A boundary control method and system for intelligent dialogue management

Abstract

The present invention discloses a boundary control method and system for intelligent dialogue management, which includes: obtaining and parsing a description file related to making dialogue scene management; preprocessing the parsing result; using the processed data as training data to train a large model to obtain an industry large model with good evaluation performance; constructing a corresponding prompt scene template according to the usage scenario corresponding to the trained industry large model; matching the corresponding prompt scene template according to the user's question query during the dialogue interaction process, and receiving the user's modification feedback result of the prompt scene template. The method preprocesses the description file and constructs a private knowledge base; by adding a private knowledge base to the large model, multiple rounds of dialogue with the user are realized in complex scenarios; it has the ability to solve complex reasoning problems; even if errors occur, effective error correction can be achieved through memory and feedback mechanisms; so as to improve the user's interactive experience.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent dialogue, and in particular to a boundary control method and system for intelligent dialogue management. Background Art

[0002] Intelligent dialogue is in the form of one question and one answer, accurately locates the answer required by the user's question, and provides personalized information services through interaction with the user. With the continuous development of intelligent language processing technology, the business scenarios applicable to intelligent dialogue are becoming more and more extensive, playing a fundamental role in business fields such as intelligent customer service and smart education.

[0003] The common intelligent dialogue systems on the market are usually based on natural language processing technology. However, due to the diversity of user questions and the complexity of usage scenarios, existing dialogue systems are unable to conduct complex multi-round dialogues or solve complex reasoning problems, and lack memory and feedback mechanisms, making it impossible to perform automatic and self-service corrections.

[0004] Therefore, the intelligent dialogue systems in the existing technology are unable to conduct complex multi-round dialogues or solve complex reasoning problems, and are unable to effectively manage dialogue boundary control issues, which need to be urgently addressed by fellow practitioners. Summary of the invention

[0005] The purpose of the present invention is to provide a boundary control method and system for intelligent dialogue management, so as to solve the problem that the intelligent dialogue system in the prior art cannot conduct complex multi-round dialogues or solve complex reasoning problems, and cannot effectively manage dialogue boundary control problems.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] In a first aspect, an embodiment of the present invention provides a boundary control method for intelligent dialogue management, comprising the following steps:

[0008] S1: Obtain and parse the description files related to making dialogue scene management; pre-process the parsing results;

[0009] S2: Use the processed data as training data to train the large model, and obtain an industry large model with good evaluation performance;

[0010] S3: Build a corresponding prompt scenario template according to the usage scenario corresponding to the trained industry big model;

[0011] S4: Match the corresponding prompt scenario template according to the user's question query during the dialogue interaction process, and receive the user's modification feedback results on the prompt scenario template.

[0012] Furthermore, the step S1 comprises:

[0013] S11, obtaining description files related to making dialogue scene management, using python-docx library, pdfminer library, and chardet library to parse the file structure and format of the description files respectively;

[0014] S12, parsing metadata of the description file;

[0015] S13, extracting text content, style, format and layout information of the description document;

[0016] S14. Handle complex document layout and formatting, including page size, headers, footers, and column layout;

[0017] S15, segmenting and sentence-by-sentence processing of the parsed text content, and converting the result into vector data;

[0018] S16. Obtain corresponding vector data, and obtain marking results of unsatisfactory parts of the vector data to obtain processed vector data.

[0019] Furthermore, the step S2 comprises:

[0020] S21. Configure the relevant parameters of the industry, training rounds, and training performance of the large model;

[0021] S22, importing the processed vector data into the large model, constructing the training task of the model, and starting, monitoring and stopping the model;

[0022] S23. Evaluate the training effect of the large model. When it meets expectations, stop training to obtain an industry large model with better evaluation performance.

[0023] Furthermore, in step S3, the prompt scene template includes: scene setting rules and output rules;

[0024] The scenario setting rules include: context / background information, roles, verbs, task objects, and output length;

[0025] Output rules include: relevant data content sources, writing style and target population.

