Response method and device, electronic equipment and storage medium

Abstract

The invention provides a reply method and device, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining a target maintenance document in response to question information for equipment maintenance, and obtaining equipment fault information in the question information; according to the equipment fault information, obtaining at least one piece of historical question information associated with the question information; wherein the content similarity between the question information and the historical question information is greater than a preset similarity threshold; and processing the question information, the historical question information and the target maintenance document based on a preset large language model, and determining reply information corresponding to the question information. The equipment maintenance efficiency can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a reply method, device, electronic device, and computer-readable storage medium. Background Art

[0002] When equipment fails, equipment maintenance currently still mainly relies on human experience or equipment manuals.

[0003] Maintenance personnel need to determine the cause of equipment failure based on the equipment failure phenomenon and the equipment manual.

[0004] The entire maintenance process is difficult to standardize and uniformly manage, and is highly dependent on the experience of maintenance personnel, resulting in low equipment maintenance efficiency and inability to effectively eliminate equipment failures. Summary of the invention

[0005] The embodiments of the present application provide a reply method, apparatus, electronic device, and computer-readable storage medium, aiming to improve equipment maintenance efficiency.

[0006] In a first aspect, an embodiment of the present application provides a reply method, the method comprising:

[0007] In response to a question about equipment maintenance, obtaining a target maintenance document and obtaining equipment failure information in the question;

[0008] Acquire at least one historical question information associated with the question information according to the device failure information; wherein the content similarity between the question information and the historical question information is greater than a preset similarity threshold;

[0009] The question information, the historical question information and the target maintenance document are processed based on a preset large language model to determine answer information corresponding to the question information.

[0010] Optionally, acquiring at least one historical question information associated with the question information according to the device fault information includes:

[0011] According to the device failure information in the question information, query the historical question-answering dialogue, and obtain the content similarity between each historical question information in the historical question-answering dialogue and the question information;

[0012] If the content similarity between the historical question information in the historical question-and-answer dialogue and the question information is greater than a preset similarity threshold, setting the historical question information as the historical question information associated with the question information;

[0013] The method further comprises:

[0014] If there are multiple historical question information associated with the question information, the weight coefficient corresponding to each historical question information is determined according to the similarity between the question information and each historical question information, so as to input the question information, the target maintenance document and the historical question information with each weight coefficient into the large language model.

[0015] Optionally, obtaining the content similarity between each historical question information in the historical question-and-answer dialogue and the question information includes:

[0016] Extracting historical fault information from each of the historical question information according to the structural information, logical information and / or semantic information of each of the historical question information in the historical question-answer pair, wherein the historical fault information includes fault description information and / or fault code;

[0017] The similarity between each of the historical question information and the device fault information in the question information is calculated according to the fault description information and / or the fault code.

[0018] Optionally, the processing the question information, the historical question information and the target maintenance document based on a preset large language model to determine the answer information corresponding to the question information includes:

[0019] Processing the question information and the target maintenance document through the large language model to obtain first answer information corresponding to the question information;

[0020] Processing the historical question information and the target maintenance document through the large language model to obtain second answer information corresponding to the historical question information;

[0021] The first reply information and the second reply information are processed by the large language model to determine the reply information corresponding to the question information.

[0022] Optionally, the processing the first reply information and the second reply information by using the large language model to determine the reply information corresponding to the question information includes:

[0023] Comparing the overlapping characters between the first reply information and the second reply information by using the large language model;

[0024] If there are no overlapping characters between the first reply information and the second reply information, setting the first reply information as the reply information by using the large language model;

[0025] If there are some overlapping characters between the first reply information and the second reply information, optimizing the first reply information according to the second reply information by using the large language model to obtain reply information;

[0026] If the first reply information is the same as the second reply information, the first reply information is set as the reply information through the large language model, and the historical reply information in the corresponding historical question and answer pair is optimized according to the reply information.

[0027] Optionally, in response to the inquiry information regarding equipment maintenance, obtaining a target maintenance document includes:

[0028] In response to the question information about equipment maintenance, query the target knowledge base to obtain multiple knowledge slices in the target knowledge base, and splice the multiple knowledge slices to obtain a target maintenance document;

[0029] The steps of constructing the target knowledge base include:

[0030] Obtain maintenance documents for multiple devices;

[0031] For each equipment maintenance document, performing layout analysis on the equipment maintenance document to identify target elements in the equipment maintenance document;

[0032] According to the target element, the equipment maintenance document is sliced ​​to obtain a plurality of knowledge slices;

[0033] For each of the knowledge slices, vectorization is performed on the knowledge slice, and the vectorized knowledge slice is imported into the target knowledge base.

[0034] Optionally, the step of training the large language model includes:

[0035] Acquire question training information, and extract maintenance training documents corresponding to the question training information from a local knowledge base;

[0036] Inputting the question training information and the maintenance training document into an initial large language model to obtain answer prediction information;

[0037] Calculating the recall accuracy and / or question-answering accuracy of the initial large language model according to the answer prediction information and the answer annotation information corresponding to the question training information;

[0038] If at least one of the recall accuracy and the question-answer accuracy is less than a corresponding accuracy threshold, fine-tuning the initial large language model to adjust a weight parameter of the initial large language model;

[0039] Until the recall accuracy and the question-answer accuracy both reach corresponding accuracy thresholds, a second largest language model is obtained.

[0040] Optionally, the processing of the question information, the historical question information and the target maintenance document based on a preset large language model to determine the answer information corresponding to the question information includes:

[0041] Get multiple sets of question information and corresponding answer information;

[0042] According to the question information, query the historical question information in the historical question and answer pairs;

[0043] If the similarity between the question information and the historical question information is less than a preset similarity, the question information and the corresponding answer information are set as a historical question-answer pair for storage to update the historical question-answer pair.

[0044] In a second aspect, an embodiment of the present application provides a reply device, including:

[0045] A first acquisition module, configured to acquire a target maintenance document in response to a question about equipment maintenance, and acquire equipment failure information in the question;

[0046] A second acquisition module is used to acquire at least one historical question information associated with the question information according to the equipment fault information; wherein the content similarity between the question information and the historical question information is greater than a preset similarity threshold;

[0047] A determination module is used to process the question information, the historical question information and the target maintenance document based on a preset large language model to determine the answer information corresponding to the question information.

