candidate object information.

[0301] When implementing this invention, the terminal can run a mobile terminal operating system or a desktop terminal operating system, and run a browser application or a native client application. The terminal provides the user with a text input area, a question display area, and a result display area through a graphical user interface, and sends and receives data to and from the server via network protocols.

[0302] In one implementation, the server receives abstract natural language request information from the terminal via a network interface. The server represents this abstract request information as a character sequence in memory and assigns a unique session identifier to each session in the database, establishing a data association between the abstract request information and user identification and time information. The server uses a record structure with fields in persistent storage, such as session identifier, user identifier, request text, and status fields, to facilitate subsequent retrieval and updates.

[0303] When generating a prompt to invoke a generative AI model, the server constructs a control text within its program. The server takes abstract requirement information as input and inserts it into the prompt template to form the complete prompt. In one example, the server can generate the following prompt: "You are a needs clarification assistant. The user's current needs are:" Looking for new business partners.

[0304] Please generate 2-3 specific clarification questions in Simplified Chinese. These issues should be addressed separately: industry or sector, form of cooperation or objective, and geographical scope.

[0305] The question should be brief and clear. In another example, the server can generate the following prompt: "The user says: 'I want to find new business partners.' Please analyze the user's needs and clarify them step by step with 2-3 specific questions, which should cover the industry, the form of cooperation, and the geographical region, respectively." The server constructs the model input text by combining the aforementioned prompts and abstract requirement text, and encodes the model input text into a token sequence. For this purpose, the server can use a tokenizer or sub-word encoder to map the input character sequence into a discrete token sequence for the generative artificial intelligence model to perform numerical calculations.

[0306] In a preferred embodiment, the server sets the generative artificial intelligence model as a language model based on a multi-layer Transformer architecture. This language model includes an embedding layer, a multi-layer self-attention network layer, and an output layer. The embedding layer maps each tag identifier to a fixed-dimensional vector; the self-attention network layer calculates the correlation between tags at different positions through a self-attention mechanism and generates a context-sensitive representation; the output layer calculates the probability distribution of the next tag based on the context representation. When the server invokes this model, it concatenates the prompt statement with abstract requirement information to form an input sequence, calculates the output tag sequence through forward inference, and the output tag sequence is decoded back into natural language text, which is the question information or specific requirement information.

[0307] To improve the stability and accuracy of model output, servers can set specific parameters when calling the model, such as maximum output length, temperature parameters, and sampling strategies. For example, setting the temperature parameter to a lower value reduces output randomness and makes the question information more stable; limiting the maximum number of output tags reduces unnecessary verbose output, thereby lowering network transmission load and computational resource consumption.

[0308] The server uses a string processing algorithm to identify question separators or number prefixes in the generated question information, splitting the model's output text into several independent questions. The server creates a record in the database for each question information, which includes a session identifier, question number, question text, and a status flag field. The status flag is used to indicate whether the question has been answered by the user.

[0309] After receiving the question information from the server, the terminal parses the received data locally and displays the question text in the question display area on the screen. The terminal can display the questions in a sequential list or as speech bubbles. Users input their answers to each question on the terminal via touch input, keyboard input, or voice-to-text conversion. The terminal pairs the answers with the corresponding questions and forms a data structure containing a list of answers locally.

[0310] When a terminal sends a response to a server, it transmits the response text along with metadata such as the session identifier and question number via network protocol. Upon receiving the response, the server writes it to the response field of the corresponding record in the database and updates the status flag field to "answered." The server also records a timestamp for subsequent analysis of session evolution based on the response time sequence.

[0311] When further clarification is needed, the server generates additional prompts based on stored question and answer information. When constructing these additional prompts, the server selectively incorporates session history information into the prompt text. For example, the server might generate the following additional prompts: "The user's original need was: 'I want to find new business partners.'"

[0312] The currently known information is as follows: - Industry: Primarily the IT industry; - Form of cooperation: We hope to cooperate on technology and jointly develop products; - Location: Ideally, partners should be located in the local city.

[0313] Please generate 1-2 more specific questions to further confirm the company size, budget range, or cooperation period, making the requirements more specific. When constructing the aforementioned appended prompts, the server does not simply concatenate the entire session history. Instead, it controls the content and amount of context based on session state information. The server can select the key information that most significantly impacts the specificity of the requirements as context, based on pre-defined rules or statistical characteristics. For example, it can retain only the latest or most recent question-and-answer summaries, thereby reducing input length, lowering model computational complexity, and minimizing noise caused by redundant context. This mechanism of controlling the amount of context information based on session state helps reduce model inference time and improves the relevance and conciseness of the output questions.

[0314] After obtaining sufficient response information, the server can integrate the abstract requirement information with all key answers and use a generative artificial intelligence model to generate specific requirement information. In one example, the server can construct the following prompt: "The user's original need was: 'I want to find new business partners.'"

[0315] The following information has been clarified: - Industry: Primarily the IT industry; - Forms of cooperation: technical cooperation and joint product development; - Location: We are looking for partners in the local city.

[0316] Please use 1-2 sentences to generate a clear and specific description of the requirements, so that any reader can accurately understand what kind of partner the user is looking for. The server inputs the above prompt along with relevant parameters into the generative artificial intelligence model, and obtains the specific requirement information through the model's output. For example, the model may output the following text: "Users are looking for business partners in the IT industry in their local cities, mainly for technical cooperation and joint product development." After obtaining specific requirements information, the server stores this information in a database and uses it as structured input for subsequent matching of collaborating partners. The server can pre-store a set of candidate partner information in a relational database or search engine, with each candidate partner possessing attributes such as industry, region, business type, and scale. During the matching phase, the server can employ a combination of algorithms: on one hand, filtering based on key fields to select partners by industry category and region; on the other hand, it can transform the specific requirements information into semantic vectors using a vectorization model, and similarly transform the candidate partner descriptions into vectors, ranking the candidates based on vector similarity calculations. The server can then use cosine similarity, dot product, or other distance metrics to rank the candidates, thereby selecting collaborating partners with higher matching degrees.

[0317] By employing the aforementioned multi-round prompt generation, generative AI model invocation, and contextual information control techniques, the server can significantly improve the convergence speed and accuracy from abstract to concrete requirements compared to methods that only use the model to generate results once. Because the server constrains the input text based on the session state and structurally restructures the output question in each model invocation round, it can obtain structured, information-dense concrete requirements with fewer model invocation rounds, thereby reducing overall computational resource consumption and the amount of data transmitted over the network.

[0318] When training generative AI models, servers can employ supervised or self-supervised learning methods, using training sets that include large-scale natural language dialogue data and task-oriented question-answering data. During training, the server measures the difference between the model's predicted labeled sequences and the target sequences by defining a loss function (e.g., cross-entropy loss) and updates the model parameters based on the gradient descent algorithm. The server can use batch training, dividing the training samples into mini-batches, calculating gradients and updating weights in each batch to improve training stability and efficiency. Data augmentation methods can be used during training, such as paraphrasing or structural rewriting text with the same intent, to enhance the model's robustness to different expressions.

[0319] Because the server employs a multi-turn question-and-answer-based requirement clarification pipeline and a controllable prompt generation strategy, it internally forms a non-traditional, human-written rule-writing process. Instead of simply mapping user input to labels using a fixed set of rules, the server dynamically generates questions through a generative artificial intelligence model, and the server-side control logic prunes and organizes the model's input and output. This processing method creates a new data flow topology within the computer, including: an abstract requirement information caching module, a prompt generation module, a model invocation module, a question and answer management module, a session state management module, and a collaboration object matching module. These modules cooperate through structured data interfaces, enabling the server to compress and structure high-dimensional natural language information in stages, achieving automatic transformation from free text to structured matching conditions.

[0320] In this embodiment of the invention, in addition to displaying questions and results, the terminal can also cache client-side data and provide input assistance. The terminal can locally cache the user's previous input history to reduce repetitive input in case of network anomalies. The terminal can provide input suggestions and historical answer recommendations, thereby reducing the amount of manual input by the user. However, these functions are only extensions to the user experience and do not affect the core algorithm logic on the server side.

[0321] In this invention, users input their initial needs in natural language, rather than pre-selected fixed options. Because the server uses a generative artificial intelligence model and multi-turn question answering to progressively decompose the user's text into computable attribute conditions, the user does not need to understand the internal data structure or retrieval syntax. In this way, the server encapsulates complex natural language parsing, semantic abstraction, and structured mapping within the model and program logic, thereby transferring user-side complexity to efficient algorithms and hardware resources on the server side.

