A Rule-Data Fusion Intelligent Question Answering Method and System Based on a Large Model

By integrating data indicators and knowledge bases from the power sector into a large language model, the accuracy problem of question answering in the power vertical field was solved, thereby improving the accuracy of intelligent question answering and user satisfaction.

CN118734962BActive Publication Date: 2026-11-17HUNAN ELECTRIC POWER DISPATCH HIGH TECH DEV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410631031.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-21
Publication Date
2026-11-17
Estimated Expiration
2044-05-21

AI Technical Summary

Technical Problem

Large language models perform poorly in knowledge-based question answering of power indicators and vertical power sectors, lacking accuracy.

Method used

By acquiring user-input question information, preprocessing and word segmentation are performed, keywords are extracted, and matched with preset data indicator modules and knowledge bases. Combined with data indicator databases and knowledge bases in the power industry, data and rules integration is achieved, improving the accuracy of answers.

Benefits of technology

It has achieved greater accuracy in intelligent question answering in the power sector, eliminating the illusion of large models and improving the accuracy of answers and user satisfaction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118734962B_ABST
    Figure CN118734962B_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for intelligent question answering based on a large model and fusion of rules and data. The method includes: acquiring question information and user data input from the user terminal; preprocessing the question-and-answer information input from the user terminal to obtain standard question information; sending the standard question information to a preset word segmentation module to obtain question information output in a specified format; extracting keywords from the question information output in the specified format; if the keywords in the question information output in the specified format are the same as preset keywords, extracting the corresponding keywords; inputting the same keywords and corresponding user data into a preset data indicator module to obtain indicators for the corresponding keywords; and sending the indicators for the same keywords to the user terminal for display. This invention eliminates the illusion of a large model, achieves rule-data fusion, and improves the accuracy of intelligent answers.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent question answering technology, and more specifically, to a method and system for intelligent question answering based on large-scale model-based data fusion. Background Technology

[0002] While large language models have strong advantages in natural language understanding and text generation, they perform poorly in specialized vertical knowledge domains and power index question answering. Although knowledge base retrieval methods based on large models and vector retrieval have emerged in the market, the strong generalization ability of large models has not solved the problem well in power index and power vertical domain knowledge question answering.

[0003] Therefore, existing technologies have shortcomings and urgently need improvement. Summary of the Invention

[0004] In view of the above problems, the purpose of this invention is to provide a method and system for intelligent question answering based on large model fusion, which can improve the accuracy of intelligent answers.

[0005] The first aspect of this invention provides an improved intelligent question-answering method based on large-scale model fusion, comprising:

[0006] Obtain user-input questions and user profile information;

[0007] The user-input question and answer information is preprocessed to obtain standard question information;

[0008] The standard question information is sent to the preset word segmentation module to obtain the question information output in the specified format;

[0009] Extract keywords from the query information output in a specified format;

[0010] If the keywords in the output query information in the specified format are the same as the preset keywords, extract the corresponding keywords;

[0011] Input the same keywords and corresponding user data into the preset data indicator module to obtain the indicators for the same keywords;

[0012] Send metrics with the same keywords to the user's device for display;

[0013] The preset word segmentation module stores a large model with word segmentation capabilities.

[0014] This plan also includes:

[0015] If the keywords in the question information output in the specified format are different from the preset keywords, then the keywords in the question information output in the specified format will be matched with the keywords in the preset knowledge base to obtain the first similarity matching value set;

[0016] If there is a first similar matching value in the first similar matching value set that is greater than or equal to the preset first similar matching threshold, then the first similar matching values ​​are sorted to obtain the largest first similar matching value.

[0017] The knowledge corresponding to the largest first similarity match value is sent to the user's terminal for display;

[0018] If the first similarity match value in the first similarity match value set is less than the preset first similarity match threshold, the preset output statement is obtained;

[0019] The preset output statement is sent to the user terminal for display.

[0020] In this solution, the step of preprocessing the question-and-answer information input by the user to obtain standard question information specifically includes:

[0021] Based on the input format selected by the user, determine the input format of the corresponding question and answer information;

[0022] Determine whether the input format of the question and answer information is the same as the preset format. If so, set the corresponding question and answer information as the standard question information.

