Intelligent man-machine interaction system and man-machine interaction method

Abstract

The invention relates to the technical field of man-machine interaction, in particular to an intelligent man-machine interaction system which comprises a network interface module, an Action Agent module, a data conversion module, a knowledge management system, a conversion engine, a system recognition module and a knowledge query and integration interface. According to the method, complex natural language instructions including fuzzy semantics, special contexts and newly appearing expression forms can be accurately processed, so that the intention of a user is more accurately analyzed, services and operation results better meeting the requirements of the user are provided for the user, a universal mechanism is designed, and the user experience is improved. Natural language instructions of users can be conveniently converted into operations on different systems (such as APIs, SQL and BIs), dependence on specific conversion tools or middleware is reduced, development and maintenance cost is reduced, accuracy and integrity of data conversion are ensured, an RAG knowledge base and a knowledge graph technology are effectively integrated, association between knowledge is fully utilized, and the user experience is improved. The comprehensiveness and depth of answering the user questions are improved, and the user interaction experience is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of human-computer interaction, and in particular to an intelligent human-computer interaction system and a human-computer interaction method. Background Art

[0002] With the rapid development of Internet technology, the application of human-computer interaction in web environment is becoming more and more extensive. At present, human-computer interaction systems are mainly based on traditional text input and simple command response modes. In terms of information perception, users mainly input content into the text input box on the web page, and the system receives and processes these text information. For the recognition of user intentions, rule-based or simple machine learning models are mostly used, such as determining the user's possible needs based on keyword matching. In terms of data processing, existing systems have limitations in processing cross-system data interactions. For example, when it comes to converting users' natural language instructions into operations on different systems (such as APIs, SQL databases, business intelligence (BI) systems), complex manual configuration or the use of conversion tools specifically for a single system is often required. In terms of knowledge utilization, some systems have begun to use knowledge bases, but most of them are simple retrieval knowledge bases, lacking in-depth integration and utilization of knowledge. For example, the application of RAG (retrieval enhanced generation) knowledge bases and knowledge graph technology in human-computer interaction is not mature enough.

[0003] The existing technology includes rule-based human-computer interaction systems, human-computer interaction systems assisted by simple machine learning models, and partial data cross-system interaction solutions; wherein the rule-based human-computer interaction system parses the text input by the user through pre-set rules. For example, in a simple web customer service system, when the user enters a keyword such as "query order status", the system will query and return the corresponding order information in the order database according to the preset rules. The implementation method is to establish a keyword-operation mapping table to associate the keywords entered by the user with the corresponding database query operation or function module call; the human-computer interaction system assisted by a simple machine learning model uses a machine learning model, such as a naive Bayes classifier or a decision tree model, to classify the text entered by the user. For example, in an online shopping website, the model can determine whether the text entered by the user is a product query, a purchase operation, or an after-sales service request, and then guide the user to the corresponding function page or perform the corresponding operation based on the classification result. These models are usually trained based on a large amount of historical user data to improve the accuracy of classification;In some enterprise applications, there are limited cross-system data interaction tools. For example, specially developed middleware is used to implement the conversion from text instructions entered in the user interface to specific APIs. This middleware usually needs to be customized for specific APIs, and different types of APIs (such as APIs developed by different companies with different interface specifications) need to be redeveloped or configured. Rule-based methods cannot handle complex natural language expressions and semantic changes. For example, when a user enters "I want to see the logistics information of something I recently purchased, including the estimated delivery time and transportation route", simple keyword matching may only focus on "logistics information" and ignore other key details, resulting in incomplete or inaccurate query results. Although simple machine learning models can process semantics to a certain extent, they have limited generalization capabilities for some newly emerging, semantically ambiguous or special contextual user instructions, and are prone to misclassification or inability to accurately understand user intent. Existing cross-system data interaction solutions lack universality and are not suitable for different systems (such as API, SQL, BI, etc.). ) Different conversion tools or middleware are required, and the development and maintenance costs are high. Moreover, when new systems or new API interfaces appear, it is difficult to quickly achieve compatibility and data interaction. Information loss or format mismatch problems are prone to occur during data conversion. For example, when converting the user's natural language into SQL query statements, the generated SQL statements may have logical errors or fail to correctly handle complex query conditions due to semantic understanding deviations. Traditional knowledge base retrieval methods cannot fully utilize the associations between knowledge. When answering user questions, relevant knowledge cannot be effectively integrated, resulting in incomplete and in-depth answers. For example, when users ask questions about product use and maintenance, simply relying on simple keyword retrieval of the knowledge base may not be able to organically combine relevant knowledge such as product function introduction and common fault handling. The application of RAG knowledge base and knowledge graph technology in existing systems is immature and has not fully utilized its potential in enhancing user interaction experience. For example, it is impossible to dynamically obtain relevant knowledge from the knowledge graph based on the user's context information to improve the processing of user instructions. ;