[0048] In a third aspect, an embodiment of the present application provides an electronic device, which further includes: a memory and a processor, wherein the memory stores a plurality of instructions; and the processor loads instructions from the memory to execute the steps of the reply method of the present application.

[0049] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, which stores a plurality of instructions, and the instructions are suitable for a processor to load to execute the steps in any one of the reply methods provided in the embodiments of the present application.

[0050] Beneficial effects of the embodiments of the present application:

[0051] Since the present application can utilize both current question information and historical question information, and combine them with maintenance documents as inputs to the large language model, compared to simply inputting the current question information, the large language model can more accurately predict the user's true intentions, and extract the required equipment maintenance reply information for the user from the maintenance documents, thereby effectively improving the accuracy of the reply information. On this basis, the present application, based on the knowledge aggregation and understanding capabilities of the large language model, can solve various types of questions asked by users, provide users with effective maintenance strategies for different equipment failures, assist users in equipment repairs, improve equipment maintenance efficiency, and also reduce equipment maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0053] Figure 1 This is a first schematic diagram of a reply process provided by an embodiment of the present application;

[0054] Figure 2 It is a schematic diagram of field information provided by an embodiment of the present application;

[0055] Figure 3 is a schematic diagram of a system provided by an embodiment of the present application;

[0056] Figure 4 It is a schematic diagram of knowledge base construction provided by an embodiment of the present application;

[0057] Figure 5 This is a first schematic diagram of large language model training provided by an embodiment of the present application;

[0058] Figure 6 This is a second schematic diagram of large language model training provided by an embodiment of the present application;

[0059] Figure 7 is a schematic diagram of a reply device provided in an embodiment of the present application;

[0060] Figure 8 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0062] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, which are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying 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 only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0063] In the embodiment of the present application, "and / or" describes the association relationship of the associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / ", unless otherwise specified, generally indicates that the associated objects before and after are in an "or" relationship.

[0064] In this application, the word "exemplary" is used to mean "serving as an example, illustration, or illustration." Any embodiment described in this application as "exemplary" is not necessarily to be construed as being preferred or advantageous over other embodiments. The following description is given to enable any person skilled in the art to implement and use the invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the invention can be implemented without using these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the invention with unnecessary details. Therefore, the present application is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present application.

[0065] Combined with the above background technology description of this application, when equipment fails, most industries currently rely mainly on human experience or equipment manuals for equipment maintenance. Maintenance personnel need to judge the cause of the equipment failure based on the equipment failure phenomenon and the equipment manual. The entire maintenance process is difficult to standardize and uniformly manage, and is highly dependent on the experience of maintenance personnel, resulting in low equipment maintenance efficiency and failure to effectively eliminate equipment failures.

[0066] In order to solve the above problems, the present application proposes a reply method, device, electronic device, computer-readable storage medium and computer program product, which aim to return corresponding reply information for the question information of equipment failure, so that maintenance personnel can directly refer to the reply information to repair the equipment, which can not only improve the accuracy of equipment fault diagnosis and equipment maintenance efficiency, but also reduce labor costs, lower the threshold for maintenance personnel, and realize unified management of equipment failure maintenance strategies.

[0067] Specifically, the reply method in the embodiment of the present application is as follows: Figure 1 As shown, the following steps may be included:

[0068] S10, in response to the inquiry information regarding equipment maintenance, obtaining a target maintenance document and obtaining equipment failure information in the inquiry information;

[0069] It should be noted that in this embodiment, when a device fails, maintenance personnel can input question information about the device failure to the corresponding terminal device, and the question information input by the user can include device maintenance related information such as fault code and fault description information. For example, maintenance personnel can input "Why is the device indicator light always on red" to the terminal device.

[0070] Among them, the terminal equipment can have a maintenance assistant function, support fault query and analysis functions, and provide maintenance assistance to users.

[0071] The terminal device may be a network device such as a mobile phone, a tablet, a notebook, a server, etc. This embodiment does not specifically limit the type of the terminal device.

[0072] On this basis, the terminal device can respond to the user's question information regarding equipment maintenance and determine the target maintenance document corresponding to the question information. In addition, the terminal device can also obtain the equipment failure information in the question information.

[0073] For example, the terminal device can analyze the question information and extract the equipment failure information in the question information. This embodiment does not limit the type of question information, such as plain text type, picture type (for example, users can take photos of equipment failures and upload them), table type, etc., and no specific limitation is made. In this way, the terminal device can extract corresponding equipment failure information according to corresponding extraction strategies for different types of question information.

[0074] For example, if the question information is in the form of an image, the fault information in the image can be extracted through optical character recognition (OCR) and converted into an editable and searchable text format.

[0075] Furthermore, the terminal device can obtain the target maintenance document according to the question information.

[0076] In a specific embodiment, the terminal device may query the target database according to the question information to obtain the target maintenance document.

[0077] For example, in this embodiment, the database may include a local knowledge base and / or a cloud knowledge base, wherein the target database stores equipment-related documents, such as product manuals, maintenance records, equipment-related videos, drawings, spare parts library files, and major accident reports, etc. The cloud knowledge base can be understood as Internet information, such as websites, forums, and other information about equipment. In this embodiment, the cloud knowledge base can be used as a supplement and support for the local knowledge base to provide users with more comprehensive and accurate response information and assist users in equipment maintenance.

[0078] S20, acquiring historical question information associated with the question information according to the device failure information; wherein the content similarity between the question information and the historical question information is greater than a preset similarity threshold;

[0079] In this embodiment, after the terminal device extracts the equipment fault information from the question information, it can obtain the historical question information associated with the question information based on the equipment fault information; wherein, in this embodiment, the content similarity between the question information and the historical question information is greater than a preset similarity threshold, and the preset similarity threshold can indicate that the question information and the historical question information are question information input for the same object, for example, the question information and the historical question information are both questions asked about the continuous flashing red light of the device.