[0023] If not, input the corresponding question and answer information into the preset format conversion module to obtain standard question information with the same preset format.

[0024] In this solution, the step of inputting the same keywords and corresponding user data into the preset data indicator module to obtain the indicators for the same keywords specifically includes:

[0025] Based on the corresponding user terminal data, determine the corresponding user terminal's ID value;

[0026] Based on the corresponding user terminal ID value, match the data indicator sub-database corresponding to the ID value;

[0027] Match the same keywords with the keywords in the corresponding data indicator sub-library to obtain the second similarity matching value set;

[0028] If there is a second similarity matching value in the second similarity matching value set that is greater than or equal to the preset second similarity matching threshold, then the second similarity matching values ​​are sorted to obtain the largest second similarity matching value;

[0029] Set the index corresponding to the keyword in the data index corresponding to the second most similar matching value in the word library as the index for the same keyword;

[0030] The preset data indicator module includes a data indicator library, which is composed of multiple data indicator sub-libraries.

[0031] This plan also includes:

[0032] If the second similarity matching value in the second similarity matching value set is less than the preset second similarity matching threshold, the keywords in the question information output in the specified format will be matched with the keywords in the preset knowledge base to obtain the knowledge or preset output statement corresponding to the largest first similarity matching value, and then verified with the user.

[0033] If the user's verification of the authenticity of the question and answer is negative, then extract the actual feedback information provided by the user.

[0034] Based on a preset first time period, the number of times the user provides real feedback information in the same question is obtained;

[0035] If the number of times a user provides genuine feedback in the same question exceeds a preset threshold for the first time, the corresponding genuine feedback will be sent to a preset management terminal for processing.

[0036] This plan also includes:

[0037] When the knowledge corresponding to the largest first similarity match value is not verified, record that the keywords in the question information output in the specified format corresponding to the largest first similarity match value and the keywords in the preset knowledge base are mismatched once.

[0038] Get the total number of incorrect matches;

[0039] If the total number of matches is greater than the preset second threshold, the total number of matching errors is multiplied by the preset first similarity matching threshold to obtain the first similarity matching revision value of the keywords in the question information output in the specified format corresponding to the largest first similarity matching value and the keywords in the preset knowledge base.

[0040] Subtract the first similarity match revision value from the largest first similarity match value to obtain the first similarity match value after revision between the keywords in the question information output in the specified format corresponding to the largest first similarity match value and the keywords in the preset knowledge base.

[0041] This plan also includes:

[0042] Obtain historical power index information;

[0043] Historical power index information is preprocessed to obtain a historical power index training set;

[0044] Based on the preset ql ora, freeze the preset trained model weight parameters and add a new network layer to the preset trained large model;

[0045] The new network layer is trained based on the historical electricity index training set to obtain the parameters of the new network layer. The pre-trained model weight parameters are then unfrozen to obtain the adjusted large model.

[0046] Obtain historical Q&A information and corresponding accurate indicator answers to those historical Q&A information;

[0047] Input historical question-and-answer information into the adjusted large model, output indicator answers, and determine the accuracy rate based on the accurate indicator answer information corresponding to the historical answer information.

[0048] When the accuracy is greater than or equal to the preset accuracy threshold, training of new network layers is stopped;

[0049] When the accuracy is less than the preset accuracy threshold, the preset trained model weight parameters are frozen by the preset ql ora, and the new network layer is continued to be trained by the historical power index training set.

[0050] A second aspect of the present invention provides a large-model-based intelligent question-answering system based on metric-data fusion, comprising a memory and a processor. The memory stores a program for a large-model-based intelligent question-answering method, which, when executed by the processor, performs the following steps:

[0051] Obtain user-input questions and user profile information;

[0052] The user-input question and answer information is preprocessed to obtain standard question information;

[0053] The standard question information is sent to the preset word segmentation module to obtain the question information output in the specified format;

[0054] Extract keywords from the query information output in a specified format;

[0055] If the keywords in the output query information in the specified format are the same as the preset keywords, extract the corresponding keywords;

[0056] Input the same keywords and corresponding user data into the preset data indicator module to obtain the indicators for the same keywords;

[0057] Send metrics with the same keywords to the user's device for display;

[0058] The preset word segmentation module stores a large model with word segmentation capabilities.