[0004] Therefore, to address the above-mentioned problems of inaccurate user intent recognition, difficulty in cross-system data flow, and insufficient knowledge utilization, an intelligent human-computer interaction system and human-computer interaction method can be designed. Summary of the invention

[0005] In order to overcome the problems of inaccurate user intent identification, difficulty in cross-system data flow and insufficient knowledge utilization.

[0006] The technical solution of the present invention is: an intelligent human-computer interaction system, including a network interface module, an Action Agent module, a data conversion module, a knowledge management system, a conversion engine, a system identification module, a knowledge query and integration interface;

[0007] The network interface module is used to receive user input, receiving user instructions from a text input box on a web page or other input devices;

[0008] The Action Agent module integrates a pre-trained language model based on the Transformer architecture to accurately process natural language instructions related to web applications. The Action Agent module includes a functional unit, a receiving unit, and an intent analysis unit.

[0009] The data conversion module automatically identifies the target system involved in the user's instruction, and accurately converts the natural language instruction into an operation statement or request acceptable to the corresponding system according to the predefined mapping rules and semantic analysis results;

[0010] The knowledge management system integrates the RAG knowledge base and the knowledge graph. When processing user instructions, the Action Agent module queries and integrates relevant knowledge in the knowledge management system based on the instruction content and user context information to provide users with more comprehensive and accurate answers. The RAG knowledge base preprocesses and indexes text materials (such as product documents, user manuals, and FAQs) for fast retrieval. The knowledge graph models the entity relationships between knowledge, such as the relationship between products and functions, and between faults and solutions.

[0011] The conversion engine contains a predefined grammar and semantic rule library, which includes mapping rules from natural language to API call specifications, SQL query syntax, and BI operation instructions;

[0012] The system identification module determines the target system type involved in the instruction through preliminary analysis of the user instruction;

[0013] When the Action Agent module processes user instructions, it queries the knowledge management system for relevant knowledge through this interface.

[0014] A human-computer interaction method, which includes the intelligent human-computer interaction system as described above, comprises the following steps:

[0015] S1: User input command

[0016] S11: The user inputs specific instructions to the system through a corresponding input interface, such as keyboard input or voice input. The instruction content is various demand expressions such as completing a task or obtaining specific information;

[0017] S12: The user input command flows from the client to the Action Agent module in the form of text or other suitable data format, serving as the starting data of the entire process;

[0018] S2: Action Agent module receives instructions

[0019] S21: The Action Agent module has a functional unit for receiving user input commands, monitors the user input interface in real time, and once a command is input, it obtains it and temporarily stores it in the internal cache area for subsequent processing;

[0020] S21: The command input by the user is completely transmitted from the user input interface to the receiving unit of the Action Agent module. The data flow is unidirectional, flowing from the external client to the Action Agent module.

[0021] S3: Perform intent analysis to determine whether cross-system operations are involved

[0022] S31: Operation content

[0023] S311: The intention analysis unit in the Action Agent module will perform semantic understanding and analysis on the received instructions, and use natural language processing technology to parse the real intention of the instructions and determine the specific task content that the user wants the system to complete. Natural language processing technology includes but is not limited to lexical analysis, syntactic analysis, and semantic role labeling;

[0024] S312: At the same time, based on the preliminary understanding of the task and the system predefined rules and RAG knowledge base information, it is determined whether the task needs to involve cross-system operations. If the task involves calling resources and services of other external systems or interacting with other independently running subsystems, then it is determined to involve cross-system operations. Otherwise, it is determined not to involve cross-system operations.