[0026] Furthermore, the step S4 comprises:

[0027] S41, receiving questions from users regarding content of interest, and converting the questions from users into vector data;

[0028] S42, matching the vector data with the data in the database, and extracting the corresponding prompt question template;

[0029] S43, according to the prompt question template, the trained industry model is queried, and the prompt answer template is obtained according to the selection of the weight library;

[0030] S44. When the user is not satisfied with the prompt answer template, receive the user's feedback on the modification of the prompt question template and update it to the weight library.

[0031] In a second aspect, an embodiment of the present invention further provides a border control system for intelligent dialogue management, including:

[0032] Obtaining a parsing module, used to obtain and parse the description files related to making dialogue scene management; pre-processing the parsing results;

[0033] The training module is used to train the large model using the processed data as training data to obtain an industry large model with better evaluation performance;

[0034] A construction module, used to construct a corresponding prompt scenario template according to the usage scenario corresponding to the trained industry big model;

[0035] The modification feedback module is used to match the corresponding prompt scene template according to the user's question query during the dialogue interaction process, and receive the user's modification feedback results on the prompt scene template.

[0036] In a third aspect, an embodiment of the present invention further provides a computer device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0037] Memory, used to store computer programs;

[0038] The processor is used to implement a boundary control method for intelligent dialogue management as described in any one of the first aspects when executing a program stored in the memory.

[0039] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the processor executes a boundary control method for intelligent dialogue management as described in any one of the first aspects.

[0040] The descriptions of the second to fourth aspects of the present invention can refer to the detailed description of the first aspect; and the beneficial effects of the descriptions of the second to fourth aspects can refer to the analysis of the beneficial effects of the first aspect, which will not be repeated here.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The boundary control method of intelligent dialogue management provided by the embodiment of the present invention includes: obtaining and parsing the description file related to making dialogue scene management; preprocessing the parsing result; using the processed data as training data to train the big model to obtain the industry big model with good evaluation performance; constructing the corresponding prompt scene template according to the usage scenario corresponding to the trained industry big model; matching the corresponding prompt scene template according to the question query of the user in the dialogue interaction process, and receiving the user's modification feedback result of the prompt scene template. The method receives and preprocesses the description file to construct a private knowledge base; by adding the private knowledge base, multiple rounds of dialogue with the user can be realized in complex scenarios. In the process of dialogue, it has the ability to solve complex reasoning problems. Even if errors occur, effective error correction can be achieved through memory and feedback mechanisms; so as to better interact with users and understand user needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A flow chart of a boundary control method for intelligent dialogue management provided by an embodiment of the present invention;

[0044] Figure 2 A flowchart of step S1 provided in an embodiment of the present invention;

[0045] Figure 3 A flowchart of step S2 provided in an embodiment of the present invention;

[0046] Figure 4 A flowchart of step S4 provided in an embodiment of the present invention;

[0047] Figure 5 A schematic diagram of a boundary control method for intelligent dialogue management provided by an embodiment of the present invention;

[0048] Figure 6 A block diagram of a border control system for intelligent dialogue management provided by an embodiment of the present invention;

[0049] Figure 7 A structural block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0051] In the description of the present invention, it should be noted that the terms "upper", "lower", "inner", "outer", "front end", "rear end", "two ends", "one end", "the other end" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0052] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "provided with", "connected", etc. should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0053] Embodiment 1:

[0054] Reference Figure 1 As shown, the present invention provides a boundary control method for intelligent dialogue management, comprising the following steps:

[0055] S1: Obtain and parse the description files related to making dialogue scene management; pre-process the parsing results;

[0056] S2: Use the processed data as training data to train the large model, and obtain an industry large model with good evaluation performance;

[0057] S3: Build a corresponding prompt scenario template according to the usage scenario corresponding to the trained industry big model;

[0058] S4: Match the corresponding prompt scenario template according to the user's question query during the dialogue interaction process, and receive the user's modification feedback results on the prompt scenario template.