[0322] The present invention achieves the following technical effects through the above technical solutions: The server reduces unnecessary model input and improves inference speed by controlling the content of prompt statements and the length of the context; the server improves matching accuracy by transforming noisy abstract requirements into less redundant structured requirements through multi-turn question-and-answer management; the server reduces the burden of long sequence inputs on the model through a context pruning mechanism driven by session state information, thereby achieving more stable output quality under the same hardware resources; and the server improves retrieval efficiency and shortens the time from when a user submits a request to when they obtain the result through a candidate object selection mechanism that combines vectorized matching and field filtering.

[0323] This invention can also have other variations. For example, the server can replace the generative artificial intelligence model with a language model with different parameter sizes to adapt to different latency and resource constraints; the server can store the candidate object information set in different types of data storage devices, such as columnar databases or vector databases, to optimize a certain type of retrieval task; the server can adjust the question generation strategy according to the application scenario, for example, prioritizing time constraints or budget conditions in some scenarios, while prioritizing spatial location or object quantity in other scenarios. Through the above various embodiments, this invention can improve the efficiency and accuracy of natural language processing on the server side through technical means.

[0324] use Figure 13 The processing flow is explained.

[0325] Step 1: Users input abstract requirement information at the terminal.

[0326] Users input natural language text, such as "Looking for new business partners," into the terminal's input interface via keyboard or touchscreen. The input is a sequence of characters typed by the user. The terminal caches this character sequence in a memory string variable and displays it on the interface in real time. The terminal performs basic checks on the string (whether it is empty, whether its length exceeds the limit). If the checks pass, the string is prepared as abstract request information and sent. The output is a valid natural language request text.

[0327] Step 2: The terminal sends abstract requirement information to the server.

[0328] The terminal takes user identifier, session identifier, and abstract requirement information as input, constructs a data structure containing multiple fields locally, and serializes it into text format. The terminal establishes an encrypted connection through the network protocol stack and sends this serialized text as the message payload to the designated interface of the server. Data processing involves encoding the structured fields in memory into a network transmission format. The output is a request message reaching the server.

[0329] Step 3: The server receives and stores abstract requirement information.

[0330] The server takes the request message received at the network layer as input, decodes it using the application layer parsing module, and restores it to a record containing the user identifier, session identifier, and abstract requirement text. The server performs length and character set validation on the abstract requirement text and writes records that pass the validation to a database table. Data processing involves extracting fields from the message and mapping them to database rows. The output is an abstract requirement record registered in persistent storage and its corresponding session state.

[0331] Step 4: The server generates the initial prompt statement.

[0332] The server takes abstract requirement text and session identifiers as input, and uses string concatenation logic to populate the abstract requirements into a predefined prompt statement template. For example, the server generates the following prompt statement: "You are a requirement clarification assistant. The user's current requirement is: 'Looking for new business partners.' Please generate 2-3 specific clarification questions in Simplified Chinese. These questions should focus on: industry or field, cooperation form or goal, and geographical scope. The questions should be concise and clear." Data processing involves embedding variable content into a fixed template to form complete control text. The output is a prompt statement that can be directly provided to a generative artificial intelligence model.

[0333] Step 5: The server converts the prompt statement into a model input vector.

[0334] The server takes the prompt text as input, calls the word segmentation or sub-word encoding module to decompose the text into a sequence of tokens, then maps each token to a numerical identifier, and further converts it into a real number vector through an embedding lookup table. Data computation includes: indexing using a lookup table, assembling the token IDs into a two-dimensional matrix for subsequent neural network computation. The output is a sequence of embedding vectors representing the prompt text.

[0335] Step 6: The server uses a generative artificial intelligence model to generate question information.

[0336] The server takes the embedded vector sequence obtained in step 5 as input and performs multi-layer neural network forward propagation on the GPU or CPU. This includes multi-layer self-attention calculation and feedforward network operations, iteratively predicting the probability distribution of the next label, and selecting labels according to a set sampling strategy until the final label is generated. Data calculus involves matrix multiplication, normalization, and nonlinear transformations on the high-dimensional vectors. The output is a sequence of label IDs representing the model's output text.

[0337] Step 7: The server decodes the model's output labeled sequence into natural language question text.

[0338] The server takes the sequence of token IDs output by the model as input, maps each token ID back to a word or character using a reverse lookup table, and concatenates them sequentially into natural language text. The server then segments the text based on newline characters, number prefixes, or punctuation marks, breaking it down into multiple independent questions. The data processing involves generating strings from the ID sequence and then splitting those strings. The output is a list of questions, with each question representing a segment of natural language text.

[0339] Step 8: The server stores the question information and updates the session state.

[0340] The server takes a list of questions and a session identifier as input, creates a record for each question in the database, and sets a question number and a status flag (unanswered). The server also records the number of questions and rounds that have been issued in the session state table. Data processing involves assigning a unique identifier to each question and establishing a relationship with the session. The output is a persistent collection of question records and updated session state data.

[0341] Step 9: The server sends a question message to the terminal.

[0342] The server takes a list of questions and a session identifier as input, packages them into a response structure, serializes it into text format, and sends it to the terminal via the network interface. Data processing involves encoding the internal structured data into a response format that the terminal can parse. The output is a response message sent over the network.

[0343] Step 10: The terminal parses the question and displays it to the user.

[0344] The terminal takes the response message returned by the server as input, recovers the session identifier and question list through the parsing module, and then generates the corresponding display components according to the question order in the interface rendering module, displaying each question on the screen. Data processing involves converting the text format into a list structure in memory and generating interface elements. The output is a set of questions displayed on the terminal screen.

[0345] Step 11: Users answer questions on the terminal.

[0346] Users input the questions displayed on the terminal as references, entering their answers in the input area below each question, such as "mainly in the IT industry," "hoping for technical cooperation and joint product development," or "preferably a partner from a local city." The terminal collects the user's input text as a string and establishes a pairing relationship with the corresponding questions in order. Data processing involves storing multiple answers into an answer list structure. The output is a list of answer texts.

[0347] Step 12: The terminal sends the response information to the server.

[0348] The terminal takes a session identifier and a list of answers as input, constructs structured data containing the session identifier, question number, and answer text, encodes it into a transmission format, and sends it to the server over the network. Data processing involves converting the answer list into a transmission record with metadata. The output is the answer message that arrives at the server.

[0349] Step 13: The server receives and stores the response information.

[0350] The server takes the response message sent by the terminal as input, parses it to obtain the session identifier, question number, and response text. The server looks up the corresponding question record in the database, writes the response text into the response field, and updates the status flag to "answered." Simultaneously, the server updates the latest response time and key fields of interest in the session status record. Data processing involves associating the response text with the specific question and session, and persisting this association. The output is the updated question record and session status.

[0351] Step 14: The server determines whether additional clarification is needed.

[0352] The server takes the abstract requirement information associated with the current session, the generated questions and corresponding answers as input, and executes rule-based judgments or calls the auxiliary model judgment module to check whether key attributes such as industry, region, and cooperation form have been fully covered. Data computation involves: performing logical judgments on existing fields (e.g., whether a field is empty, whether it meets preset coverage conditions). If any uncovered attributes are found, the server generates a flag indicating that further clarification is needed; otherwise, it generates a flag indicating that specific requirements can be generated. The output is a decision result: "Continue clarification?"

[0353] Step 15: The server generates additional prompts if further clarification is required.

[0354] The server takes abstract requirement information and key answers as input, reads relevant fields from the database, and organizes them into text fragments in the form of lists or entries. Based on an append prompt template, the server embeds these text fragments to generate new control text, for example: "The user's original requirement is: 'Looking for new business partners.' Currently known information includes: - Industry: Primarily IT; - Cooperation type: Desiring technical cooperation and joint product development; - Location: Preferably a local partner. Please generate 1-2 more specific questions to further confirm the company size, budget range, or cooperation period, making the requirement more specific." Data processing involves restoring the multi-field structure to a formatted natural language context. The output is the append prompt text.

[0355] Step 16: The server then invokes the generative artificial intelligence model again to generate additional questions based on the appended prompts.

[0356] The server takes the appended prompts as input, repeats the word segmentation, encoding, vector embedding, and neural network forward propagation processes to obtain a new output token sequence, which is then decoded into natural language question text. The server performs question segmentation again on the output text to obtain a list of appended questions. The data computation is the same as the aforementioned model inference, but the input context includes more conversation history content. The output is a set of appended question texts.

[0357] Step 17: The server generates specific requirements information without requiring further clarification.