[0080] It should be noted that in this embodiment, in addition to obtaining the question information currently input by the user, historical question and answer pairs can also be combined to extract historical question information, so that the current question information and historical question information can be simultaneously input into the large language model for answer prediction, so as to accurately predict the user's true intentions and provide the user with the required equipment maintenance strategy.

[0081] S30, processing the question information, the historical question information and the target maintenance document based on a preset large language model, and determining answer information corresponding to the question information.

[0082] In this embodiment, after the terminal device determines the question information, historical question information, and the target maintenance document corresponding to the question information, it can input the question information, historical question information, and the target maintenance document corresponding to the question information into a preset large language model to obtain reply information corresponding to the question information output by the large language model.

[0083] It should be noted that, in this embodiment, large language models (LLMs) are a type of natural language processing models with large-scale parameters and computing power. They are trained with a large amount of data and parameters to generate descriptive text similar to humans or answer natural language questions, such as GPT (Generative Pre-trained Transformer), Bard, T5 (Text-to-Text Transfer Transformer), PaLM (PathwaysLanguage Model), etc. This embodiment does not limit the type and specific construction method of the large language model.

[0084] The large language model in this embodiment can be used to generate answer information corresponding to the question information.

[0085] It is understandable that in the field of artificial intelligence, especially in natural language processing (NLP), "prompt" is a technique that can be used to guide a model to generate a specific output or perform a specific task. In the context of a large language model, a prompt refers to an input provided to the model, which is usually a question, a text, or any type of instruction to guide the model to generate a specific output. In this embodiment, the prompt can specifically be domain information, such as device-related domain knowledge.

[0086] In this embodiment, an instruction information prompt may be pre-constructed to guide the large language model to generate reply information. The instruction information prompt may be manually written by the user or automatically generated by the system based on the question information and maintenance documents, and there is no specific limitation on this.

[0087] For example, Figure 2 The prompt shown includes response strategy information, such as answer generation guidelines, and document format information, such as local knowledge base format instructions.

[0088] Therefore, in this embodiment, the terminal device can respond to the user's question information for equipment maintenance and determine the target maintenance document corresponding to the question information. In addition, the terminal device can also obtain the equipment failure information in the question information. After the terminal device extracts the equipment failure information from the question information, it can obtain the historical question information associated with the question information according to the equipment failure information, and then the question information, the historical question information and the target maintenance document corresponding to the question information can be input into the preset large language model to obtain the reply information corresponding to the question information output by the large language model. It can be seen that since the present application can simultaneously use the current question information and the historical question information, and combine the maintenance document as the input of the large language model, compared with only inputting the current question information, the large language model can more accurately predict the user's true intention, extract the required equipment maintenance reply information for the user from the maintenance document, and effectively improve the accuracy of the reply information. On this basis, the present application can solve various types of questions asked by users based on the knowledge aggregation and understanding ability of the large language model, provide users with effective maintenance strategies for different equipment failures, assist users in equipment maintenance, improve equipment maintenance efficiency, and also reduce equipment maintenance costs.

[0089] In one embodiment, in the above 20, “acquiring historical question information associated with the question information according to the device fault information” may include:

[0090] S201, querying historical question-and-answer dialogues according to the device failure information in the question information;

[0091] S202, determining the content similarity between each historical question information in the historical question-and-answer dialogue and the question information;

[0092] S203, if the content similarity between the historical question information in the historical question-and-answer dialogue and the question information is greater than a preset similarity threshold, setting the historical question information as the historical question information associated with the question information;

[0093] On this basis, the reply method in this application may also include:

[0094] If there are multiple historical question information associated with the question information, the weight coefficient corresponding to each historical question information is determined according to the similarity between the question information and each historical question information, so as to input the question information, the target maintenance document and the historical question information with each weight coefficient into the large language model.

[0095] In this embodiment, after obtaining the question information and the device failure information in the question information, the terminal device can query the historical question and answer dialogues based on the device failure information in the question information to determine the content similarity between each question information in the historical question and answer dialogue and the device failure information.

[0096] It should be noted that, in this embodiment, the historical question-and-answer dialogue may include historical question information input by each user, maintenance documents corresponding to the question information, and answer information corresponding to the question information, and the above historical question-and-answer dialogue may be stored in the cloud.

[0097] Furthermore, the terminal device may select question information whose content similarity with the device fault information is greater than a preset similarity threshold, and set the question information as historical question information associated with the currently input question information.

[0098] It is understandable that since the amount of information in the question information input by the user is very limited, and there may be a problem of missing key information, for example, the user only describes the equipment failure phenomenon, but does not enter equipment-related information, such as the equipment name, purpose, etc. Therefore, this embodiment can input the current question information and the historical question information associated with the question information into the large language model to enrich the amount of effective question information and improve the response content prediction accuracy of the large language model.

[0099] On this basis, if there are multiple historical question information associated with the question information, the weight coefficient corresponding to each historical question information is determined according to the similarity between the question information and each historical question information, so as to input the question information, the target maintenance document and the historical question information with each weight coefficient into the large language model.

[0100] It can be understood that in this embodiment, if there are multiple historical question information corresponding to a similarity greater than a preset similarity threshold, then the multiple historical question information can be sorted according to the similarity, and a preset number of historical question information with higher similarity can be used as historical question information associated with the current question information, and a corresponding weight coefficient can be assigned to each historical question information based on the similarity, wherein the higher the similarity, the greater the proportion of the historical weight system can be.

[0101] In one embodiment, in the above S202, “determining the content similarity between each question information in the historical question-and-answer dialogue and the device fault information” may include:

[0102] S2021, extracting historical fault information from each of the historical question information according to the structural information, logical information and / or semantic information of each of the historical question information in the historical question-answer pair, wherein the historical fault information includes fault description information and / or fault code;

[0103] S2022: Calculate the similarity between each of the historical question information and the device fault information in the question information according to the fault description information and / or the fault code.

[0104] In this embodiment, for each historical question information in the historical question pair, the structural information, logical information and / or semantic information of the historical question information can be obtained, wherein the structural information may include text paragraphs, etc., the logical information may include "and / or" information, and the semantic information may be the user intention and emotion contained in the text, etc.