[0059] This plan also includes:

[0060] If the keywords in the question information output in the specified format are different from the preset keywords, then the keywords in the question information output in the specified format will be matched with the keywords in the preset knowledge base to obtain the first similarity matching value set;

[0061] If there is a first similar matching value in the first similar matching value set that is greater than or equal to the preset first similar matching threshold, then the first similar matching values ​​are sorted to obtain the largest first similar matching value.

[0062] The knowledge corresponding to the largest first similarity match value is sent to the user's terminal for display;

[0063] If the first similarity match value in the first similarity match value set is less than the preset first similarity match threshold, the preset output statement is obtained;

[0064] The preset output statement is sent to the user terminal for display.

[0065] In this solution, the step of preprocessing the question-and-answer information input by the user to obtain standard question information specifically includes:

[0066] Based on the input format selected by the user, determine the input format of the corresponding question and answer information;

[0067] Determine whether the input format of the question and answer information is the same as the preset format. If so, set the corresponding question and answer information as the standard question information.

[0068] If not, input the corresponding question and answer information into the preset format conversion module to obtain standard question information with the same preset format.

[0069] This invention discloses a method and system for intelligent question answering based on large-scale model fusion, which eliminates the illusion of large-scale models, realizes fusion of rules and data, and improves the accuracy of intelligent answers. Attached Figure Description

[0070] Figure 1 A flowchart of a rule-based intelligent question-answering method based on a large model is shown below.

[0071] Figure 2 The diagram shows a block diagram of a rule-based intelligent response system based on a large model according to the present invention. Detailed Implementation

[0072] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0073] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0074] Figure 1 The flowchart of a metric fusion intelligent question answering method based on a large model according to the present invention is shown.

[0075] like Figure 1 As shown, this invention discloses a rule-based data fusion intelligent question-answering method based on a large model, comprising:

[0076] S101, Obtain the question information and user data input by the user;

[0077] S102, preprocess the question and answer information input by the user to obtain standard question information;

[0078] S103, send the standard question information to the preset word segmentation module to obtain the question information output in the specified format;

[0079] S104, Extract keywords from the query information output in the specified format;

[0080] S105, If the keywords in the question information output in the specified format are the same as the preset keywords, extract the corresponding keywords;

[0081] S106, Input the same keywords and corresponding user terminal data into the preset data indicator module to obtain the indicators for the same keywords;

[0082] S107, Send the metrics for the same keywords to the user terminal for display;

[0083] According to an embodiment of the present invention, the preset word segmentation module stores a large model with word segmentation capability; the standard specifications of the standard question information are set according to the field to which the question and answer belong. For example, if it is the field of electricity, the corresponding standard question information contains keywords such as time, address, and electricity sales. The keywords in the question information output in the specified format are matched with the preset keywords to find the indicator corresponding to the threshold. For example, if it is the field of electricity, the corresponding indicator is the electricity indicator. The preset data indicator module stores a data indicator library for storing data indicators of the user end.

[0084] According to an embodiment of the present invention, it further includes:

[0085] If the keywords in the question information output in the specified format are different from the preset keywords, then the keywords in the question information output in the specified format will be matched with the keywords in the preset knowledge base to obtain the first similarity matching value set;

[0086] If there is a first similar matching value in the first similar matching value set that is greater than or equal to the preset first similar matching threshold, then the first similar matching values ​​are sorted to obtain the largest first similar matching value.

[0087] The knowledge corresponding to the largest first similarity match value is sent to the user's terminal for display;

[0088] If the first similarity match value in the first similarity match value set is less than the preset first similarity match threshold, the preset output statement is obtained;

[0089] The preset output statement is sent to the user terminal for display.

[0090] It should be noted that when all first similarity matching values ​​are less than the preset first similarity matching threshold, it means that the current user's input question does not involve the indicator or the input question is vague and unclear. A preset output statement is used to prompt the user, such as "Hello, the indicator Q&A assistant is still learning, please wait patiently".