[0059] In this embodiment, the method can be applied to an intelligent dialogue system. By constructing a prompt scenario template specific to an industry or usage scenario based on training data and a large model, a more personalized and customized dialogue scenario can be achieved to meet the needs of different users and businesses. By applying this method, the intelligent dialogue system can handle complex multi-round dialogues, allowing users to interact more deeply in the dialogue, rather than just a simple question and answer. This is crucial for solving complex problems and providing more detailed information services. In addition, the method has a memory and feedback mechanism, which can receive user feedback on the modification of the template and automatically correct it. This means that the intelligent dialogue system can continuously improve itself, better meet user needs, and provide a higher quality dialogue experience; thereby better enabling the intelligent dialogue system to interact with users and understand user needs.

[0060] The following is a detailed description of each of the above steps.

[0061] Step 1: Obtain and parse the description files related to making dialogue scene management; pre-process the parsing results;

[0062] like Figure 2 As shown, step S1 includes:

[0063] S11, obtaining description files related to making dialogue scene management, using python-docx library, pdfminer library, and chardet library to parse the file structure and format of the description files respectively;

[0064] S12, parsing metadata of the description file;

[0065] S13, extracting text content, style, format and layout information of the description document;

[0066] S14. Handle complex document layout and formatting, including page size, headers, footers, and column layout;

[0067] S15, segmenting and sentence-by-sentence processing of the parsed text content, and converting the result into vector data;

[0068] S16. Obtain corresponding vector data, and obtain marking results of unsatisfactory parts of the vector data to obtain processed vector data.

[0069] In this embodiment, for example, in the field of intelligent customer service or smart education, business experts upload business files on the front-end page. The file formats include but are not limited to docx, pdf, and txt. The business files contain common business knowledge. After the remote server receives the business file, it reads the file format and calls the corresponding parsing module for parsing according to the file format. The python-docx library, pdfminer library, and chardet library can be used to parse word documents, pdf documents, and text documents, respectively.

[0070] Taking PDF document parsing as an example, 1: Understand the structure and characteristics of PDF documents, including objects, pages, fonts, colors, etc. 2: Parse the metadata of PDF documents. 3: Extract text content, including body, title, header and footer, etc. 4: Process document layout and format, such as page size, header and footer, column layout, etc. 5: Delete useless content in the document. 6: Segment and sentence processing of text. 7: Convert the data split in the previous step into vectors. Among them, it may involve feature transformation, feature importance evaluation, feature selection, feature generation, etc., including but not limited to normalization, standardization, discretization, one-hot encoding, etc.

[0071] After the business document parsing is completed, experts can manually mark the dissatisfied parts of the parsed business document data through the system's built-in expert tag library. The manually processed data is used to develop industry-specific models and stored in the database.

[0072] This step allows you to add a private knowledge base to the large model, thereby implementing an intelligent dialogue system that can provide high-quality answers to questioners even in complex usage scenarios.

[0073] Step 2: Use the processed data as training data to train the large model and obtain an industry large model with good evaluation performance;

[0074] like Figure 3 As shown, step S2 specifically includes:

[0075] S21. Configure the relevant parameters of the industry, training rounds, and training performance of the large model;

[0076] S22, importing the processed vector data into the large model, constructing the training task of the model, and starting, monitoring and stopping the model;

[0077] S23. Evaluate the training effect of the large model. When it meets expectations, stop training to obtain an industry large model with better evaluation performance.

[0078] In this embodiment, for example, the business expert first fine-tunes the parameters of the vicuna-6b model built into the system, then selects the basic industry parameters required by the industry model, and finally calls the data processed in the previous step for model training. After the data results of the model training are output, the automatic evaluation method will be used to evaluate the text quality of the industry model. When the automatic evaluation method calculates that the text generated by the industry model is greater than 0.8 similar to the correct text, the quality assessment of the industry model is qualified. If the automatic evaluation method calculates that the text generated by the industry model is less than 0.8 similar to the correct text, the parameters of the industry model are adjusted and the model is retrained. After the training is completed, the performance of the industry model is re-evaluated until it is satisfactory.