[0105] Furthermore, the terminal device can extract historical fault information from the historical question information based on the structural information, logical information and / or semantic information of each historical question information in the historical question and answer pair. In this embodiment, the historical fault information may include fault description information and fault code, etc.

[0106] The terminal device can then compare the fault description information and / or fault code with the device fault information in the question information to calculate the similarity between each historical question information and the current question information, and then set the historical question information corresponding to the similarity greater than the preset similarity threshold as the historical question information associated with the current question information.

[0107] In one embodiment, in the above S30, “processing the question information, the historical question information and the target maintenance document based on a preset large language model to determine the answer information corresponding to the question information” may include:

[0108] S301, processing the question information and the target maintenance document through the large language model to obtain first answer information corresponding to the question information;

[0109] S302, processing the historical question information and the target maintenance document through the large language model to obtain second answer information corresponding to the historical question information;

[0110] S303: Process the first reply information and the second reply information through the large language model to determine reply information corresponding to the question information.

[0111] In this embodiment, if Figure 3As shown, after obtaining the current question information and the historical question information, the terminal device can input the question information and the target maintenance document into the large language model, process the question information and the target maintenance document through the large language model, and obtain the first reply information corresponding to the question information. In addition, the historical question information and the target maintenance document are also input into the large language model, and the question information and the target maintenance document are processed through the large language model to obtain the second reply information corresponding to the historical question information.

[0112] The first reply information and the second reply information can then be processed through a large language model to determine the reply information corresponding to the question information.

[0113] It should be noted that the instruction information prompt in this embodiment may include task requirements:

[0114] (1) Check the title and content of the knowledge base fragments one by one to evaluate their relevance to the user's question; if relevant, add the index value of the document fragment to related_paras;

[0115] (2) If the topic, subject, or object of the user's question matches the title or content of the reference segment, the segment is deemed to be related to the user's question, and its index value is added to related_paras;

[0116] (3) If the snippet contains the answer to the user’s question, it is considered relevant and its index value is added to related_paras.

[0117] (4) If a segment cannot directly answer the user's question but can provide useful auxiliary information, it should also be considered relevant and its index value should be added to related_params;

[0118] (5) If no segment is related to the user question, set relation = false; if there is one or more related segments, set relation = true;

[0119] (6) Output requirements: Use JSON format: {"relation":true / false,"related_paras":[1,2,3]}. No additional explanation is required.

[0120] On this basis, the terminal device can calculate the relevance score between the question information and the target maintenance document according to the title information, maintenance content information and question information, and obtain the segment corresponding to the relevance score greater than or equal to the preset relevance threshold.

[0121] Then, the fragments corresponding to the correlation scores greater than or equal to the preset correlation threshold can be marked, and finally all the marked fragments can be assembled to obtain the first reply information and the second reply information.

[0122] In one embodiment, in the above S303, “processing the first reply information and the second reply information by the large language model to determine the reply information corresponding to the question information” may include:

[0123] S3031, comparing overlapping characters between the first reply information and the second reply information by using the large language model;

[0124] S3032: If there are no overlapping characters between the first reply information and the second reply information, setting the first reply information as the reply information by using the large language model;

[0125] S3033, if there are some overlapping characters between the first reply information and the second reply information, optimizing the first reply information according to the second reply information by using the large language model to obtain reply information;

[0126] S3034: If the first reply information is the same as the second reply information, the first reply information is set as the reply information through the large language model, and the corresponding historical question information is optimized according to the target question information to obtain an updated historical question and answer pair.

[0127] In this embodiment, after acquiring the first reply information and the second reply information through the large language model, the terminal device can compare the overlapping characters between the first reply information and the second reply information.

[0128] If there are no overlapping characters between the first reply information and the second reply information, it means that the first reply information and the second reply information are not related at this time, and the second reply information does not meet the user's true intention. At this time, the first reply information can be directly set as the reply information through the large language model.

[0129] If there are some overlapping characters between the first reply information and the second reply information, the first reply information can be optimized according to the second reply information through a large language model to obtain the reply information. For example, the first reply information and the second reply information can be fused to obtain the reply information.

[0130] If the first reply information is the same as the second reply information, the first reply information can be set as the reply information through the large language model, and the historical reply information in the corresponding historical question and answer pairs can be optimized according to the reply information to continuously optimize the historical question and answer pairs, improve the prediction accuracy of the large language model, and be more in line with the user's true intention.

[0131] It is worth noting that in this embodiment, the reply information in this embodiment not only includes the reply content corresponding to the question information, but also includes the source information of the reply content, for example, whether the reply content comes from the local knowledge base and / or the cloud knowledge base, and may also include reference path information. For example, if the reply content comes from the cloud knowledge base, the reference path information may be web link information, so that the user can jump to the source document through the web link information. In addition, the reply information may also include knowledge search options, maintenance record query options, and fault code query options.

[0132] In one embodiment, in the above S10, “obtaining a target maintenance document in response to the inquiry information regarding equipment maintenance” may include:

[0133] S101, in response to a question about equipment maintenance, querying a target knowledge base, obtaining a plurality of knowledge slices in the target knowledge base, and splicing the plurality of knowledge slices to obtain a target maintenance document;

[0134] In this embodiment, the terminal device can query the target database based on the question information, obtain multiple knowledge slices, and splice the multiple knowledge slices to obtain the target maintenance document. For example, the terminal device can splice the multiple knowledge slices to obtain the target maintenance document based on the context information between the multiple knowledge slices.

[0135] According to the above-mentioned embodiment, the database may include a local knowledge base and / or a cloud knowledge base, wherein the target database stores equipment-related documents, such as product manuals, maintenance records, equipment-related videos, drawings, spare parts library files, and major accident reports, etc. The cloud knowledge base may be understood as Internet information, such as websites, forums, and other equipment-related information, wherein the present embodiment may utilize the local knowledge base and the cloud knowledge base to provide users with more comprehensive and accurate response information, and assist users in equipment maintenance.