[0025] S32: The input command data is transmitted from the receiving unit to the intention analysis unit in the Action Agent module. The intention analysis unit refers to the predefined rule base and RAG knowledge base data in the Action Agent module during the analysis process. It mainly processes the input command. The analysis result will flow to different subsequent branch processes. The analysis result is a yes or no judgment result and a description of the understanding of the intention.

[0026] S4: Enter the data conversion module to convert instructions

[0027] S41: When the judgment result is that the cross-system operation is involved, the instruction will be transmitted to the data conversion module, which will perform corresponding conversion operations on the original instruction according to the interface specifications and data format requirements of the target cross-system to ensure that the instruction can be correctly executed in the cross-system environment;

[0028] S42: the original instruction data transmitted from the intention analysis unit enters the data conversion module, and the converted instruction data flows out of the data conversion module in a format that meets the cross-system requirements, and flows to the subsequent module that needs to use the converted instruction, that is, the Action Agent module queries the knowledge management system for knowledge and task list modules;

[0029] S5: Action Agent module queries the knowledge management system for knowledge and task lists

[0030] S51: Operation content

[0031] S511: Regardless of whether cross-system operations are involved, the Action Agent module interacts with the knowledge management system and sends a query request to the knowledge management system based on the task intent obtained from the previous analysis and the converted instruction content;

[0032] S512: After receiving the request, the knowledge management system searches for knowledge information and a task list related to the task in its internal RAG knowledge base;

[0033] S513: The Action Agent module receives the query results returned by the knowledge management system and temporarily stores the results in its own corresponding cache area for reference in subsequent task execution;

[0034] S52: the instruction-related data transmitted from the data conversion module or the intention analysis unit is sent from the Action Agent module to the knowledge management system as part of the query request, and the knowledge and task list data queried by the knowledge management system are returned from the knowledge management system to the Action Agent module, and the data flow is a two-way interaction;

[0035] S6: Execute the mission

[0036] S61: The Action Agent module actually executes the task based on the knowledge and task list obtained from the knowledge management system, combined with its own execution capabilities and algorithm logic;

[0037] S62: During the task execution, there will be multiple data flows. On the one hand, the Action Agent module reads relevant data from its internal cache area for task execution; on the other hand, new data will be generated during the execution process. These newly generated data will flow between different modules within the Action Agent module to support the continuous advancement of the task until the task execution is completed;

[0038] S7: Integrate the processing results

[0039] S71: When the task is completed, the Action Agent module collects and integrates various result data generated during the execution process. The integration operation involves unifying the format of the data, filtering the content, and arranging it in a certain logical order, so as to present the processing results in a clear, complete form that meets user needs;

[0040] S72: The result data collected from each module executing the task and the cache area flows to the integration module, and the integrated data flows out of the integration module in the form of final processing results, ready to be returned to the user;

[0041] S8: Return the processing result to the user

[0042] S81: The integrated processing results are presented to the user in an appropriate manner through the output interface of the Action Agent module, so that the user can intuitively understand the task status completed by the system after the user inputs the command and the final answer obtained;

[0043] S82: The integrated processing result is transmitted from the integration module inside the Action Agent module to the client through the output interface. The data flow is unidirectional, flowing from the Action Agent module to the user.

[0044] Preferably, in S41, corresponding conversion operations are performed on the original instructions, including adjustment of the data format, conversion of one data structure into another data structure that meets the requirements of the target system, adaptation of the instruction syntax, and reorganization of the instruction content according to the instruction syntax rules that the target system can understand.

[0045] Preferably, the knowledge information in S512 includes but is not limited to concept explanations in related fields, historical execution records, and solutions to similar tasks, and the task list includes priority ranking of tasks and suggestions for the order in which tasks are to be executed.

[0046] Preferably, the actual execution of the task in S61 involves calling various functional modules, algorithm programs, etc. within itself to complete the specific steps of the task, such as data processing, calculation operations, further interaction with other related systems or modules, etc. The specific execution content depends on the nature and requirements of the task.

[0047] Preferably, the cache area in S62 is a cache for storing knowledge and task lists; new data includes intermediate calculation results, new information obtained through interaction with other systems, etc.