[0091] It should be noted that by using data indicators from a preset data indicator library and knowledge from a preset knowledge base to answer questions in a specified format, the data and rules are integrated, thus improving the accuracy of question and answer responses.

[0092] According to an embodiment of the present invention, the step of preprocessing the question-and-answer information input by the user terminal to obtain standard question information specifically includes:

[0093] Based on the input format selected by the user, determine the input format of the corresponding question and answer information;

[0094] Determine whether the input format of the question and answer information is the same as the preset format. If so, set the corresponding question and answer information as the standard question information.

[0095] If not, input the corresponding question and answer information into the preset format conversion module to obtain standard question information with the same preset format.

[0096] It should be noted that users can choose the input format for their questions according to their own needs, such as voice or text. If the preset format is text, but the user's input question is in voice format, then the voice question needs to be converted to text format by the voice recognition system.

[0097] According to an embodiment of the present invention, the step of inputting the same keywords and corresponding user terminal information into a preset data indicator module to obtain indicators corresponding to the same keywords specifically includes:

[0098] Based on the corresponding user terminal data, determine the corresponding user terminal's ID value;

[0099] Based on the corresponding user terminal ID value, match the data indicator sub-database corresponding to the ID value;

[0100] Match the same keywords with the keywords in the corresponding data indicator sub-library to obtain the second similarity matching value set;

[0101] If there is a second similarity matching value in the second similarity matching value set that is greater than or equal to the preset second similarity matching threshold, then the second similarity matching values ​​are sorted to obtain the largest second similarity matching value;

[0102] Set the index corresponding to the keyword in the data index corresponding to the second most similar matching value in the word library as the index for the same keyword;

[0103] The preset data indicator module includes a data indicator library, which is composed of multiple data indicator sub-libraries.

[0104] It should be noted that each user terminal ID corresponds to a data metric sub-library, and each data metric sub-library contains one or more user terminal data metrics.

[0105] According to an embodiment of the present invention, it further includes:

[0106] If the second similarity matching value in the second similarity matching value set is less than the preset second similarity matching threshold, the keywords in the question information output in the specified format will be matched with the keywords in the preset knowledge base to obtain the knowledge or preset output statement corresponding to the largest first similarity matching value, and then verified with the user.

[0107] If the user's verification of the authenticity of the question and answer is negative, then extract the actual feedback information provided by the user.

[0108] Based on a preset first time period, the number of times the user provides real feedback information in the same question is obtained;

[0109] If the number of times a user provides genuine feedback in the same question exceeds a preset threshold for the first time, the corresponding genuine feedback will be sent to a preset management terminal for processing.

[0110] It should be noted that when the second similarity match value in the second similarity match value set is less than the preset second similarity match threshold, it means that the current data indicator library does not contain data indicator answers related to the user's question information. Therefore, the answer is provided using knowledge from the preset knowledge base, and the knowledge or preset output statement corresponding to the largest first similarity match value is verified based on the user's feedback. If the verification fails, it means that the user is not satisfied with the knowledge or preset output statement corresponding to the largest first similarity match value. By setting a preset first-time threshold, when the number of times the user provides real feedback information in the same question exceeds the preset first-time threshold, the corresponding real feedback information is sent to the preset management terminal. The preset management terminal processes the knowledge with user feedback, such as deleting or adding.

[0111] According to an embodiment of the present invention, it further includes:

[0112] When the knowledge corresponding to the largest first similarity match value is not verified, record that the keywords in the question information output in the specified format corresponding to the largest first similarity match value and the keywords in the preset knowledge base are mismatched once.

[0113] Get the total number of incorrect matches;

[0114] If the total number of matches is greater than the preset second threshold, the total number of matching errors is multiplied by the preset first similarity matching threshold to obtain the first similarity matching revision value of the keywords in the question information output in the specified format corresponding to the largest first similarity matching value and the keywords in the preset knowledge base.

[0115] Subtract the first similarity match revision value from the largest first similarity match value to obtain the first similarity match value after revision between the keywords in the question information output in the specified format corresponding to the largest first similarity match value and the keywords in the preset knowledge base.