[0136] The steps of constructing the target knowledge base include:

[0137] Obtain maintenance documents for multiple devices;

[0138] For each equipment maintenance document, performing layout analysis on the equipment maintenance document to identify target elements in the equipment maintenance document;

[0139] According to the target element, the equipment maintenance document is sliced ​​to obtain a plurality of knowledge slices;

[0140] For each of the knowledge slices, vectorization is performed on the knowledge slice, and the vectorized knowledge slice is imported into the target knowledge base.

[0141] In this embodiment, the terminal device can obtain multiple equipment maintenance documents. The equipment maintenance documents can be product manuals, historical maintenance records, equipment drawings and other documents. For example, a large number of documents can be collected from the local and cloud, and the documents in the cloud can be downloaded to the local. This embodiment does not limit the type of documents, such as plain text type, graphic type (doc / docx, pdf), table (excel), audio and video, etc.

[0142] On this basis, if Figure 4 As shown, for each device maintenance document, the terminal device can perform layout analysis on the device maintenance document to identify target elements in the device maintenance document, wherein the target elements may include tables, pictures, text, etc. in the device maintenance document, wherein the layout analysis may include but is not limited to optical character recognition (OCR), image extraction, table extraction, layout recovery, etc., which will not be elaborated here.

[0143] Then, the terminal device can slice the equipment maintenance document according to the target element to obtain multiple knowledge slices, such as knowledge slice 1, knowledge slice 2, ..., knowledge slice n, etc. The knowledge slices are embedded and vectorized, and finally the vectorized knowledge slices are imported into the local knowledge base.

[0144] In this embodiment, the cloud knowledge base can be constructed in a variety of ways according to different technologies and requirements. For example, the following steps may be included:

[0145] Building a cloud database can follow the following steps and principles:

[0146] (1) Architecture design:

[0147] Layered architecture: Divide the database system into multiple layers, such as access layer, computing layer, storage layer, etc., to improve the scalability and maintainability of the system;

[0148] Distributed architecture: Distributed database technologies such as sharding and replication are used to achieve horizontal expansion of data and automatic failure transfer;

[0149] High availability design: Ensure high availability of the database through master-slave replication, multi-copy deployment, etc.

[0150] (2) Technology selection:

[0151] Database system: Choose a suitable database system based on business needs, including relational databases (such as MySQL, PostgreSQL), NoSQL databases (such as MongoDB, Cassandra), or NewSQL databases (such as TiDB, CockroachDB);

[0152] Virtualization technology: Use containerization technologies such as Docker and Kubernetes to achieve rapid deployment, resource isolation, and elastic scaling of databases;

[0153] Cloud service platform: Select a suitable cloud service platform as the underlying infrastructure, and use the computing, storage, network and other resources it provides to build a cloud database environment.

[0154] (3) Deployment and implementation:

[0155] Environment preparation: Create virtual machines, configure networks, mount storage and other resources on the cloud service platform to prepare for database deployment;

[0156] Database installation: Based on the technical selection results, download and install the database software, and configure the necessary parameters and permissions.

[0157] (4) Selection of cloud database service providers, such as Alibaba Cloud, Tencent Cloud, and Baidu Cloud.

[0158] Through the above steps and principles, a stable, reliable, and elastically scalable cloud database service can be built.

[0159] Through the above operations, this embodiment can construct a vectorized local knowledge base and a cloud database. On this basis, this embodiment can use the local knowledge base and combine it with the maintenance documents in the cloud knowledge base to train the large language model in this embodiment.

[0160] In one embodiment, the step of training the large language model includes:

[0161] Acquire question training information, and extract maintenance training documents corresponding to the question training information from a local knowledge base;

[0162] Inputting the question training information and the maintenance training document into an initial large language model to obtain answer prediction information;

[0163] Calculating the recall accuracy and / or question-answering accuracy of the initial large language model according to the answer prediction information and the answer annotation information corresponding to the question training information;

[0164] If at least one of the recall accuracy and the question-answer accuracy is less than a corresponding accuracy threshold, fine-tuning the initial large language model to adjust a weight parameter of the initial large language model;

[0165] Until the recall accuracy and the question-answer accuracy both reach corresponding accuracy thresholds, a second largest language model is obtained.

[0166] like Figure 5 As shown, in this embodiment, the terminal device can input the question training information and the maintenance training document into the initial large language model, so that the initial large language model processes the question training information and the maintenance training document to obtain the answer prediction information.

[0167] The terminal device then calculates the recall accuracy and / or question-answering accuracy of the initial large language model based on the answer prediction information and the answer annotation information corresponding to the question training information.

[0168] If it is determined that at least one of the recall accuracy and the question-answering accuracy is less than the corresponding accuracy threshold, it means that the initial large language model cannot accurately predict the answer information corresponding to the question information. In this case, the initial large language model can be fine-tuned to adjust the weight parameters of the initial large language model. For example, the initial large language model can include a Transformer encoder layer and a Transformer decoder layer. When training the initial large language model, the weight parameters of the initial large language model can be adjusted according to the output recall accuracy and / or question-answering accuracy.

[0169] By repeatedly adjusting the weight parameters of the initial large language model until the recall accuracy and question-answer accuracy reach the corresponding accuracy thresholds, it means that the initial large language model can accurately predict the user's intention based on the question information input by the user and feedback the corresponding answer information. Therefore, the initial large language model at this time can be set as the language large model.

[0170] Specifically, for example, Figure 6 As shown, this embodiment can detect whether the initial large language model is trained through multiple evaluation parameters such as recall accuracy, answer fidelity, and answer completeness output by the initial large language model. If it is detected that the recall accuracy is 99% at this time, which has exceeded the corresponding accuracy threshold (such as 98%), and the question-answer accuracy is 97.34%, which has reached the corresponding accuracy threshold (such as 95%), then the initial large language model at this time can be set as the large language model.

[0171] In one embodiment, after the above S30, "processing the question information, the historical question information and the target maintenance document based on a preset large language model to determine the answer information corresponding to the question information", the following may also be included:

[0172] S40, obtaining multiple sets of question information and corresponding answer information;

[0173] S50, searching for historical question information in historical question-answer pairs according to the question information;

[0174] S60: If the similarity between the question information and the historical question information is less than a preset similarity, the question information and the corresponding answer information are set as a historical question-answer pair for storage to update the historical question-answer pair.