[0358] The server takes abstract requirement information and all key answers as input, extracts relevant fields from the database, and constructs a summary input text according to a set order (e.g., industry first, then cooperation type, then region). The server embeds this summary text into a specific prompt template, forming a prompt such as: "The user's original requirement is: 'Looking for new business partners.' The clarified information is as follows: - Industry: Primarily IT; - Cooperation type: Technical cooperation and joint product development; - Region: Looking for partners in the local city. Please generate a clear and specific description of the requirement in 1-2 sentences, so that any reader can accurately understand what kind of partner the user is looking for." The server inputs this prompt into a generative AI model, which, through reasoning and decoding, obtains a concise, specific requirement text. Data processing involves condensing multiple pieces of information into a structured summary, which is further refined by the model into human-readable text. The output is a specific requirement message.

[0359] Step 18: The server selects collaborating partners based on specific requirements.

[0360] The server takes the specific requirement text as input. On one hand, it extracts key fields from the text, parsing values ​​for fields such as industry, region, and cooperation type. On the other hand, it converts the complete requirement description into a vector using a semantic vectorization module. From the candidate object information set, the server first uses field filtering to eliminate objects that do not meet the basic conditions. Then, it vectorizes the description text of the remaining objects and calculates their similarity to the requirement vector. Data processing includes string parsing, field matching, and vector similarity calculation. The output is a list of collaboration object information sorted by matching degree.

[0361] Step 19: The server sends the specific requirements and collaboration information to the terminal.

[0362] The server takes the specified requirement text and a list of collaborating objects as input, constructs a response structure containing a requirement summary and multiple object entries, encodes it into a transmission format, and sends it to the terminal via a network interface. Data processing involves assembling the results of internal multi-table queries into a hierarchical response message. The output is the response data for display on the terminal.

[0363] Step 20: The terminal displays specific requirements and collaborative results to the user.

[0364] The terminal takes the response data returned by the server as input, parses out the specific requirement text and the list of collaborating partners, and presents them separately on the interface. For example, the top displays "The user wants to find business partners in the IT industry in their local city, mainly for technical cooperation and joint product development." The bottom lists the name, region, business type, and matching degree of each collaborating partner in card format. Data processing involves extracting fields from the response structure and converting them into interface components. The output is a visual display of results on the terminal screen, allowing the user to perform further operations or make decisions.

[0365] 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".

[0366] Existing computer information processing and recommendation technologies suffer from the following technical shortcomings: (1) For users’ vague requests input in natural language, the request parsing logic often relies on fixed rules or simple keyword matching. The computer system has difficulty in timely and accurate determination of the request category and clarity, which leads to the inability of the subsequent processing flow to adaptively generate appropriate interactive content. Users often need to input multiple times inefficiently, increasing the interaction cost.

[0367] (2) Although generative artificial intelligence models have the ability to generate questions or suggestions in natural language, traditional systems usually use the model only as a "single response tool" and do not build a closed-loop mechanism of "automatic generation of prompts → model call → result feedback → structured utilization" at the system level. This makes it difficult for the model output to be deeply linked with the internal data structure (user profile, candidate object features, recommendation algorithm), and the intelligence and controllability of the overall computer processing flow are insufficient.

[0368] (3) In multi-round dialogue scenarios, existing systems often lack explicit modeling and programmatic implementation of the “refinement process of requirements”. The computer cannot dynamically generate new prompts based on each round of responses and guide the generative artificial intelligence model to raise the next question, resulting in rigid dialogue logic. It is difficult to structure the user’s vague requirements into parameters that can be directly used by the recommendation algorithm in a few rounds.

[0369] (4) Regarding the use of emotional information, traditional recommendation systems mostly build user profiles based on behavioral features such as clicks, browsing, and purchases, ignoring the possibility of extracting emotional features from natural language and introducing them into the candidate evaluation and prompt generation process. Therefore, at the algorithm level, it is impossible to dynamically adjust the recommendation strategy according to the user's current emotional state, resulting in insufficient recommendation results in terms of relevance and experience.

[0370] (5) From the perspective of data flow within the computer, the existing technology usually separates the link of “user behavior history and attribute information → clustering or similarity calculation → potential relationship / purchase demand mining → recommendation logic” from the link of “natural language dialogue and generative artificial intelligence model output”. The lack of a unified program architecture and data structure makes it difficult for the server to coordinate the structured feature quantity and the natural language capability of the generative artificial intelligence model in the same set of operation process, resulting in poor overall scalability and maintainability of the system.

[0371] Therefore, a new computer implementation is needed. This involves introducing a natural language parsing unit, a prompt generation unit, a phased requirement refinement unit, a candidate object evaluation unit, a recommendation description generation unit, and a user characteristic update unit into the server. This would construct an automatic prompt generation and closed-loop utilization mechanism for generative artificial intelligence models, thereby improving the computer's ability to understand fuzzy natural language requirements, utilize emotional features, and dynamically evaluate and recommend candidates. Ultimately, this would lead to an overall improvement in recommendation accuracy and interaction efficiency.

[0372] 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.

[0373] In this invention, the server includes a natural language parsing unit, a prompt statement generation unit, a phased requirement refinement unit, a candidate object evaluation unit, a recommendation description generation unit, and a user characteristic update unit. Each unit is executed by a computing device and works collaboratively under a predetermined data structure. Specifically, the natural language parsing unit is configured to perform lexical and semantic analysis on natural language request statements based on user input and behavioral information, determine the request category and clarity, and extract keywords, intent, and sentiment features. The prompt statement generation unit is configured to automatically generate prompt statements for input into a generative artificial intelligence model based on the request category, clarity, and sentiment features, and send these prompt statements to the generative artificial intelligence model to obtain dialogue statements or recommendation descriptions for user interaction. The phased requirement refinement unit is configured to send the dialogue statements to the user terminal, receive user response statements, and again call the natural language parsing unit to parse the user response statements, thereby progressively improving the granularity of the requirement description in multiple rounds, transforming fuzzy natural language requirements into structured requirement parameters. The candidate object evaluation unit is configured to retrieve information from an information storage device... The system obtains candidate object information based on user behavior and attribute information, performs clustering or similarity calculation to generate features representing potential relationship or purchase needs, and scores and ranks the candidate object information based on the highly refined demand parameters and the emotional features to select recommended object information. A recommendation description generation unit is configured to generate a second type of prompt statement for the recommended object information, input the prompt statement into the generative artificial intelligence model to obtain a user-oriented natural language recommendation description, and output the recommendation description to the user terminal after associating it with the recommended object information. A user characteristic update unit is configured to receive the user's operation history for the recommended object information, store the operation history in an information storage device, and dynamically update the user characteristic information based on the storage results, so that the natural language parsing, prompt statement generation, and candidate object evaluation can reflect the latest user preferences and behavioral patterns in subsequent calls. This allows for a closed-loop data processing flow within the computer: parsing → generating prompts → calling generative AI models → structured utilization → recommendation → feedback updates. This enables system-level integration of the capabilities of generative AI models, significantly improving server efficiency and recommendation accuracy in handling fuzzy natural language requirements, evaluating candidates using sentiment features, and adaptively generating recommended content. Ultimately, this improves the overall performance of human-computer dialogue recommendation systems in the field of computer technology.

[0374] A “computing device” refers to an electronic computing device that processes and performs calculations on data by executing program instructions, including a central processing unit, a storage unit, and control circuits that work in conjunction with it.

[0375] The "Natural Language Parsing Unit" refers to a functional module executed by a computing device to perform word segmentation, part-of-speech tagging, syntactic analysis, and semantic understanding on text input by users in human natural language, thereby obtaining structured information such as keywords, intent, and sentiment features.

[0376] "User input information" refers to the request content, response content, and explanatory information related to the needs that users input into the system through the terminal in the form of text, speech-to-text, or other natural language.

[0377] "User behavior information" refers to the operation records generated by users in the system, including actions such as clicking, browsing, favorites, joining groups, purchasing, and logging out, as well as corresponding data such as time and object identifiers.

[0378] "Natural language request statements" refer to a sequence of texts entered by a user in natural language that expresses a need, question, or intention, such as a shopping intention, an intention to find a partner, or a request to obtain information.

[0379] "Request category" refers to the functional category or business type to which the request belongs, as determined by the natural language parsing unit based on the natural language request statement, such as product recommendation request, relationship matching request, content recommendation request, etc.

[0380] "Request clarity" refers to the evaluation result of the natural language parsing unit on the completeness of information such as constraints, target objects, and scope contained in the request statement. It is used to distinguish between vague requests and explicit requests.

[0381] "Prompt statements" are control texts input into generative artificial intelligence models to instruct the models to perform specific generative tasks. These statements include descriptions of the user's background, current context, desired output format, and generation goals.