[0048] Preferably, the various result data in S71 include final results and intermediate process data reflecting relevant process information.

[0049] Preferably, the processing results integrated in S81 are presented to the user in a suitable manner through the output interface of the Action Agent module, and the results may be displayed as text on a display screen or as voice broadcast results.

[0050] The beneficial effects of the present invention are as follows: it can accurately process complex natural language instructions, including those with semantic ambiguity, special contexts and newly emerging expressions, so as to more accurately parse user intentions and provide users with services and operation results that better meet their needs. It designs a universal mechanism that can conveniently convert users' natural language instructions into operations for different systems (such as API, SQL, BI), reduce dependence on specific conversion tools or middleware, reduce development and maintenance costs, and ensure the accuracy and completeness of data conversion. It effectively integrates RAG knowledge base and knowledge graph technology, makes full use of the relationship between knowledge, and dynamically acquires and utilizes relevant knowledge according to the user's context and instructions, thereby improving the comprehensiveness and depth of answers to user questions and enhancing the user interaction experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 The following is a schematic diagram of the framework structure of the intelligent human-computer interaction system of the present invention;

[0052] Figure 2 What is shown is the workflow diagram of the human-computer interaction method of the present invention. DETAILED DESCRIPTION

[0053] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0054] See also Figure 1-Figure 2 ,The present invention provides an embodiment: an intelligent human-computer interaction system, including a network interface module, an Action Agent module, a data conversion module, a knowledge management system, a conversion engine, a system identification module, a knowledge query and integration interface;

[0055] The network interface module is used to receive user input, receiving user instructions from a text input box on a web page or other input devices;

[0056] The Action Agent module integrates a pre-trained language model based on the Transformer architecture to accurately process natural language instructions related to web applications. The Action Agent module includes a functional unit, a receiving unit, and an intent analysis unit.

[0057] The data conversion module automatically identifies the target system involved in the user's instruction, and accurately converts the natural language instruction into an operation statement or request acceptable to the corresponding system according to the predefined mapping rules and semantic analysis results;

[0058] The knowledge management system integrates the RAG knowledge base and the knowledge graph. When processing user instructions, the Action Agent module queries and integrates relevant knowledge in the knowledge management system based on the instruction content and user context information to provide users with more comprehensive and accurate answers. The RAG knowledge base preprocesses and indexes text materials (such as product documents, user manuals, and FAQs) for fast retrieval. The knowledge graph models the entity relationships between knowledge, such as the relationship between products and functions, and between faults and solutions.

[0059] The conversion engine contains a predefined grammar and semantic rule library, which includes mapping rules from natural language to API call specifications, SQL query syntax, and BI operation instructions;

[0060] The system identification module determines the target system type involved in the instruction through preliminary analysis of the user instruction;

[0061] When the Action Agent module processes user instructions, it queries the knowledge management system for relevant knowledge through this interface.

[0062] A human-computer interaction method, which includes the intelligent human-computer interaction system as described above, comprises the following steps:

[0063] S1: User input command

[0064] S11: The user inputs specific instructions to the system through a corresponding input interface, such as keyboard input or voice input. The instruction content is various demand expressions such as completing a task or obtaining specific information;

[0065] S12: The user input command flows from the client to the Action Agent module in the form of text or other suitable data format, serving as the starting data of the entire process;

[0066] S2: Action Agent module receives instructions

[0067] S21: The Action Agent module has a functional unit for receiving user input commands, monitors the user input interface in real time, and once a command is input, it obtains it and temporarily stores it in the internal cache area for subsequent processing;

[0068] S21: The command input by the user is completely transmitted from the user input interface to the receiving unit of the Action Agent module. The data flow is unidirectional, flowing from the external client to the Action Agent module.

[0069] S3: Perform intent analysis to determine whether cross-system operations are involved

[0070] S31: Operation content

[0071] S311: The intention analysis unit in the Action Agent module will perform semantic understanding and analysis on the received instructions, and use natural language processing technology to parse the real intention of the instructions and determine the specific task content that the user wants the system to complete. Natural language processing technology includes but is not limited to lexical analysis, syntactic analysis, and semantic role labeling;

[0072] S312: At the same time, based on the preliminary understanding of the task and the system predefined rules and RAG knowledge base information, it is determined whether the task needs to involve cross-system operations. If the task involves calling resources and services of other external systems or interacting with other independently running subsystems, then it is determined to involve cross-system operations. Otherwise, it is determined not to involve cross-system operations.