[0175] In this embodiment, after the terminal device obtains the answer information corresponding to the question information through the large language model, for each set of question information and the corresponding answer information, the historical question information in the historical question-answer pair can be queried according to the question information.

[0176] If the similarity between the question information and the historical question information is greater than or equal to the preset similarity, it means that the question information is a repeated question information. In this case, the question information and the corresponding answer information can be ignored to avoid generating repeated question and answer pairs.

[0177] If the similarity between the question information and the historical question information is less than the preset similarity, it means that the question information is not a repeated question information. In this case, the question information and the corresponding answer information can be set as a historical question and answer pair for storage to enrich the historical question and answer pairs.

[0178] Accordingly, the embodiment of the present application also provides a reply device, such as Figure 7 As shown, the device may include:

[0179] The first acquisition module 1001 is used to obtain a target maintenance document in response to a question about equipment maintenance, and to obtain equipment failure information in the question;

[0180] The second acquisition module 1002 is used to acquire at least one historical question information associated with the question information according to the device fault information; wherein the content similarity between the question information and the historical question information is greater than a preset similarity threshold;

[0181] The determination module 1003 is used to process the question information, the historical question information and the target maintenance document based on a preset large language model to determine the answer information corresponding to the question information.

[0182] Optionally, the second acquisition module 1002 is further configured to:

[0183] According to the device failure information in the question information, query the historical question-answering dialogue, and obtain the content similarity between each historical question information in the historical question-answering dialogue and the question information;

[0184] If the content similarity between the historical question information in the historical question-and-answer dialogue and the question information is greater than a preset similarity threshold, setting the historical question information as the historical question information associated with the question information;

[0185] The reply device in this application also includes:

[0186] The second determination module is used to determine the weight coefficient corresponding to each historical question information according to the similarity between the question information and each historical question information if there are multiple historical question information associated with the question information, so as to input the question information, the target maintenance document and the historical question information with each weight coefficient into the large language model.

[0187] Optionally, the second acquisition module 1002 is further configured to:

[0188] Extracting historical fault information from each of the historical question information according to the structural information, logical information and / or semantic information of each of the historical question information in the historical question-answer pair, wherein the historical fault information includes fault description information and / or fault code;

[0189] The similarity between each of the historical question information and the device fault information in the question information is calculated according to the fault description information and / or the fault code.

[0190] Optionally, the determination module 1003 is further configured to:

[0191] Processing the question information and the target maintenance document through the large language model to obtain first answer information corresponding to the question information;

[0192] Processing the historical question information and the target maintenance document through the large language model to obtain second answer information corresponding to the historical question information;

[0193] The first reply information and the second reply information are processed by the large language model to determine the reply information corresponding to the question information.

[0194] Optionally, the determination module 1003 is further configured to:

[0195] Comparing the overlapping characters between the first reply information and the second reply information by using the large language model;

[0196] If there are no overlapping characters between the first reply information and the second reply information, setting the first reply information as the reply information by using the large language model;

[0197] If there are some overlapping characters between the first reply information and the second reply information, optimizing the first reply information according to the second reply information by using the large language model to obtain reply information;

[0198] If the first reply information is the same as the second reply information, the first reply information is set as the reply information through the large language model, and the historical reply information in the corresponding historical question and answer pair is optimized according to the reply information.

[0199] Optionally, the first acquisition module 1001 is further used to:

[0200] In response to the question information about equipment maintenance, query the target knowledge base to obtain multiple knowledge slices in the target knowledge base, and splice the multiple knowledge slices to obtain a target maintenance document;

[0201] The steps of constructing the target knowledge base include:

[0202] Obtain maintenance documents for multiple devices;

[0203] For each equipment maintenance document, performing layout analysis on the equipment maintenance document to identify target elements in the equipment maintenance document;

[0204] According to the target element, the equipment maintenance document is sliced ​​to obtain a plurality of knowledge slices;

[0205] For each of the knowledge slices, vectorization is performed on the knowledge slice, and the vectorized knowledge slice is imported into the target knowledge base.

[0206] Optionally, the training process of the large language model includes:

[0207] Acquire question training information, and extract maintenance training documents corresponding to the question training information from a local knowledge base;

[0208] Inputting the question training information and the maintenance training document into an initial large language model to obtain answer prediction information;

[0209] Calculating the recall accuracy and / or question-answering accuracy of the initial large language model according to the answer prediction information and the answer annotation information corresponding to the question training information;

[0210] If at least one of the recall accuracy and the question-answer accuracy is less than a corresponding accuracy threshold, fine-tuning the initial large language model to adjust a weight parameter of the initial large language model;

[0211] Until the recall accuracy and the question-answer accuracy both reach corresponding accuracy thresholds, a second largest language model is obtained.

[0212] Optionally, the reply device in the present application further includes:

[0213] A storage module is used to obtain multiple groups of question information and corresponding answer information; based on the question information, query the historical question information in the historical question and answer pair; if the similarity between the question information and the historical question information is less than a preset similarity, set the question information and the corresponding answer information as a historical question and answer pair for storage to update the historical question and answer pair.

[0214] The specific implementation of the above operations can be found in the previous embodiments, which will not be described in detail here.

[0215] Accordingly, the present application also provides an electronic device, such as Figure 8 As shown, Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device 1100 includes a processor 1101 having one or more processing cores, a memory 1102 having one or more computer-readable storage media, and a computer program stored in the memory 1102 and executable on the processor. The processor 1101 is electrically connected to the memory 1102. It will be understood by those skilled in the art that the electronic device structure shown in the figure does not constitute a limitation on the electronic device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0216] The processor 1101 is the control center of the electronic device 1100, and uses various interfaces and lines to connect various parts of the entire electronic device 1100. By running or loading software programs and / or units stored in the memory 1102, and calling data stored in the memory 1102, the processor 1101 executes various functions of the electronic device 1100 and processes data, thereby monitoring the electronic device 1100 as a whole. The processor 1101 can be a processor CPU, a graphics processor GPU, a network processor (Network Processor, NP), etc., and can implement or execute the various methods, devices, steps and logic block diagrams disclosed in the embodiments of the present application.