[0382] "Generative artificial intelligence models" refer to artificial intelligence models based on deep learning technology that can automatically generate natural language text based on input prompts, including large-scale language models.

[0383] "Dialogue statements" refer to the natural language output content generated by generative artificial intelligence models based on prompts and used by the system to display to users, including questions, introductory statements, or explanatory statements.

[0384] The "phased requirement refinement unit" refers to a functional module executed by the computing device that receives user responses through multiple rounds of interaction and repeatedly calls natural language parsing to gradually refine and structure user requirements.

[0385] "User response statements" refer to the text of a user's reply in natural language after receiving the dialogue statements output by the system, which is used to supplement, correct or refine the original requirements.

[0386] "Highly refined requirement content" refers to user requirement information represented in the form of structured parameters after multiple rounds of natural language parsing and requirement refinement, including clear categories, constraints, and preferences.

[0387] "Information storage device" refers to a storage system used to persistently store user behavior information, attribute information, and candidate object information, including databases, file storage systems, or other data storage media.

[0388] "Candidate object information" refers to various object data obtained from information storage devices that may be recommended to users, including attribute information of goods, services, content, user accounts, or other resources.

[0389] "User attribute information" refers to static or relatively stable data related to a user's personal characteristics, including age group, region, occupation, interest tags, skill tags, etc.

[0390] "Emotional state" refers to the type and intensity of emotions inferred from user text or behavior based on natural language parsing or sentiment analysis, such as joy, sadness, anxiety, neutrality, etc.

[0391] The "candidate evaluation unit" refers to a functional module executed by a computing device to score, sort, and filter candidate information based on highly refined demand content, emotional state, and user characteristics.

[0392] "Recommended object information" refers to one or more object data that are selected and actually recommended to users based on the evaluation results after evaluating candidate object information.

[0393] The "recommendation description generation unit" refers to a functional module executed by the computing device, which generates corresponding prompt statements for the recommended object information, calls the generative artificial intelligence model to generate recommendation description statements and associates them with the recommended object information.

[0394] "Recommended explanatory statements" refer to the natural language text output by the generative artificial intelligence model based on the prompts provided by the recommendation explanation generation unit, which is used to explain or introduce information about the recommended object.

[0395] "User terminal" refers to human-computer interaction devices that allow users to input and receive output, including mobile terminals, desktop terminals, tablet devices, or other devices with network communication and interface display functions.

[0396] "User action history" refers to the interaction records generated by users after receiving recommendation information, including actions such as clicking on recommendations, viewing details, purchasing, ignoring or rejecting, and their related attributes.

[0397] "User characteristic information" refers to the feature data used to characterize user preferences, interest distribution, and behavioral patterns, which are calculated by the system based on user input information, behavioral information, and operation history.

[0398] "Statistical language processing program" refers to software components that use statistical methods to analyze and process natural language text, including programs for word segmentation, word frequency statistics, language modeling, topic modeling, etc.

[0399] A "machine learning program" is a software component that uses training data to learn parameters or models in order to perform tasks such as classification, clustering, regression, or feature extraction.

[0400] "Emotional features" refer to the feature data extracted from text through natural language parsing or sentiment analysis that quantitatively represents a user's emotional state, including emotion category, confidence level, and intensity.

[0401] "Clustering" refers to an unsupervised learning process that groups multiple users or objects based on feature vectors, so that samples within the same group have high similarity in the feature space, while samples between different groups have low similarity.

[0402] "Similarity calculation processing" refers to the process of calculating the similarity between the feature vectors of users or objects based on a predetermined similarity metric, thereby obtaining an index of the degree of similarity between the two.

[0403] "Potential relationship needs" refer to the needs for new interpersonal or collaborative connections that users may need to establish, which are inferred by analyzing users' behavioral history and attribute information but have not yet been explicitly expressed.

[0404] "Potential purchase demand" refers to the intention to purchase goods or services that a user may not have explicitly expressed but could potentially generate in the future, inferred by analyzing user behavior history and attribute information.

[0405] In the following embodiments, the system includes at least a server, a terminal, and a user interacting with the server through the terminal. The server executes multiple functional units through a computing device, which can be implemented as software modules, microservices, or hardware logic; the terminal interacts with the user through a graphical user interface and exchanges data with the server through a communication network. The embodiments of this invention focus on the technical processing within the server, introducing a natural language parsing unit, a prompt generation unit, a phased requirement refinement unit, a candidate object evaluation unit, a recommendation description generation unit, and a user characteristic update unit. This constructs a closed-loop data processing architecture within the computer for generative artificial intelligence models, thereby improving the computational performance and accuracy of natural language requirement processing, sentiment feature utilization, and candidate object recommendation.

[0406] I. Overall Hardware and Software Composition of the System A server can be a general-purpose processor-based computer system. A server includes: a central processing unit (CPU), an optional graphics processing unit (GPU), main memory (RAM), a large-capacity storage device (such as a solid-state drive), a network interface controller, and a bus architecture for executing the operating system and applications. The operating system running on the server can be a Unix-based system or other multitasking operating system. The server runs multiple software components on top of the operating system, including but not limited to: The server uses a natural language processing library to perform word segmentation, part-of-speech tagging, dependency parsing, and intent recognition on text data; the server uses a data processing library to perform tabulation, cleaning, feature construction, and statistical operations on user behavior data, attribute data, and candidate object data; the server uses a machine learning library to perform clustering, similarity calculation, and classification; and the server uses a deep learning framework to load and execute recommendation models, sentiment classification models, and encoding / decoding submodules for generative artificial intelligence models.

[0407] The terminal can be a smartphone, tablet computing device, or personal computer. The terminal includes a touchscreen display, processor, memory, and wireless or wired communication modules. The terminal runs a terminal operating system and user interface-based applications, which can be built using user interface libraries or cross-platform mobile frameworks to create dialog interfaces, recommendation list interfaces, and interactive controls.

[0408] Users input natural language text through the terminal and perform actions such as clicking, browsing, saving, purchasing, and joining communities. User input and behavior are encoded into structured data through the terminal and transmitted to the server via communication protocols.

[0409] II. Implementation Forms of Natural Language Parsing Units The server stores the natural language parsing program on storage media. The natural language parsing unit, executed by the processing unit, performs fine-grained analysis of the user's input natural language requests. The server employs a multi-layered processing architecture within this unit: The server first uses word segmentation and sub-word splitting algorithms to transform the input sentence into a sequence of words or sub-words. Then, the server encodes the sentence using a trained sequence encoding model, which can be an encoder structure based on an attention mechanism. The server extracts sentence vectors from the encoding results and inputs them into an intent classifier, which can be a multi-layer fully connected network. The server then outputs request category labels through this classifier, such as "product recommendation request," "relationship matching request," or "content recommendation request."

[0410] The server further annotates the input sequence using a sequence labeling model to identify slot information such as price, category, brand, and time. This sequence labeling model can be a combination of a conditional random field model and a bidirectional recurrent neural network or convolutional structure, or it can be a labeling module utilizing a self-attention mechanism. The server outputs structured parameters in key-value pair form through these models, such as: category = "fashion", budget ≈ "200".

[0411] The server also integrates a sentiment analysis submodule within its natural language parsing unit. The server inputs sentence vectors into a sentiment classification network, which may include several fully connected layers and an output layer. The server pre-trains this network using a cross-entropy loss function, employing corpora with sentiment labels as training samples. During runtime, the server outputs sentiment states based on the classification results, such as "joy," "neutral," and "anxious," and outputs the corresponding confidence scores as sentiment features.

[0412] Through the above structure, the server can perform high-dimensional vector encoding and joint extraction of intent, slots, and sentiment for any natural language request without relying on simple keyword matching or fixed rules, thereby providing fine-grained features for subsequent prompt generation and candidate evaluation. This invention improves the accuracy and generalization ability of traditional rule-based systems when parsing complex sentences by introducing deep sequence coding and sequence labeling models.

[0413] III. Implementation Form of the Prompt Statement Generation Unit In the prompt statement generation unit, the server constructs prompt statements for invoking the generative artificial intelligence model based on the request category, request clarity, keywords, slot parameters, and sentiment features output by the natural language parsing unit.

[0414] The server predefines a set of prompt templates for different tasks. Each template contains a fixed instruction section, a variable context section, and a generation goal description section. The server selects the appropriate template based on the request category. For example, when the request category is "shopping_fuzzy" and the request is not very clear, the server selects the template for the "clarification category"; when the request category is "relationship matching" and there are already clear skills and industry parameters, the server selects the template for "generating collaboration suggestions".