[0116] It should be noted that, for example, if the first similarity match value between keyword 'a' and keyword A in the preset knowledge base is the largest, and the user's feedback at that time is negative, then a match error between the current keyword 'a' and keyword A in the preset knowledge base is recorded once. When there are multiple match errors between the previous keyword 'a' and keyword A in the preset knowledge base, and the total number of corresponding match errors is greater than the preset second threshold, the first similarity match value between keyword 'a' and keyword A in the preset knowledge base is revised based on the first similarity match revision value, in order to improve the accuracy of the first similarity match value between the keywords in the query information output in the specified format and the keywords in the preset knowledge base.

[0117] According to an embodiment of the present invention, it further includes:

[0118] Obtain historical power index information;

[0119] Historical power index information is preprocessed to obtain a historical power index training set;

[0120] Based on the preset ql ora, freeze the preset trained model weight parameters and add a new network layer to the preset trained large model;

[0121] The new network layer is trained based on the historical electricity index training set to obtain the parameters of the new network layer. The pre-trained model weight parameters are then unfrozen to obtain the adjusted large model.

[0122] Obtain historical Q&A information and corresponding accurate indicator answers to those historical Q&A information;

[0123] Input historical question-and-answer information into the adjusted large model, output indicator answers, and determine the accuracy rate based on the accurate indicator answer information corresponding to the historical answer information.

[0124] When the accuracy is greater than or equal to the preset accuracy threshold, training of new network layers is stopped;

[0125] When the accuracy is less than the preset accuracy threshold, the preset trained model weight parameters are frozen by the preset ql ora, and the new network layer is continued to be trained by the historical power index training set.

[0126] It should be noted that the large model is a dataset of the indicator at a certain time point. For example, if it is an electricity indicator, then it is an electricity dataset at a certain time point. By fine-tuning the large model, each sentence is identified and output in a specified pattern.

[0127] Figure 2 The diagram shows a block diagram of a rule-based intelligent response system based on a large model according to the present invention.

[0128] like Figure 2 As shown, a second aspect of the present invention provides a large-model-based intelligent question-answering system 2, comprising a memory 21 and a processor 22. The memory stores a program for a large-model-based intelligent question-answering method. When the processor executes the program for the large-model-based intelligent question-answering method, it performs the following steps:

[0129] Obtain user-input questions and user profile information;

[0130] The user-input question and answer information is preprocessed to obtain standard question information;

[0131] The standard question information is sent to the preset word segmentation module to obtain the question information output in the specified format;

[0132] Extract keywords from the query information output in a specified format;

[0133] If the keywords in the output query information in the specified format are the same as the preset keywords, extract the corresponding keywords;

[0134] Input the same keywords and corresponding user data into the preset data indicator module to obtain the indicators for the same keywords;

[0135] Send metrics with the same keywords to the user's device for display;

[0136] The preset word segmentation module stores a large model with word segmentation capabilities.

[0137] This plan also includes:

[0138] If the keywords in the question information output in the specified format are different from the preset keywords, then the keywords in the question information output in the specified format will be matched with the keywords in the preset knowledge base to obtain the first similarity matching value set;

[0139] If there is a first similar matching value in the first similar matching value set that is greater than or equal to the preset first similar matching threshold, then the first similar matching values ​​are sorted to obtain the largest first similar matching value.

[0140] The knowledge corresponding to the largest first similarity match value is sent to the user's terminal for display;

[0141] If the first similarity match value in the first similarity match value set is less than the preset first similarity match threshold, the preset output statement is obtained;

[0142] The preset output statement is sent to the user terminal for display.

[0143] In this solution, the step of preprocessing the question-and-answer information input by the user to obtain standard question information specifically includes:

[0144] Based on the input format selected by the user, determine the input format of the corresponding question and answer information;

[0145] Determine whether the input format of the question and answer information is the same as the preset format. If so, set the corresponding question and answer information as the standard question information.

[0146] If not, input the corresponding question and answer information into the preset format conversion module to obtain standard question information with the same preset format.