[0217] In the embodiment of the present application, the processor 1101 in the electronic device 1100 will load instructions corresponding to the processes of one or more application programs into the memory 1102 according to the following steps, and the processor 1101 will run the application programs stored in the memory 1102 to implement various functions, such as:

[0218] In response to a question about equipment maintenance, obtaining a target maintenance document and obtaining equipment failure information in the question;

[0219] Acquire, according to the device fault information, at least one historical question information associated with the question information; wherein the content similarity between the question information and the historical question information is greater than a preset similarity threshold;

[0220] The question information, the historical question information and the target maintenance document are processed based on a preset large language model to determine answer information corresponding to the question information.

[0221] The specific implementation of the above operations can be found in the previous embodiments, which will not be described in detail here.

[0222] Optional, such as Figure 8 As shown, the electronic device 1100 also includes: a touch display screen 1103 , a radio frequency circuit 1104 , an audio circuit 1105 , an input unit 1106 , and a power supply 1107 .

[0223] In addition, the electronic device 1100 further includes: a security module 1108 , a device chip 1109 and a hard disk drive 1110 .

[0224] The processor 1101 is electrically connected to the touch screen 1103, the radio frequency circuit 1104, the audio circuit 1105, the input unit 1106 and the power supply 1107. Those skilled in the art will appreciate that Figure 8 The electronic device structure shown in the figure does not constitute a limitation of the electronic device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0225] The touch display screen 1103 can be used to display a graphical user interface and receive operation instructions generated by the user acting on the graphical user interface. The touch display screen 1103 may include a display panel and a touch panel. Among them, the display panel can be used to display information input by the user or information provided to the user and various graphical user interfaces of the electronic device, and these graphical user interfaces can be composed of graphics, text, icons, videos and any combination thereof. Optionally, the display panel can be configured in the form of a liquid crystal display (LCD, Liquid Crystal Display), an organic light-emitting diode (OLED, Organic Light-EmittingDiode) and the like. The touch panel can be used to collect the user's touch operation on or near it (such as the user uses any suitable object or attachment such as a finger, a stylus, etc. on the touch panel or near the touch panel), and generate corresponding operation instructions, and the operation instructions execute corresponding programs. Optionally, the touch panel may include two parts: a touch detection device and a touch controller. Among them, the touch detection device detects the user's touch orientation, detects the signal brought by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device, converts it into the touch point coordinates, and then sends it to the processor 1101, and can receive the command sent by the processor 1101 and execute it. The touch panel can cover the display panel. When the touch panel detects a touch operation on or near it, it is transmitted to the processor 1101 to determine the type of touch event, and then the processor 1101 provides a corresponding visual output on the display panel according to the type of touch event. In an embodiment of the present application, the touch panel and the display panel can be integrated into the touch display screen 1103 to realize the input and output functions. However, in some embodiments, the touch panel and the touch panel can be used as two independent components to realize the input and output functions. That is, the touch display screen 1103 can also be used as a part of the input unit 1106 to realize the input function.

[0226] The radio frequency circuit 1104 may be used to send and receive radio frequency signals, so as to establish wireless communication with a network device or other electronic devices through wireless communication, and to send and receive signals between the network device or other electronic devices.

[0227] The audio circuit 1105 can be used to provide an audio interface between the user and the electronic device through a speaker and a microphone. The audio circuit 1105 can transmit the electrical signal converted from the received audio data to the speaker, which is converted into a sound signal for output; on the other hand, the microphone converts the collected sound signal into an electrical signal, which is received by the audio circuit 1105 and converted into audio data, and then the audio data is output to the processor 1101 for processing, and then sent to another electronic device through the radio frequency circuit 1104, or the audio data is output to the memory 1102 for further processing. The audio circuit 1105 may also include an earplug jack to provide communication between an external headset and an electronic device.

[0228] The input unit 1106 may be used to receive input numbers, character information or user feature information (such as fingerprint, iris, facial information, etc.), and generate keyboard, mouse, joystick, optical or trackball signal input related to user settings and function control.

[0229] The power supply 1107 is used to supply power to various components of the electronic device 1100. Optionally, the power supply 1107 can be logically connected to the processor 1101 through a power management system, so that the power management system can manage charging, discharging, and power consumption. The power supply 1107 can also include one or more DC or AC power supplies, recharging systems, power failure detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0230] The security module 1108 can be used to respond to a device startup event on the electronic device 1100 and read encryption information and device information from the non-volatile memory of the device chip 1109; the encryption information and device information stored in the non-volatile memory are obtained by authentication when the electronic device first accesses the cloud; a drive encryption key is generated based on the encryption information and device information, and the hard disk drive of the device chip 1109 is activated; the hard disk drive 1110 can be used to receive data operation instructions triggered by the target application on the activated hard disk drive, and operate the target data based on the drive encryption key.

[0231] The specific implementation of the above operations can be found in the previous embodiments, which will not be described in detail here.

[0232] although Figure 8 Not shown, the electronic device 1100 may also include a camera, a sensor, a wireless fidelity module, a Bluetooth module, etc., which will not be described in detail here.

[0233] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0234] A person of ordinary skill in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be completed by instructions, or by controlling related hardware through instructions. The instructions may be stored in a computer-readable storage medium and loaded and executed by a processor.

[0235] To this end, an embodiment of the present application provides a computer-readable storage medium, in which a plurality of computer programs are stored. The computer program can be loaded by a processor to execute any one of the reply methods provided in the embodiment of the present application. The computer program can execute the steps of the following reply method:

[0236] In response to a question about equipment maintenance, obtaining a target maintenance document and obtaining equipment failure information in the question;

[0237] Acquire, according to the device fault information, at least one historical question information associated with the question information; wherein the content similarity between the question information and the historical question information is greater than a preset similarity threshold;

[0238] The question information, the historical question information and the target maintenance document are processed based on a preset large language model to determine answer information corresponding to the question information.