[0415] The server fills the template placeholders with the user's original statement, parsed slot information, and sentiment features in a predetermined order, thereby generating a complete prompt statement. For example, when a user enters "I want to buy something new," the server can generate the following prompt statement: The user says, "I want to buy some new things." Ask the user one or two questions in Chinese to understand their preferred product categories (e.g., fashion, electronics, books), using a natural and friendly tone.

[0416] When the server needs to generate relationship suggestions based on behavioral history, it can generate the following prompt statement: The user recently participated in the following data analysis-related activities: "Data Analysis Fundamentals Workshop" and "Python Data Visualization Sharing Session". Please generate three recommended types of collaborators or online communities for this user in Chinese, explaining the reasoning for each suggestion in one sentence.

[0417] Through the aforementioned template and parameter combination mechanism, the server upgrades the generation of prompt statements from rule concatenation to template filling based on structured features. This ensures that the prompt statements have sufficient contextual information and a clear task description when calling generative artificial intelligence models, thereby improving the relevance and controllability of the generated results.

[0418] IV. Invocation and Internal Processing of Generative Artificial Intelligence Models The server invokes a generative AI model deployed on a dedicated computing node via a network interface. This generative AI model can employ a multi-layer transformer architecture, consisting of multiple layers of self-attention encoders and decoders. During the pre-training phase, the model uses a large-scale text corpus as training data, aiming to predict the next word or generate a complete sentence in a given context; the loss function can be cross-entropy loss. In the fine-tuning phase, the server can use dialogue corpora and recommendation corpora to perform supervised fine-tuning of the model, making it more suitable for dialogue clarification and recommendation tasks.

[0419] The server encodes the generated prompts as conditional inputs into the model's input sequence. The model maps the text to vector representations through an embedding layer and captures long-range dependencies in the sequence using a multi-head self-attention module. During the decoding phase, the server uses a beam search algorithm to generate multiple candidate output sequences and selects the best output as either the dialogue prompt or the recommended explanatory statement based on probability scores and a length normalization strategy.

[0420] For example, when the server inputs the above clarification-category prompts into the generative artificial intelligence model, the model's output might be: Which type of product would you prefer to buy right now? For example, fashion apparel, digital products, or books? The server utilizes the sequence generation capabilities of generative artificial intelligence models in this way. However, because the prompts contain structured features and explicit task instructions, the model's degrees of freedom are controlled, thereby reducing the uncertainty of the generated content and improving practical usability.

[0421] V. Operation Mode of Phased Requirements Refinement Unit In the phased requirements refinement unit, the server receives the dialogue statements output by the generative artificial intelligence model and sends them to the terminal via a communication interface. The terminal displays the statements in a graphical interface and waits for user input and response.

[0422] Users input response statements through the terminal, such as answering "fashionable" to a category question posed by the server, or answering "around 200 yuan" to a budget question. The terminal then sends this response statement, along with the user identifier and session identifier, to the server.

[0423] The server invokes the natural language parsing unit again to perform intent and slot identification on the response statement, extracting the category "fashion" and budget range information. The server then overlays these new slot parameters into the session context data structure, forming a progressively more complete requirement description. The server maintains a session context object in the phased requirement refinement unit, which stores confirmed parameters in key-value pairs, for example: Category = "Fashion"; Budget minimum = 100; Budget cap = 300; Emotional state = "joy".

[0424] When the server detects that the clarity of the requirement has reached a preset threshold (e.g., all key slots are filled), it stops generating further clarifying prompts and instead initiates the candidate selection and scoring process through the candidate evaluation unit. This phased refinement mechanism transforms what would otherwise be a difficult-to-parse, one-time fuzzy natural language requirement into a multi-turn dialogue-driven structured parameter construction process within the computer, thereby reducing the complexity of a single parsing operation and improving parsing accuracy.

[0425] VI. Algorithm and Data Structure of Candidate Evaluation Unit In the candidate evaluation unit, the server retrieves candidate object information relevant to high-precision requirements from the information storage device. The candidate object information is stored in tabular form and includes fields such as object unique identifier, category tag, price, attribute characteristics, historical click-through rate, and conversion rate.

[0426] The server first filters out objects that meet basic constraints using query conditions, such as filtering products based on category and budget range, or filtering potential collaborators based on skill tags and industry tags. Then, the server uses a data processing library to load the filtering results into a data table, imputing missing values ​​and normalizing numerical features. The server constructs feature vectors based on object type and user task, such as combining price, popularity, similar user purchase rate, and similarity to the user's historical preferences into a multi-dimensional vector.

[0427] The server can use various machine learning models to score candidate objects: In one implementation, the server uses tree models such as gradient boosting trees to perform non-linear mapping on the input feature vectors and output a relevance score; In another implementation, the server uses a deep recommendation model based on matrix factorization or multilayer perceptron to map user feature vectors and object feature vectors to the latent space and calculate the matching score through dot product or non-linear combination.

[0428] During model training, the server uses historical interaction data as samples, with sample labels including click markers, purchase markers, or dwell time. The loss function can be binary cross-entropy or ranking loss. The server updates model weights using gradient descent, learning the complex relationship between user preferences and object features.

[0429] During the runtime phase, the server incorporates highly refined requirement parameters and emotional features into the candidate object scoring process. For example, when the emotional state is "joy," the server adds an emotional dimension to the feature vector and weights the scores of objects related to celebration or advancement in the output layer or post-processing stage, thereby technically achieving quantitative utilization of emotional information.

[0430] Furthermore, servers can use clustering algorithms or similarity calculations to generate potential demand features from user behavior and attributes. For example, a server can use the k-means algorithm to cluster user feature vectors, assigning users to corresponding groups and learning typical demand patterns for each group. During the candidate evaluation phase, the server references these group characteristics, increasing the scores of objects that align with the group's typical needs, thereby uncovering latent relationship needs or potential purchasing needs that have not yet been explicitly expressed.

[0431] VII. Recommendations on the specific processing of generated units After scoring and ranking the candidate objects, the server selects a number of recommended objects. To improve user understanding and acceptance, the server generates natural language descriptions for these recommended objects in the recommendation description generation unit.

[0432] The server prepares a second type of prompt template for this unit. The template includes the title of the recommended object, key attributes, and user-defined constraints. The server fills the template with the selected result to generate the prompt statement. For example, the server can generate the following prompt statement: Based on the following product list, generate a Chinese recommendation message of no more than 80 characters for the user, in a friendly tone: Product 1: Spring Fashion Coat, Price 199 Yuan; Product 2: Minimalist Dress, Price 259 Yuan.

[0433] The server invokes a generative artificial intelligence model to obtain a piece of natural language text for the user, for example: Based on your budget of around 200 yuan, we have selected several fashionable spring items for you, such as lightweight jackets and simple dresses, which are both versatile and stylish. Feel free to take a look and see if there are any you like.

[0434] The server packages the recommendation description and the corresponding recommended object information into response data and sends it to the terminal. The terminal displays the recommendation as dialogue content and presents the recommended objects in the form of a list or cards.

[0435] Because the explanatory text is generated by a generative AI model based on structured object information, the server does not need to manually write scripts for each combination, resulting in high system scalability. Furthermore, since the prompts contain explicit object attributes and user constraints, the model-generated text is highly semantically consistent with the actual recommended content, thereby reducing comprehension errors and improving the accuracy of the recommendation explanations.

[0436] VIII. User Feature Update Unit and Adaptive Improvement After viewing the recommendation list on the terminal, users can perform actions such as clicking, adding to favorites, purchasing, or ignoring. The terminal sends these actions as user operation history to the server. In the user characteristic update unit, the server writes the operation history into the information storage device and aggregates and statistically analyzes user behavioral characteristics based on a certain time window, such as click-through rate, conversion rate, and preference for different categories.

[0437] The server uses these aggregated features to update user characteristic information, which is stored in vector form and includes dimensions such as interest distribution, price sensitivity, brand preference, and emotional response patterns. During the next natural language parsing, prompt generation, and candidate evaluation, the server inputs the updated user characteristic vector into the relevant model, thus enabling the system behavior to dynamically adjust as user behavior evolves over time.

[0438] Through this automatic update mechanism, the server forms a feedback loop within the computer: recommendation performance affects user behavior, user behavior updates user characteristics, and user characteristics change the logic for generating the next round of ratings and prompts, thereby gradually improving the accuracy and stability of the overall recommendation system.