[0073] S32: The input command data is transmitted from the receiving unit to the intention analysis unit in the Action Agent module. The intention analysis unit refers to the predefined rule base and RAG knowledge base data in the Action Agent module during the analysis process. It mainly processes the input command. The analysis result will flow to different subsequent branch processes. The analysis result is a yes or no judgment result and a description of the understanding of the intention.

[0074] S4: Enter the data conversion module to convert instructions

[0075] S41: When the judgment result is that the cross-system operation is involved, the instruction will be transmitted to the data conversion module, which will perform corresponding conversion operations on the original instruction according to the interface specifications and data format requirements of the target cross-system to ensure that the instruction can be correctly executed in the cross-system environment;

[0076] S42: the original instruction data transmitted from the intention analysis unit enters the data conversion module, and the converted instruction data flows out of the data conversion module in a format that meets the cross-system requirements, and flows to the subsequent module that needs to use the converted instruction, that is, the Action Agent module queries the knowledge management system for knowledge and task list modules;

[0077] S5: Action Agent module queries the knowledge management system for knowledge and task lists

[0078] S51: Operation content

[0079] S511: Regardless of whether cross-system operations are involved, the Action Agent module interacts with the knowledge management system and sends a query request to the knowledge management system based on the task intent obtained from the previous analysis and the converted instruction content;

[0080] S512: After receiving the request, the knowledge management system searches for knowledge information and a task list related to the task in its internal RAG knowledge base;

[0081] S513: The Action Agent module receives the query results returned by the knowledge management system and temporarily stores the results in its own corresponding cache area for reference in subsequent task execution;

[0082] S52: the instruction-related data transmitted from the data conversion module or the intention analysis unit is sent from the Action Agent module to the knowledge management system as part of the query request, and the knowledge and task list data queried by the knowledge management system are returned from the knowledge management system to the Action Agent module, and the data flow is a two-way interaction;

[0083] S6: Execute the mission

[0084] S61: The Action Agent module actually executes the task based on the knowledge and task list obtained from the knowledge management system, combined with its own execution capabilities and algorithm logic;

[0085] S62: During the task execution, there will be multiple data flows. On the one hand, the Action Agent module reads relevant data from its internal cache area for task execution; on the other hand, new data will be generated during the execution process. These newly generated data will flow between different modules within the Action Agent module to support the continuous advancement of the task until the task execution is completed;

[0086] S7: Integrate the processing results

[0087] S71: When the task is completed, the Action Agent module collects and integrates various result data generated during the execution process. The integration operation involves unifying the format of the data, filtering the content, and arranging it in a certain logical order, so as to present the processing results in a clear, complete form that meets user needs;

[0088] S72: The result data collected from each module executing the task and the cache area flows to the integration module, and the integrated data flows out of the integration module in the form of final processing results, ready to be returned to the user;

[0089] S8: Return the processing result to the user

[0090] S81: The integrated processing results are presented to the user in an appropriate manner through the output interface of the Action Agent module, so that the user can intuitively understand the task status completed by the system after the user inputs the command and the final answer obtained;

[0091] S82: The integrated processing result is transmitted from the integration module inside the Action Agent module to the client through the output interface. The data flow is unidirectional, flowing from the Action Agent module to the user.

[0092] Preferably, in S41, corresponding conversion operations are performed on the original instructions, including adjustment of the data format, conversion of one data structure into another data structure that meets the requirements of the target system, adaptation of the instruction syntax, and reorganization of the instruction content according to the instruction syntax rules that the target system can understand.

[0093] Preferably, the knowledge information in S512 includes but is not limited to concept explanations in related fields, historical execution records, and solutions to similar tasks, and the task list includes priority ranking of tasks and suggestions for the order in which tasks are to be executed.

[0094] Preferably, the actual execution of the task in S61 involves calling various functional modules, algorithm programs, etc. within itself to complete the specific steps of the task, such as data processing, calculation operations, further interaction with other related systems or modules, etc. The specific execution content depends on the nature and requirements of the task.