[0239] The specific implementation of the above operations can be found in the previous embodiments, which will not be described in detail here.

[0240] The computer-readable storage medium may include: a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0241] Since the computer program stored in the computer-readable storage medium can execute any one of the reply methods provided in the embodiments of the present application, the beneficial effects that can be achieved by any one of the reply methods provided in the embodiments of the present application can be achieved. Please refer to the previous embodiments for details and will not be repeated here.

[0242] In the above-mentioned electronic device, computer-readable storage medium, and electronic device embodiments, the description of each embodiment has its own emphasis. For parts not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process and beneficial effects of the electronic device, computer-readable storage medium, electronic device and its corresponding units described above can refer to the description of the reply method in the above embodiment, and will not be repeated here.

[0243] The above is a detailed introduction to a reply method, device, electronic device, and computer-readable storage medium provided in the embodiments of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for technical personnel in this field, according to the ideas of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A reply method, characterized in that: The method comprises: In response to a question about equipment maintenance, obtaining a target maintenance document and obtaining equipment failure information in the question; Acquire, according to the device fault information, at least one historical question information associated with the question information; wherein the content similarity between the question information and the historical question information is greater than a preset similarity threshold; The question information, the historical question information and the target maintenance document are processed based on a preset large language model to determine answer information corresponding to the question information.

2. The reply method according to claim 1, characterized in that: The acquiring, according to the device fault information, at least one historical question information associated with the question information comprises: According to the equipment failure information in the question information, query the historical question-answering dialogue, and obtain the content similarity between each historical question information in the historical question-answering dialogue and the question information; If the content similarity between the historical question information in the historical question-and-answer dialogue and the question information is greater than a preset similarity threshold, setting the historical question information as the historical question information associated with the question information; The method further comprises: If there are multiple historical question information associated with the question information, the weight coefficient corresponding to each historical question information is determined according to the similarity between the question information and each historical question information, so as to input the question information, the target maintenance document and the historical question information with each weight coefficient into the large language model.

3. The reply method according to claim 2, characterized in that: The obtaining of the content similarity between each historical question information in the historical question-and-answer dialogue and the question information includes: Extracting historical fault information from each of the historical question information according to the structural information, logical information and / or semantic information of each of the historical question information in the historical question-answer pair, wherein the historical fault information includes fault description information and / or fault code; The similarity between each of the historical question information and the device fault information in the question information is calculated according to the fault description information and / or the fault code.

4. The reply method according to claim 1, characterized in that: The processing of the question information, the historical question information and the target maintenance document based on a preset large language model to determine the answer information corresponding to the question information includes: Processing the question information and the target maintenance document through the large language model to obtain first answer information corresponding to the question information; Processing the historical question information and the target maintenance document through the large language model to obtain second answer information corresponding to the historical question information; The first reply information and the second reply information are processed by the large language model to determine the reply information corresponding to the question information.

5. The reply method according to claim 4, characterized in that: The processing of the first reply information and the second reply information by the large language model to determine the reply information corresponding to the question information includes: Comparing the overlapping characters between the first reply information and the second reply information by using the large language model; If there are no overlapping characters between the first reply information and the second reply information, setting the first reply information as the reply information by using the large language model; If there are some overlapping characters between the first reply information and the second reply information, optimizing the first reply information according to the second reply information by using the large language model to obtain reply information; If the first reply information is the same as the second reply information, the first reply information is set as the reply information through the large language model, and the historical reply information in the corresponding historical question and answer pair is optimized according to the reply information.

6. The reply method according to claim 1, characterized in that: The step of obtaining a target maintenance document in response to the inquiry information regarding equipment maintenance includes: In response to the question information about equipment maintenance, query the target knowledge base to obtain multiple knowledge slices in the target knowledge base, and splice the multiple knowledge slices to obtain a target maintenance document; The steps of constructing the target knowledge base include: Obtain maintenance documents for multiple devices; For each equipment maintenance document, performing layout analysis on the equipment maintenance document to identify target elements in the equipment maintenance document; According to the target element, the equipment maintenance document is sliced ​​to obtain a plurality of knowledge slices; For each of the knowledge slices, vectorization is performed on the knowledge slice, and the vectorized knowledge slice is imported into the target knowledge base.

7. The reply method according to claim 1, characterized in that: The training steps of the large language model include: Acquire question training information, and extract maintenance training documents corresponding to the question training information from a local knowledge base; Inputting the question training information and the maintenance training document into an initial large language model to obtain answer prediction information; Calculating the recall accuracy and / or question-answering accuracy of the initial large language model according to the answer prediction information and the answer annotation information corresponding to the question training information; If at least one of the recall accuracy and the question-answer accuracy is less than a corresponding accuracy threshold, fine-tuning the initial large language model to adjust a weight parameter of the initial large language model; Until the recall accuracy and the question-answer accuracy both reach corresponding accuracy thresholds, a second largest language model is obtained.

8. The reply method according to any one of claims 1 to 7, characterized in that: After the question information, the historical question information and the target maintenance document are processed based on the preset large language model and the answer information corresponding to the question information is determined, the method includes: Get multiple sets of question information and corresponding answer information; According to the question information, query the historical question information in the historical question and answer pairs; If the similarity between the question information and the historical question information is less than a preset similarity, the question information and the corresponding answer information are set as a historical question-answer pair for storage to update the historical question-answer pair.

9. A reply device, characterized in that: The device comprises: A first acquisition module, configured to acquire a target maintenance document in response to a question about equipment maintenance, and acquire equipment failure information in the question; A second acquisition module is used to acquire at least one historical question information associated with the question information according to the equipment fault information; wherein the content similarity between the question information and the historical question information is greater than a preset similarity threshold; A determination module is used to process the question information, the historical question information and the target maintenance document based on a preset large language model to determine the answer information corresponding to the question information.

10. An electronic device, characterized in that: The electronic device further comprises: a memory and a processor, wherein the memory stores a plurality of instructions; and the processor loads instructions from the memory to execute the steps of the reply method as claimed in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the steps of the reply method according to any one of claims 1 to 8.

Images