[0079] In the actual implementation process, through the cooperation between technical personnel and industry experts, a suitable and effective industry large model can be obtained through parameter adjustment, model setting, parameter setting, model training and other actions.

[0080] Business experts can configure the industry model's industry, training rounds, training performance and other related parameters. In the configuration unit, they can also configure prompt scenario templates, including [context / background information], [role], [verb], [task and object], [output and length]. The output can include relevant [data and content sources], writing [style], and the [target group / audience] that the content is mainly aimed at.

[0081] Step 3: Build a corresponding prompt scenario template according to the usage scenario corresponding to the trained industry big model;

[0082] For example, on the LangChain-based visual prompt optimization page, business experts set prompt scenario templates for all possible scenarios in which the industry model trained in the previous step may be used. The scenario templates include [context / background information], [roles], [verbs], [tasks and objects], [output and length]. The output can include relevant [data and content sources], writing [style], and the [target population / audience] that the content is mainly aimed at.

[0083] In the specific implementation, the prompt template library is built and applied to the Chain-of-Thought (CoT) project, which is a technology used to improve the reasoning ability of large language models. It solves multi-step problems by prompting the model to generate a series of reasoning steps.

[0084] Step 4: Match the corresponding prompt scenario template according to the user's question query during the dialogue interaction process, and receive the user's modified feedback on the prompt scenario template.

[0085] Reference Figure 4 As shown,

[0086] Step S4 includes:

[0087] S41, receiving questions from users regarding content of interest, and converting the questions from users into vector data;

[0088] S42, matching the vector data with the data in the database, and extracting the corresponding prompt question template;

[0089] S43, according to the prompt question template, the trained industry model is queried, and the prompt answer template is obtained according to the selection of the weight library;

[0090] S44. When the user is not satisfied with the prompt answer template, receive the user's feedback on the modification of the prompt question template and update it to the weight library.

[0091] For example, when a user interacts with an intelligent dialogue system on the user-facing end, the user first asks a question about something they are concerned about. After asking the question, the question is converted into a vector, which is matched with the data in the database, and finally the corresponding prompt question template is extracted from the database.

[0092] In the previous step 3, when building the template library, each template determines the usage scenario and main questions and other related information. Then when the user asks a new question, the input text will be decomposed into words and phrases using natural language processing, the topic of the question will be extracted, and then converted into a vector. Then, the prompt question template with a high similarity to the user's question topic vector will be searched in the database.

[0093] Then, the LLM model is queried based on the prompt question template, and then the prompt answer template is obtained based on the selection of the weight library. If the user is not satisfied with the answer, the prompt question template can be edited.

[0094] Users' edits to the prompt question template will be fed back to the weight library to improve the quality of the next answer response.

[0095] The boundary control method for intelligent dialogue management provided by the present invention is applied to an intelligent dialogue system. Take product after-sales as an example: a user says that his computer cannot access the Internet. At this time, the intelligent dialogue system will feedback multiple prompt answer templates, but each step in each prompt answer template may not completely meet the user's situation. The user can choose one of the solutions and modify a step in the solution.

[0096] for example

[0097] 1. Check the network connection.

[0098] 1.1 Make sure the desktop is connected to the router. You can check if the desktop's network adapter is turned on and connected to the correct network port.

[0099] 1.2 Make sure the router is turned on. You can check the router's indicator lights to make sure all lights are on.

[0100] 1.3 Try restarting the router. You can unplug the router for 30 seconds and then plug it back in.

[0101] 2. Check the network settings.

[0102] 2.1 Make sure the network settings of the desktop computer are correct. You can open the network settings of the desktop computer and check the IP address, subnet mask and DNS server address.

[0103] 2.2 If you are unsure about your network settings, you can try using the settings provided by your network administrator.