[0095] Preferably, the cache area in S62 is a cache for storing knowledge and task lists; new data includes intermediate calculation results, new information obtained through interaction with other systems, etc.

[0096] Preferably, the various result data in S71 include final results and intermediate process data reflecting relevant process information.

[0097] Preferably, the processing results integrated in S81 are presented to the user in a suitable manner through the output interface of the Action Agent module, and the results may be displayed as text on a display screen or as voice broadcast results.

[0098] Through the above steps, complex natural language instructions can be processed accurately, so as to more accurately analyze user intentions and provide users with services and operation results that better meet their needs. It can easily convert users' natural language instructions into operations for different systems, effectively integrate RAG knowledge base and knowledge graph technology, make full use of the relationship between knowledge, dynamically acquire and utilize relevant knowledge according to user context and instructions, improve the comprehensiveness and depth of answers to user questions, and enhance user interaction experience to solve the problems of inaccurate user intention identification, difficulty in cross-system data flow and insufficient knowledge utilization.

[0099] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of those skilled in the art without departing from the spirit of the present invention.

Claims

1. An intelligent human-computer interaction system; characterized in that: It includes network interface module, Action Agent module, data conversion module, knowledge management system, conversion engine, system identification module, knowledge query and integration interface; The network interface module is used to receive user input, receiving user instructions from a text input box on a web page or other input devices; The Action Agent module integrates a pre-trained language model based on the Transformer architecture to accurately process natural language instructions related to web applications. The Action Agent module includes a functional unit, a receiving unit, and an intent analysis unit. The data conversion module automatically identifies the target system involved in the user's instruction, and accurately converts the natural language instruction into an operation statement or request acceptable to the corresponding system according to the predefined mapping rules and semantic analysis results; The knowledge management system integrates the RAG knowledge base and the knowledge graph. When processing user instructions, the Action Agent module queries and integrates relevant knowledge in the knowledge management system based on the instruction content and user context information to provide users with more comprehensive and accurate answers. The RAG knowledge base preprocesses and indexes text materials (such as product documents, user manuals, and FAQs) for fast retrieval. The knowledge graph models the entity relationships between knowledge, such as the relationship between products and functions, and between faults and solutions. The conversion engine contains a predefined grammar and semantic rule library, which includes mapping rules from natural language to API call specifications, SQL query syntax, and BI operation instructions; The system identification module determines the target system type involved in the instruction through preliminary analysis of the user instruction; When the Action Agent module processes user instructions, it queries the knowledge management system for relevant knowledge through this interface.