[0439] IX. Explanation of Technical Effects and Causal Relationship This invention achieves specific technical effects by constructing a unified architecture within the server that combines natural language parsing, prompt generation, generative artificial intelligence model reasoning, and structured recommendation computation: By using deep natural language parsing and phased requirement refinement, the server reduces the complexity of single-round parsing, breaks down complex semantics into multiple simpler parsings, reduces parsing errors, and improves the accuracy of intent recognition and slot extraction, thereby directly improving the accuracy of candidate selection criteria.

[0440] The server uses a template-driven prompt generation mechanism to ensure that the generative AI model receives instruction text containing complete context and target description, reducing the randomness of the model's output; under the same computing resources, it increases the proportion of useful output and improves the efficiency of computing resource utilization.

[0441] By incorporating sentiment and clustering features into the candidate evaluation process, the server enables the recommendation algorithm to consider not only static interests but also dynamic sentiment and group patterns. This reduces the recommendation of irrelevant objects, increases the probability of user clicks and conversions, and, from the perspective of computational error and accuracy, reduces the deviation between the recommendation results and the user's true preferences.

[0442] By uniformly managing user characteristic information and operation history, the server internally constructs an incrementally updatable data structure, enabling the model to be frequently updated with small batches of data without having to completely retrain the large model each time. This significantly reduces computational complexity and training time, improving computational efficiency and resource consumption.

[0443] Furthermore, in this invention, the server employs a specific, unconventional processing sequence (first performing natural language parsing and sentiment extraction, then driving prompt generation and invoking the generative AI model, and finally integrating the output into the recommendation algorithm in a structured manner), making the generative AI model a controllable link in the computational process, rather than an isolated component. This architectural optimization changes the traditional "model as a black box output" approach, improving interpretability and maintainability at the system level.

[0444] 10. Other Implementation Forms and Variations In another implementation, the server can integrate the natural language parsing unit and the sentiment analysis submodule into a unified sequence-to-sequence model, and further reduce model parameter redundancy and improve inference speed through multi-task learning of the shared coding layer.

[0445] In another implementation, the server can adopt a strategy that combines rule-based and model-based approaches for the prompt generation unit. That is, a lightweight generation model rewrites some natural language expressions on the basic template, thereby increasing language diversity while maintaining structural stability.

[0446] In other implementations, the terminal can integrate a local preprocessing module to perform speech recognition on audio and facial expression recognition on images before transmitting the resulting text and features. After receiving these preprocessed results, the server still uses the unified data structure of this invention for subsequent processing to reduce network transmission load and improve end-to-end response speed.

[0447] Furthermore, the functional units described in this invention can be deployed on multiple physical servers or virtual machines in different modular ways and communicate through message queues or remote calls. As long as they logically implement the same data flow and operation order, they are all considered as embodiments of this invention.

[0448] use Figure 14 The processing flow is explained.

[0449] Step 1: Users input their initial natural language requirements using a terminal, which generates behavioral data.

[0450] In the terminal's dialogue interface, users can bring up the keyboard via the touchscreen, input natural language text such as "I want to buy something new" or "I want to find a new partner," and then click the send button. Users may also browse, click, and favorite items on the interface.

[0451] Input: Text data (natural language strings) entered by the user on the terminal and behavioral event data in the current session (such as the identifier and time of clicking on content).

[0452] The terminal's specific data processing: The terminal encapsulates user text and behavioral events into structured data objects, adds user identifiers, session identifiers and timestamps, and converts them into a unified encoding format.

[0453] Output: The request data packet sent by the terminal to the server, which includes at least the user text field, the list of behavioral events, the user identifier, and the session identifier.

[0454] Step 2: The terminal sends the encapsulated request to the server over the network.

[0455] The terminal invokes the network communication module, uses a secure transmission protocol to send the request data packet constructed in step 1 to the interface address provided by the server, and waits for the server's response.

[0456] Input: A request data packet containing user text and behavioral events.

[0457] The terminal performs specific data processing: The terminal serializes internal data objects into a transmission format, sets request header information (including authentication information and content type), and sends it through the network interface.

[0458] Output: The original request message received by the server, which serves as input for subsequent server processing.

[0459] Step 3: The server parses the received request and extracts the basic fields.

[0460] The server receives the request message on the receiving port, parses the user identifier, session identifier, natural language text, and list of behavioral events from the message body, and loads them into the session context structure in memory.

[0461] Input: The original request message received from the terminal.

[0462] The server's specific data processing involves deserializing the message body, converting text, identifiers, and events into internal data structures (such as objects or records), and checking data integrity and format validity.

[0463] Output: An internal context object containing user text, action events, user ID, and session state pointers.

[0464] Step 4: The server performs word segmentation, intent recognition, and slot extraction on the user's text in the natural language parsing unit.

[0465] The server inputs the user text from the context object into the natural language parsing module, and uses pre-loaded word segmentation and intent classification models to perform multi-level analysis on the text to identify the request category and preliminary requirement parameters.

[0466] Input: User text string from step 3 and user-related historical features (such as language preferences).

[0467] The server's specific data processing and calculations are as follows: First, the server performs word segmentation and sub-word splitting on the text, mapping it into an embedded vector sequence; then, the server inputs the sequence into the intent classification network, calculates the probability distribution of each category, and selects the request category with the highest probability; the server then identifies entities and slots, such as category words and budget words, through a sequence labeling model.

[0468] Output: Parsing result data structure, including request category, keyword list and preliminary slot information (e.g., category = unknown, budget = unknown, only keywords such as "buy" and "new" are available).

[0469] Step 5: The server performs sentiment analysis to obtain the user's current sentiment state and sentiment characteristics.

[0470] The server inputs the sentence vector of the user's text into the sentiment classification sub-model to determine the sentiment type and intensity corresponding to the text, and generates sentiment feature vectors that can be used subsequently.

[0471] Input: Sentence vector after user text encoding, and sentiment classification weight parameters stored in the model.

[0472] The server's specific data processing and calculation: The server calculates the emotion scores of each emotion in the output layer of the emotion classification network through forward propagation, and uses activation functions and normalization operations to convert the results into emotion labels (such as joy, neutral) and corresponding confidence scores; the server encodes the emotion labels into numerical features.

[0473] Output: Emotional state labels and emotional feature values, as extended user context information.

[0474] Step 6: The server generates prompts for invoking generative artificial intelligence models based on the request type and clarity.

[0475] The server checks the slot filling status and request clarity in the parsing results. If key parameters are missing, it selects an appropriate template in the prompt statement generation unit and fills in the user's original statement, known slots, and sentiment features to generate one or more prompt statements to drive the generative artificial intelligence model to generate clarification questions.

[0476] Input: Request category, request clarity, keywords, slot information, and sentiment metrics.

[0477] The server's specific data processing and calculation: The server selects the corresponding task template (such as the "clarification category" template) based on conditional judgment, places the user's original text in the context of the template, and embeds the existing parameters as explanatory text prompts; the server performs simple rule processing on sentiment characteristics, such as requiring a more friendly tone in the prompt when the user is in a positive mood.

[0478] Output: A structured prompt text, such as: "The user says: 'I want to buy something new.' Please ask the user 1-2 questions in Chinese to understand the product categories he is more interested in (e.g., fashion, digital products, books, etc.), in a natural and friendly tone." Step 7: The server invokes a generative artificial intelligence model to generate dialogue statements based on the prompts.

[0479] The server sends the prompt obtained in step 6 to the deployed generative artificial intelligence model interface. The model encodes the prompt based on a multi-layer attention network and generates natural language output. The server receives and saves this output as the content for subsequent questions to the user.

[0480] Input: Prompt text and parameters of the generative AI model (including word embedding matrix, multilayer transformer weights, etc.).

[0481] The server's specific data processing and computation: The server encodes the prompts into a vector sequence, the model performs forward computation with a multi-layer self-attention and feedforward network, and generates several candidate output sentences using a beam search strategy through the decoder; the server selects the optimal output as the dialogue statement based on the likelihood score and length penalty.

[0482] Output: User-facing dialogue statements, such as: "Which type of product would you prefer to buy right now? Fashion apparel, digital products, or books?" Step 8: The server sends the dialogue message to the terminal, which displays the message and receives the user's response.

[0483] The server packages the dialogue statement and session identifier generated in step 7 into a response message and sends it to the terminal. The terminal parses the response, displays it to the user in the form of a text bubble in the dialogue interface, and waits for the user to input a new response statement.

[0484] Input: Server-side dialogue statements and session identifiers.

[0485] The terminal's specific data processing and calculation: The terminal receives and deserializes the server response, inserts the text into the message list of the interface, and updates the scroll position of the interface; the terminal then captures new user input (such as "fashion") and repackages the request.