[0147] This invention discloses a method and system for intelligent question answering based on a large model and fusion of rules and data. The method includes: acquiring question information and user data input from the user terminal; preprocessing the question-and-answer information input from the user terminal to obtain standard question information; sending the standard question information to a preset word segmentation module to obtain question information output in a specified format; extracting keywords from the question information output in the specified format; if the keywords in the question information output in the specified format are the same as preset keywords, extracting the corresponding keywords; inputting the same keywords and corresponding user data into a preset data indicator module to obtain indicators for the corresponding keywords; and sending the indicators for the same keywords to the user terminal for display. This invention eliminates the illusion of a large model, achieves rule-data fusion, and improves the accuracy of intelligent answers.

[0148] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0149] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0150] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0151] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0152] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

Claims

1. A rule-based intelligent question-answering method based on a large model, characterized in that, include: Obtain user-input questions and user profile information; The user-input question and answer information is preprocessed to obtain standard question information; The standard question information is sent to the preset word segmentation module to obtain the question information output in the specified format; Extract keywords from the query information output in a specified format; If the keywords in the output query information in the specified format are the same as the preset keywords, extract the corresponding keywords; Input the same keywords and corresponding user data into the preset data indicator module to obtain the indicators for the same keywords; specifically including: Based on the corresponding user terminal data, determine the corresponding user terminal's ID value; Based on the corresponding user terminal ID value, match the data indicator sub-database corresponding to the ID value; Match the same keywords with the keywords in the corresponding data indicator sub-library to obtain the second similarity matching value set; If there is a second similarity matching value in the second similarity matching value set that is greater than or equal to the preset second similarity matching threshold, then the second similarity matching values ​​are sorted to obtain the largest second similarity matching value; Set the index corresponding to the keyword in the data index corresponding to the second most similar matching value in the word library as the index for the same keyword; The preset data indicator module includes a data indicator library, which is composed of multiple data indicator sub-libraries. Send metrics with the same keywords to the user's device for display; The preset word segmentation module stores a large model with word segmentation capabilities; Also includes: If the keywords in the question information output in the specified format are different from the preset keywords, then the keywords in the question information output in the specified format will be matched with the keywords in the preset knowledge base to obtain the first similarity matching value set; If there is a first similar matching value in the first similar matching value set that is greater than or equal to the preset first similar matching threshold, then the first similar matching values ​​are sorted to obtain the largest first similar matching value. The knowledge corresponding to the largest first similarity match value is sent to the user's terminal for display; If the first similarity match value in the first similarity match value set is less than the preset first similarity match threshold, the preset output statement is obtained; The preset output statement is sent to the user terminal for display.

2. The intelligent question-answering method based on large-scale model fusion according to claim 1, characterized in that, The step of preprocessing the question-and-answer information input by the user to obtain standard question information specifically includes: Based on the input format selected by the user, determine the input format of the corresponding question and answer information; Determine whether the input format of the question and answer information is the same as the preset format. If so, set the corresponding question and answer information as the standard question information. If not, input the corresponding question and answer information into the preset format conversion module to obtain standard question information with the same preset format.

3. The intelligent question-answering method based on large-scale model fusion according to claim 1, characterized in that, Also includes: If the second similarity matching value in the second similarity matching value set is less than the preset second similarity matching threshold, the keywords in the question information output in the specified format will be matched with the keywords in the preset knowledge base to obtain the knowledge or preset output statement corresponding to the largest first similarity matching value, and then verified with the user. If the user's verification of the authenticity of the question and answer is negative, then extract the actual feedback information provided by the user. Based on a preset first time period, the number of times the user provides real feedback information in the same question is obtained; If the number of times a user provides genuine feedback in the same question exceeds a preset threshold for the first time, the corresponding genuine feedback will be sent to a preset management terminal for processing.