[0104] 2.3 You can also try to use the automatic configuration feature to obtain the network settings.

[0105] 3. Contact your network administrator.

[0106] 3.1 If you still cannot connect to the Internet, you can contact your network administrator for help.

[0107] 3.2 The network administrator can help you diagnose network problems and solve your problems.

[0108] The boundary control method of intelligent dialogue management provided by the present invention has an overall process, as shown in FIG. Figure 5 As shown in the figure, the business expert first imports the business files, and then processes these files. Use these uploaded business files to train the industry model. After the industry model training and verification is completed, the business expert configures the prompt template library according to the business scenario. After the business expert's work content is completed. The user side can try to communicate with the system in the actual business scenario. If you are not satisfied with the prompt template of the answer fed back by the system, you can adjust it manually. The adjustment results will be saved in the system database.

[0109] Embodiment 2:

[0110] Based on the same inventive concept, Figure 6 As shown, the present invention also provides a border control system for intelligent dialogue management, comprising:

[0111] The acquisition and analysis module 61 is used to acquire and analyze the description files related to the management of the dialogue scenes; and pre-process the analysis results;

[0112] A training module 62 is used to train the large model using the processed data as training data to obtain an industry large model with better evaluation performance;

[0113] A construction module 63 is used to construct a corresponding prompt scenario template according to the usage scenario corresponding to the trained industry big model;

[0114] The modification feedback module 64 is used to match the corresponding prompt scene template according to the question query of the user during the dialogue interaction process, and receive the modification feedback result of the user on the prompt scene template.

[0115] The overall process of specific use: Business experts first import business files, and then process these files. Use these uploaded business files to train the industry model. After the industry model training and verification is completed, business experts configure the prompt template library according to the business scenario. After the business expert's work content is completed. On the user side, you can try to communicate with the system in the actual business scenario. If you are not satisfied with the prompt template of the answer fed back by the system, you can adjust it manually. The adjustment results will be saved in the system database.

[0116] The following is a detailed introduction of the system from the perspective of data flow and processing:

[0117] The border control system based on intelligent dialogue management includes: file parsing module, data transmission module, file labeling module, model development module, vectorization module, storage module and operation module.

[0118] File parsing module: includes processing unit and parsing unit. The processing unit includes common data processing functions. After the business expert uploads the business file, the business expert can manually process the formatted and semi-formatted files in the processing unit and select the data for training the business model, thereby improving the efficiency of the training model. The parsing unit includes a variety of tools for parsing files such as word, pdf, and txt. After receiving the business file processed by the processing unit. According to the type of business file, the corresponding parsing tool is called to read the structure and format of the file, parse the metadata, and extract the text content. The data processed by the file parsing module is sent to the file labeling module for the next step of processing.

[0119] Data transfer module: includes receiving unit, transmission unit and sending unit. The receiving unit is used to receive files sent by other modules, the transmission unit is used to transmit files received by the receiving unit, and the sending unit is used to send files transmitted by the transmission unit. The receiving unit receives files sent by the file labeling module that have been marked by business experts, and sends them to the model development module and the vectorization module through the sending unit. Receive user questions sent by the operation module, and send them to the vectorization module through the sending unit. Receive prompt question templates sent by the storage module, and send them to the operation module for users to browse and edit. Receive user feedback, and save the user feedback results to the storage unit through the transmission and sending units.

[0120] File labeling module: includes feature engineering automation unit and manual labeling unit. In the feature engineering unit, business experts can manually configure rules to realize automatic processing of data features processed by the file parsing module. Feature processing includes: feature transformation, feature importance assessment, feature selection, feature generation, etc., including but not limited to normalization, standardization, discretization, one-hot encoding, etc. If business experts are not satisfied with the results of automated processing, they can also process data by manual labeling through the manual labeling unit.