[0486] Output: The user's new response request data packet, including text such as "fashion" and an updated timestamp, is ready to be sent to the server.

[0487] Step 9: The server uses the new user response to update the requirement parameters and determine whether the requirement is clear enough.

[0488] The server receives the user's response text forwarded by the terminal, such as "fashion," and calls the natural language parsing unit again to perform intent and slot recognition on the short sentence, confirming that it is category information and writing it into the session context.

[0489] Input: User response text and original session context object.

[0490] The server performs specific data processing and calculations: The server segments and recognizes entities in the response text, mapping "fashion" to the "fashion" tag in the predefined category set; the server adds or updates the field category = "fashion" in the context, and calculates a new request clarity value based on the number of filled slots.

[0491] Output: The updated session context object, containing more complete requirement parameters, such as category = "fashion" and budget = unknown.

[0492] Step 10: When the requirements are not fully clear, the server generates prompts again and repeatedly calls the generative artificial intelligence model for further clarification.

[0493] The server checks for key parameters that are still missing in the session context, such as the budget range. If the request clarity is still below the threshold, it uses a template adapted to the budget clarification in the prompt generation unit to construct a new prompt statement and calls the generative AI model again to generate additional questions.

[0494] Input: Updated session context (including category information) and current request clarity.

[0495] The server's specific data processing and calculations: The server selects a budget clarification template, embeds category parameters, and generates a prompt statement, such as: "The user has expressed interest in the 'Fashion' category. Please ask for his budget range in Chinese and provide several examples of budget ranges."; The server then calls the generative artificial intelligence model again for forward inference to obtain new question text.

[0496] Output: Dialogue statements related to the budget, such as: "What is your approximate budget? For example, under 100 yuan, 100-300 yuan, or over 300 yuan?" Step 11: Once the requirements parameters are fully defined, the server retrieves candidate objects from the information storage device and constructs object feature vectors.

[0497] When the session context contains key slots (such as category "fashion" or budget "100-300"), the server no longer generates new prompts, but instead the candidate object evaluation unit queries the storage for object records that meet these constraints.

[0498] Input: Highly detailed requirement parameters (category, budget, etc.) and basic user characteristics.

[0499] The server's specific data processing and calculations are as follows: The server executes database queries and filters product records based on category and price range; then, the data processing library loads the query results into a data table and calculates features for each record, such as standardized price, historical click-through rate, and similar user purchase probability; the server combines these features into an object feature vector and concatenates it with the user characteristic vector.

[0500] Output: A set of candidate objects with feature vectors for use by the scoring model.

[0501] Step 12: The server scores and sorts the candidate objects, generating a list of recommended objects.

[0502] The server inputs the feature vectors of candidate objects and the user characteristic vectors into a pre-trained recommendation model. The model outputs a matching score for each candidate object. The server sorts the candidates by score and selects the objects with higher scores as the recommended objects.

[0503] Input: A set of candidate object feature vectors and a user characteristic vector.

[0504] The server's specific data processing and calculations: The server performs model inference in batches in memory, such as calculating the similarity between user vectors and object vectors through multilayer perceptrons or matrix factorization models, and outputs a normalized matching score using activation functions; the server performs a sorting operation on the scores of all candidate objects and selects the top few.

[0505] Output: A list of recommended object information, including object identifier, title, key attributes, and matching score.

[0506] Step 13: The server generates prompts for recommendation explanations and calls a generative artificial intelligence model to generate explanations for the recommendations.

[0507] In the recommendation description generation unit, the server constructs prompts for the recommendation objects obtained in step 12, embeds the name, price, features, and user needs parameters of each object into the template, and then calls the generative artificial intelligence model to generate a coherent natural language recommendation script.

[0508] Input: A list of recommended objects and required parameters (such as category and budget range).

[0509] The server's specific data processing and calculations: The server sequentially lists the brief attributes of multiple objects, such as "Product 1: Spring Fashion Coat, Price 199 Yuan; Product 2: Minimalist Dress, Price 259 Yuan"; it then combines these texts with task instructions to form prompts, such as: "Based on the following product list, generate a Chinese recommendation message of no more than 80 characters for the user, in a friendly tone: ..."; subsequently, it calls the generative artificial intelligence model to perform the text generation forward operation.

[0510] Output: We recommend using a descriptive sentence, such as: "Based on your budget of around 200 yuan, we have selected several spring fashion items for you, such as lightweight jackets and simple dresses, which are both versatile and stylish. Please take a look and see if there are any you like." Step 14: The server packages the list of recommended objects and the accompanying descriptions and sends them to the terminal. The terminal then displays the results and guides the user through the process.

[0511] The server encapsulates the structured data of the recommended items along with the generated recommendation description in a response message and sends it to the terminal. After parsing, the terminal displays the description in the dialog area, presents the recommended items to the user in the form of product cards or lists, and adds interactive buttons such as "View Details", "Buy Now" or "Contact Us" to each recommendation.

[0512] Input: The server-side prepared response data packet, including the recommended objects and their descriptions.

[0513] The terminal's specific data processing and calculation: The terminal maps structured data into interface components, displaying the title, image link, and price of each object in the card; the terminal binds click event listening logic to each card and records subsequent user operations.

[0514] Output: A list of recommendations and explanatory text displayed on the terminal interface, along with an interactive state awaiting further user action.

[0515] Step 15: When a user interacts with a recommended object on the terminal, the terminal sends the interaction history back to the server, and the server updates the user's profile information.

[0516] When a user selects a recommended object on the terminal, they can click, view, or purchase it. The terminal records the operation type, object identifier, and time, and reports this as the user's operation history to the server. The server writes the operation to storage in the user characteristic update unit and recalculates the user's preference features for natural language parsing, prompt generation, and candidate object evaluation in subsequent sessions.

[0517] Input: User actions on the terminal (such as clicking on a recommended item, completing an order, etc.).

[0518] The specific data processing and calculations on the terminal and server are as follows: The terminal encapsulates the event into an event record and sends it; the server parses the event, writes it into the user event table, and updates the user's click-through rate and conversion rate under different categories and price ranges through statistical calculations, and updates the user characteristic vector; the server may perform incremental training or parameter fine-tuning so that the adjusted model parameters can be used in the next recommendation.

[0519] Output: Updated user behavior database records and updated user characteristic information structure, thus forming a closed loop of "recommendation-feedback-update", providing a foundation for generating more accurate prompts and recommendation results in the future.

[0520] 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.

[0521] 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.

[0522] 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.

[0523] 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.

[0524] 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.

[0525] Second Implementation Method Figure 3 An example of the configuration of the data processing system 210 according to the second embodiment is shown.

[0526] 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.

[0527] 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).

[0528] 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.

[0529] 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.

[0530] 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).

[0531] 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.

[0532] 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.

[0533] 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.

[0534] 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).

[0535] 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.

[0536] 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".

[0537] 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.

[0538] 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.

[0539] 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.

[0540] 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.

[0541] 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.

[0542] 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.

[0543] 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.

[0544] 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.

[0545] 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.

[0546] Third Implementation Method Figure 5 An example of the configuration of the data processing system 310 according to the third embodiment is shown.

[0547] 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.

[0548] 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).

[0549] 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.

[0550] 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.

[0551] 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).

[0552] 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.

[0553] 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.

[0554] 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.

[0555] 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.

[0556] 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.

[0557] 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".

[0558] 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.

[0559] 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.

[0560] 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.

[0561] 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.

[0562] 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.

[0563] 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.

[0564] 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.

[0565] For example, the collection unit is implemented by the control unit 46A of the head-mounted terminal 314 or the specific processing unit 290 of the data processing device 12. For example, the acquisition unit uses the camera 42 or communication I / F 44 of the head-mounted terminal 314 to acquire step data, which is then processed by the specific processing unit 290 of the data processing device 12. For example, the analysis unit is implemented by the specific processing unit 290 of the data processing device 12, which analyzes the data from the collection unit and the acquisition unit. For example, the generation unit is implemented by the specific processing unit 290 of the data processing device 12, which uses a generation AI to generate a menu. For example, the serving unit is implemented by the speaker 240 and display 343 of the head-mounted terminal 314 or the specific processing unit 290 of the data processing device 12, which provides the generated menu to the user. The correspondence between each unit and the device or control unit is not limited to the above examples and various changes can be made.

[0566] 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.

[0567] Fourth Implementation Method Figure 7 An example of the configuration of the data processing system 410 according to the fourth embodiment is shown.

[0568] 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.

[0569] 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).

[0570] 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.

[0571] 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.

[0572] 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).

[0573] 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.

[0574] 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.

[0575] 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.

[0576] 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.