4. The intelligent question-answering method based on large-scale model fusion according to claim 3 further includes: When the knowledge corresponding to the largest first similarity match value is not verified, record that the keywords in the question information output in the specified format corresponding to the largest first similarity match value and the keywords in the preset knowledge base are mismatched once. Get the total number of incorrect matches; If the total number of matches is greater than the preset second threshold, the total number of matching errors is multiplied by the preset first similarity matching threshold to obtain the first similarity matching revision value of the keywords in the question information output in the specified format corresponding to the largest first similarity matching value and the keywords in the preset knowledge base. Subtract the first similarity match revision value from the largest first similarity match value to obtain the first similarity match value after revision between the keywords in the question information output in the specified format corresponding to the largest first similarity match value and the keywords in the preset knowledge base.

5. The intelligent question-answering method based on large-scale model fusion according to claim 1 further includes: Obtain historical power index information; Historical power index information is preprocessed to obtain a historical power index training set; Based on the preset qlora, the preset trained model weight parameters are frozen, and a new network layer is added to the preset trained large model; The new network layer is trained based on the historical electricity index training set to obtain the parameters of the new network layer. The pre-trained model weight parameters are then unfrozen to obtain the adjusted large model. Obtain historical Q&A information and corresponding accurate indicator answers to those historical Q&A information; Input historical question-and-answer information into the adjusted large model, output indicator answers, and determine the accuracy rate based on the accurate indicator answer information corresponding to the historical answer information. When the accuracy is greater than or equal to the preset accuracy threshold, training of new network layers is stopped; When the accuracy is less than the preset accuracy threshold, the preset trained model weight parameters are frozen by the preset qlora, and the new network layer is continued to be trained by the historical power index training set.

6. A rule-data fusion intelligent question-answering system based on a large model, characterized in that, The system includes a memory and a processor. The memory stores a program for a large-model-based intelligent question-answering method that integrates rules and data fusion. When the processor executes the program, the program performs the following steps: Obtain user-input questions and user profile information; The user-input question and answer information is preprocessed to obtain standard question information; The standard question information is sent to the preset word segmentation module to obtain the question information output in the specified format; Extract keywords from the query information output in a specified format; If the keywords in the output query information in the specified format are the same as the preset keywords, extract the corresponding keywords; Input the same keywords and corresponding user data into the preset data indicator module to obtain the indicators for the same keywords; specifically including: Based on the corresponding user terminal data, determine the corresponding user terminal's ID value; Based on the corresponding user terminal ID value, match the data indicator sub-database corresponding to the ID value; Match the same keywords with the keywords in the corresponding data indicator sub-library to obtain the second similarity matching value set; If there is a second similarity matching value in the second similarity matching value set that is greater than or equal to the preset second similarity matching threshold, then the second similarity matching values ​​are sorted to obtain the largest second similarity matching value; Set the index corresponding to the keyword in the data index corresponding to the second most similar matching value in the word library as the index for the same keyword; The preset data indicator module includes a data indicator library, which is composed of multiple data indicator sub-libraries. Send metrics with the same keywords to the user's device for display; The preset word segmentation module stores a large model with word segmentation capabilities; Also includes: If the keywords in the question information output in the specified format are different from the preset keywords, then the keywords in the question information output in the specified format will be matched with the keywords in the preset knowledge base to obtain the first similarity matching value set; If there is a first similar matching value in the first similar matching value set that is greater than or equal to the preset first similar matching threshold, then the first similar matching values ​​are sorted to obtain the largest first similar matching value. The knowledge corresponding to the largest first similarity match value is sent to the user's terminal for display; If the first similarity match value in the first similarity match value set is less than the preset first similarity match threshold, the preset output statement is obtained; The preset output statement is sent to the user terminal for display.

7. The intelligent question-answering system based on large-scale model fusion according to claim 6, characterized in that, The step of preprocessing the question-and-answer information input by the user to obtain standard question information specifically includes: Based on the input format selected by the user, determine the input format of the corresponding question and answer information; Determine whether the input format of the question and answer information is the same as the preset format. If so, set the corresponding question and answer information as the standard question information. If not, input the corresponding question and answer information into the preset format conversion module to obtain standard question information with the same preset format.

Citation Information

Patent Citations

  • Intelligent question and answer method, device and equipment and storage medium

    CN117591658A

  • Question and answer method and device based on large language model, equipment and medium

    CN117972032A