2. A human-computer interaction method, characterized in that The intelligent human-computer interaction system according to claim 1 comprises the following steps: S1: User input command S11: The user inputs specific instructions to the system through a corresponding input interface, such as keyboard input or voice input. The instruction content is various demand expressions such as completing a task or obtaining specific information; S12: The user input command flows from the client to the Action Agent module in the form of text or other suitable data format, serving as the starting data of the entire process; S2: Action Agent module receives instructions S21: The Action Agent module has a functional unit for receiving user input commands, monitors the user input interface in real time, and once a command is input, it obtains it and temporarily stores it in the internal cache area for subsequent processing; S21: The command input by the user is completely transmitted from the user input interface to the receiving unit of the Action Agent module. The data flow is unidirectional, flowing from the external client to the Action Agent module. S3: Perform intent analysis to determine whether cross-system operations are involved S31: Operation content S311: The intention analysis unit in the Action Agent module will perform semantic understanding and analysis on the received instructions, and use natural language processing technology to parse the real intention of the instructions and determine the specific task content that the user wants the system to complete. Natural language processing technology includes but is not limited to lexical analysis, syntactic analysis, and semantic role labeling; S312: At the same time, based on the preliminary understanding of the task and the system predefined rules and RAG knowledge base information, it is determined whether the task needs to involve cross-system operations. If the task involves calling resources and services of other external systems or interacting with other independently running subsystems, then it is determined to involve cross-system operations. Otherwise, it is determined not to involve cross-system operations. S32: The input command data is transmitted from the receiving unit to the intention analysis unit in the Action Agent module. The intention analysis unit refers to the predefined rule base and RAG knowledge base data in the Action Agent module during the analysis process. It mainly processes the input command. The analysis result will flow to different subsequent branch processes. The analysis result is a yes or no judgment result and a description of the understanding of the intention. S4: Enter the data conversion module to convert instructions S41: When the judgment result is that the cross-system operation is involved, the instruction will be transmitted to the data conversion module, which will perform corresponding conversion operations on the original instruction according to the interface specifications and data format requirements of the target cross-system to ensure that the instruction can be correctly executed in the cross-system environment; S42: the original instruction data transmitted from the intention analysis unit enters the data conversion module, and the converted instruction data flows out of the data conversion module in a format that meets the cross-system requirements, and flows to the subsequent module that needs to use the converted instruction, that is, the Action Agent module queries the knowledge management system for knowledge and task list modules; S5: Action Agent module queries the knowledge management system for knowledge and task lists S51: Operation content S511: Regardless of whether cross-system operations are involved, the Action Agent module interacts with the knowledge management system and sends a query request to the knowledge management system based on the task intent obtained from the previous analysis and the converted instruction content; S512: After receiving the request, the knowledge management system searches for knowledge information and a task list related to the task in its internal RAG knowledge base; S513: The Action Agent module receives the query results returned by the knowledge management system and temporarily stores the results in its own corresponding cache area for reference in subsequent task execution; S52: the instruction-related data transmitted from the data conversion module or the intention analysis unit is sent from the Action Agent module to the knowledge management system as part of the query request, and the knowledge and task list data queried by the knowledge management system are returned from the knowledge management system to the Action Agent module, and the data flow is a two-way interaction; S6: Execute the mission S61: The Action Agent module actually executes the task based on the knowledge and task list obtained from the knowledge management system, combined with its own execution capabilities and algorithm logic; S62: During the task execution, there will be multiple data flows. On the one hand, the Action Agent module reads relevant data from its internal cache area for task execution; on the other hand, new data will be generated during the execution process. These newly generated data will flow between different modules within the Action Agent module to support the continuous advancement of the task until the task execution is completed; S7: Integrate the processing results S71: When the task is completed, the Action Agent module collects and integrates various result data generated during the execution process. The integration operation involves unifying the format of the data, filtering the content, and arranging it in a certain logical order, so as to present the processing results in a clear, complete form that meets user needs; S72: The result data collected from each module executing the task and the cache area flows to the integration module, and the integrated data flows out of the integration module in the form of final processing results, ready to be returned to the user; S8: Return the processing result to the user S81: The integrated processing results are presented to the user in an appropriate manner through the output interface of the Action Agent module, so that the user can intuitively understand the task status completed by the system after the user inputs the command and the final answer obtained; S82: The integrated processing result is transmitted from the integration module inside the Action Agent module to the client through the output interface. The data flow is unidirectional, flowing from the Action Agent module to the user.

3. The human-computer interaction method according to claim 2, characterized in that: In S41, the original instruction is converted accordingly, including adjusting the data format, converting one data structure into another data structure that meets the requirements of the target system, adapting the instruction syntax, and reorganizing the instruction content according to the instruction syntax rules that the target system can understand.

4. The human-computer interaction method according to claim 2, characterized in that: The knowledge information in S512 includes, but is not limited to, concept explanations in related fields, historical execution records, and solutions to similar tasks. The task list includes the priority ranking of tasks and the suggested order of task execution.

5. The human-computer interaction method according to claim 2, characterized in that: The actual execution of the task in S61 involves calling various functional modules, algorithm programs, etc. within itself to complete the specific steps of the task, such as data processing, calculation operations, and further interaction with other related systems or modules. The specific execution content depends on the nature and requirements of the task.

6. The human-computer interaction method according to claim 2, characterized in that: The cache area in S62 is a cache for storing knowledge and task lists; new data includes intermediate calculation results, new information obtained through interaction with other systems, etc.

7. The human-computer interaction method according to claim 2, characterized in that: The various result data in S71 include final results and intermediate process data reflecting relevant process information.

8. The human-computer interaction method according to claim 2, characterized in that: The processing results integrated in S81 are presented to the user in a suitable manner through the output interface of the Action Agent module, and the results may be displayed as text on a display screen or as voice broadcasts.

Images