[0577] 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.

[0578] 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.

[0579] 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".

[0580] 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.

[0581] 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.

[0582] 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.

[0583] 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.

[0584] 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.

[0585] 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.

[0586] 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.

[0587] 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.

[0588] 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.

[0589] 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.

[0590] 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.

[0591] 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.

[0592] 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).

[0593] 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.

[0594] 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."

[0595] 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.

[0596] 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).

[0597] 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.

[0598] 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.

[0599] 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.

[0600] 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.

[0601] 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.

[0602] 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.

[0603] 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.

[0604] 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.

[0605] 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.

[0606] 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.

[0607] In addition, the following notes are provided in response to the above explanation.

[0608] Example 1 (Note 1) An information processing system, characterized in that it comprises: An apparatus for receiving interest information and skill information input by a user via a terminal through a processing device control unit, and for acquiring the interest information and skill information as text information; An apparatus for managing a storage area for storing text information in a structured information form based on acquired text information, and for storing interest information and skill information in association within the storage area; An apparatus for generating feature quantities for calculating the similarity between users based on the stored interest information and skill information, and for calculating a similarity index between users based on the feature quantities; A device for extracting combinations of users that meet predetermined selection conditions based on a calculated similarity index, and selecting the extracted combinations of users as connection candidates; An apparatus for generating a notification message indicating the existence of a connection candidate and information related to the connection candidate for a selected entity included in the connection candidate, and presenting it via the terminal; An apparatus for generating prompt statements that instruct a generative artificial intelligence processing unit to analyze the interests and skills of a user subject, taking the interest information, the skill information, the similarity index, and information about the connection candidates as input information; A means for generating prompt statements that instruct the generative artificial intelligence processing unit to clarify the needs of the user. A device for generating prompt statements that instruct the generative artificial intelligence processing unit to analyze the emotional state of the user and provide suggestions; A device for updating the selection criteria of the connection candidates and / or the content of the notification message based on the response information output by the generative artificial intelligence processing unit, thereby generating communication support information to facilitate communication between the users.

[0609] (Note 2) The information processing system according to Appendix 1 is characterized in that, The processing device control unit is configured to send the prompt statement to the generative artificial intelligence processing unit, and adjust the generation of the feature quantity and the calculation method of the similarity index based on the response information obtained from the generative artificial intelligence processing unit, thereby dynamically optimizing the selection of connection candidates between the subjects.

[0610] (Note 3) The information processing system according to Appendix 1 is characterized in that, The processing device control unit is configured to store the behavioral history information and attribute information of the user in the storage area, and calculate the similarity index by combining the behavioral history information and the attribute information with the interest information and the skill information to discover potential connection needs, and to enable the generative artificial intelligence processing unit to generate response information containing explanatory text and suggestion text about the potential connection needs.

[0611] Application Example 1 (Note 1) An information processing system, characterized in that it comprises: Devices for controlling the acquisition and storage of user attribute information; A device for calculating the similarity between users and extracting multiple collaboration candidates based on the user attribute information; A device for sending relevant information of the plurality of collaborative candidates to a communication terminal and presenting it to the user through the communication terminal; A device for managing bidirectional communication between the user and the collaboration candidate and establishing synchronous dialogue in virtual space; A device for generating prompt statements that are input to a generative artificial intelligence model based on the user attribute information and the content of the two-way communication, and for instructing the specification of the user's needs and the construction of dialogue content through the prompt statements; A device for sending dialogue text candidates contained in the generated results obtained from the generative artificial intelligence model to the communication terminal, and enabling the user to use the dialogue text candidates as communication content sent to the collaborative candidates by selecting or editing them; An apparatus for parsing the user attribute information and the historical information of the two-way communication to infer potential relationship needs, and for determining new collaboration candidates or information resources based on the potential relationship needs.

[0612] (Note 2) The information processing system according to Appendix 1 is characterized in that, The user attribute information includes interest information and ability information. The system is configured to store the interest information and ability information in an information management structure in a memory, and to calculate the similarity based on the common elements in the interest information and ability information.

[0613] (Note 3) The information processing system according to Appendix 1 is characterized in that, The system is configured to extract content containing ambiguous demand expressions from the historical information of the two-way communication, and incorporate the ambiguous demand expressions into prompt statements provided as input to the generative artificial intelligence model, so that the generative artificial intelligence model generates dialogue text candidates for progressively specifying the ambiguous demand expressions.

[0614] Example 2 (Note 1) An information processing system, characterized in that it comprises: A unit used in an information processing device to receive and store natural language abstraction requirement information obtained from a user device; This is used to generate prompt statements for input into a generative artificial intelligence model based on the abstract demand information, and to use the generative artificial intelligence model to parse and process the abstract demand information, and to generate a unit for specifying question information based on the parsing result. A unit for sending the question information to the user device, obtaining the user's answer information to the question information through the user device, and storing the answer information in association with the abstract requirement information; The system is used to generate additional prompts for inputting into the generative artificial intelligence model based on the historical records of the answer information and the question information, and to use the generative artificial intelligence model to generate additional question information or units for progressively specifying the abstract requirement information or specifying the requirement information. A unit for extracting collaborative object information from a data storage device storing a set of candidate object information with attribute information based on the specified requirement information, and selecting collaborative object information according to the correspondence between the specified requirement information and the collaborative object information. A unit for sending the specific requirements information and the information of the collaborating parties to the user device and outputting the information.

[0615] (Note 2) According to the information processing system described in Appendix 1, the information processing device sets the model used as the generative artificial intelligence model as a language model that takes natural language text as input and outputs natural language text, and uses the input text containing the prompt statement and the abstract requirement information to generate the question information and the specific requirement information from the language model.

[0616] (Note 3) According to the information processing system described in Appendix 1, the information processing device stores the answer information and the specific requirement information in association with user identification information, time information and session state information, and controls the content and quantity of context information included in the appended prompt statement based on the stored session state information.

[0617] Application Example 2 (Note 1) An information processing system, characterized in that it comprises: The natural language parsing unit, executed by the computing device, is configured to parse natural language request statements based on user input and behavioral information, and determine the request category and request clarity. The prompt statement generation unit executed by the computing device is configured to generate a prompt statement for inputting into a generative artificial intelligence model based on the request category and the request clarity, send the prompt statement to the generative artificial intelligence model, and obtain a dialogue statement for engaging in dialogue with the user from the generative artificial intelligence model. The phased requirement refinement unit executed by the computing device is configured to send the acquired dialogue statement to the user terminal, receive the user response statement from the user terminal, and provide the user response statement to the natural language parsing unit again to perform phased high-precision refinement of the user's requirement content. The candidate object evaluation unit, executed by the computing device, is configured to retrieve candidate object information from an information storage device storing user behavior information and attribute information, evaluate the candidate object information based on the highly refined demand content and the user's emotional state, and select recommended object information based on the evaluation results. The recommendation description generation unit, executed by the computing device, is configured to generate a prompt statement for obtaining a description statement related to the information of the recommended object, input the prompt statement into the generative artificial intelligence model, obtain the recommendation description statement from the generative artificial intelligence model, associate the recommendation description statement with the information of the recommended object, and send it to the user terminal. The user characteristic update unit, executed by the computing device, is configured to receive user operation history from the user terminal for the recommended object information, store the user operation history in the information storage device, and update the user characteristic information used for the evaluation and the generation of the prompt statement based on the storage result.

[0618] (Note 2) The information processing system according to Appendix 1 is characterized in that, The natural language parsing unit is configured to extract keywords, intents, and sentiment features from user input statements using statistical language processing and machine learning programs during natural language parsing, and input the extracted keywords, intents, and sentiment features as generation conditions for generating the prompt statements into the generative artificial intelligence model.

[0619] (Note 3) The information processing system according to Appendix 1 is characterized in that, The candidate object evaluation unit is configured to perform clustering or similarity calculation processing to classify users using user behavior history information and attribute information stored in the information storage device, generate feature quantities representing potential relationship needs or potential purchase needs based on the processing results, and reflect the generated feature quantities in the generation of the prompt statement and the selection of the recommended object information.

Claims

1. An information processing system, characterized in that, include: Processor, wherein the processor is configured to: Generate prompts to indicate user interests and skills in the analysis; Generate prompts to clarify user needs; Generate prompts to indicate the user's emotional state and make suggestions accordingly.

2. The information processing system according to claim 1, characterized in that, The processor is configured to analyze the user's interests and skills using a generative artificial intelligence model.

3. The information processing system according to claim 1, characterized in that, The processor is configured to analyze user behavior history and user profile information to uncover potential connectivity needs.

Images