[0121] Model development module: includes configuration unit, conversion unit, training unit, and evaluation unit. In the configuration unit, business experts can configure the industry to which the industry model belongs, training rounds, training performance and other related parameters. In the configuration unit, prompt scenario templates can also be configured, including [context / background information], [roles], [verbs], [tasks and objects], and [output and length]. The output can include relevant [data and content sources], [writing style], and [target population / audience] that the content is mainly aimed at. The conversion unit is used to convert the content sent by the file label module into vectors. The training unit builds model training tasks and starts, monitors, and stops the model. The evaluation unit is used to evaluate the training effect of the training unit. If qualified, the model is stored. If unqualified, the model is retrained by modifying the prompt scenario template or calling the conversion unit to re-vectorize the business data.

[0122] Vectorization module: includes splitting unit and conversion unit. The splitting unit is used to split the business data sent by the expert tag library, the prompt template data sent by the model development module, and the questions sent by the operation module into smaller pieces. The conversion unit converts the data processed by the splitting unit into vectors.

[0123] Storage module: Contains business data unit, prompt template unit, industry model unit, and weight unit. Business data unit is used to store the processed business data and its index sent by the vectorization unit. Prompt unit is used to store prompt templates and their indexes. Industry model unit is used to store the industry model developed in the model development module.

[0124] Operation module: includes interaction unit, editing unit, and feedback unit. When users use the industry model in the operation module, they can ask questions to the industry model through the interaction unit, and the results of the questions will be presented to the user in the form of prompt answer templates. If the answer is not satisfactory, the user can edit the prompt question template through the editing unit. The edited results will be updated in the storage module through the feedback unit.

[0125] Take the product after-sales intelligent dialogue management system as an example: First, the business expert imports word files such as product manuals and common product usage problems encountered by users into the system. After receiving these files, the system will call the parsing module to call the python-docx library to read the product planning plan word document, understand the file structure and format of the word document, parse the metadata of the word document, extract the text content, style, format and layout information, and process complex document objects such as lists, hyperlinks, comments, annotations, etc. The processed files will automatically perform feature engineering and display the results to the business experts. If the business experts are not satisfied with the results, they can manually annotate them. After the annotation is completed, the business experts configure the use scenario, training rounds, training performance and other related parameters for the product after-sales dialogue model on the model development page. Configure the prompt scenario template, which includes [context / background information], [role], [verb], [task and object], [output and length]. The output can include relevant [data and content source], writing [style], and [target population / audience] of the content. After the configuration is completed, the business experts start to use the previously analyzed data to start model training and make adjustments until they are satisfied with the model results. Once satisfied, the product after-sales intelligent dialogue management system client is opened to users.

[0126] When a user accesses the client, he asks a question, such as: I clicked the power button, but the product did not start. The client converts the user's question into a vector, extracts the relevant answer from the database, and displays it to the user on the client in order of weight. If the user is not satisfied with the answer, he can edit the prompt question template. After the editing is completed, the system will retrieve the prompt answer template from the database based on the user's edit, and save the prompt question template edited by the user.

[0127] Embodiment 3:

[0128] Based on the same inventive concept, the present invention also provides a computer device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0129] Memory, used to store computer programs;

[0130] The processor, when used to execute the program stored in the memory, can implement a boundary control method for intelligent dialogue management as described in any one of Example 1.

[0131] like Figure 7 As shown, the electronic device may include: a processor 71, a communications interface 72, a memory 73 and a communication bus 74, wherein the processor 71, the communications interface 72 and the memory 73 communicate with each other via the communication bus 74. The processor 71 may call the logic instructions in the memory 73 to execute a boundary control method for intelligent dialogue management, the method comprising:

[0132] S1: Obtain and parse the description files related to making dialogue scene management; pre-process the parsing results;

[0133] S2: Use the processed data as training data to train the large model, and obtain an industry large model with good evaluation performance;

[0134] S3: Build a corresponding prompt scenario template according to the usage scenario corresponding to the trained industry big model;

[0135] S4: Match the corresponding prompt scenario template according to the user's question query during the dialogue interaction process, and receive the user's modification feedback results on the prompt scenario template.

[0136] In addition, the logic instructions in the above-mentioned memory 73 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0137] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0138] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0139] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0141] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A boundary control method for intelligent dialogue management, characterized in that: The following steps are involved: S1: Obtain and parse the description files related to making dialogue scene management; pre-process the parsing results; use the processed data to develop industry-specific models and store them in the database; Add private knowledge base to large models; S2: Use the processed data as training data to train the large model, and obtain an industry large model with good evaluation performance; S3: Based on the usage scenarios corresponding to the trained industry big model, a prompt scenario template of a series of reasoning steps is constructed through the thought chain prompt engineering; S4: Match the corresponding prompt scenario template according to the user's question query during the dialogue interaction process, and receive the user's modification feedback result on the prompt scenario template; The step S4 comprises: S41, receiving questions from users regarding content of interest, and converting the questions from users into vector data; S42, matching the vector data with the data in the database, and extracting the corresponding prompt question template; S43, according to the prompt question template, the trained industry model is queried, and the prompt answer template is obtained according to the selection of the weight library; S44. When the user is not satisfied with the prompt answer template, receive the user's feedback on the modification of the prompt question template and update it to the weight library.

2. The border control method of intelligent dialogue management according to claim 1, characterized in that: The step S1 comprises: S11, obtaining description files related to making dialogue scene management, using python-docx library, pdfminer library, and chardet library to parse the file structure and format of the description files respectively; S12, parsing metadata of the description file; S13, extracting text content, style, format and layout information of the description document; S14. Handle complex document layout and formatting, including page size, headers, footers, and column layout; S15, segmenting and sentence-by-sentence processing of the parsed text content, and converting the result into vector data; S16. Obtain corresponding vector data, and obtain marking results of unsatisfactory parts of the vector data to obtain processed vector data.

3. The border control method of intelligent dialogue management according to claim 2, characterized in that: The step S2 comprises: S21. Configure the relevant parameters of the industry, training rounds, and training performance of the large model; S22, importing the processed vector data into the large model, constructing the training task of the model, and starting, monitoring and stopping the model; S23. Evaluate the training effect of the large model. When it meets expectations, stop training to obtain an industry large model with better evaluation performance.

4. The border control method of intelligent dialogue management according to claim 3, characterized in that: In step S3, the prompt scene template includes: scene setting rules and output rules; The scenario setting rules include: context / background information, roles, verbs, task objects, and output length; Output rules include: relevant data content sources, writing style and target population.

5. A border control system with intelligent dialogue management, characterized in that: include: An acquisition and parsing module is used to acquire and parse the description files related to making dialogue scene management; Preprocess the parsing results; The processed data is used to develop industry-specific models and stored in the database; Add private knowledge base to large models; The training module is used to train the large model using the processed data as training data to obtain an industry large model with better evaluation performance; A construction module is used to construct a prompt scenario template of a series of reasoning steps through a thought chain prompt engineering according to the usage scenario corresponding to the trained industry big model; The modification feedback module is used to match the corresponding prompt scenario template according to the user's question query during the dialogue interaction process, and receive the user's modification feedback result on the prompt scenario template; specifically, it is used to perform the following steps: S41, receiving questions from users regarding content of interest, and converting the questions from users into vector data; S42, matching the vector data with the data in the database, and extracting the corresponding prompt question template; S43, according to the prompt question template, the trained industry model is queried, and the prompt answer template is obtained according to the selection of the weight library; S44. When the user is not satisfied with the prompt answer template, receive the user's feedback on the modification of the prompt question template and update it to the weight library.

6. A computer device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory, used to store computer programs; The processor is used to implement a boundary control method for intelligent dialogue management as described in any one of claims 1 to 4 when executing a program stored in the memory.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, which, when executed by a processor, enable the processor to execute a boundary control method for intelligent dialogue management according to any one of claims 1 to 4.

Images