Enterprise management system question and answer control method and system based on large language model

Abstract

The invention discloses an enterprise management system question and answer control method and system based on a large language model. The method is described. The method comprises the following steps: constructing a pre-training model; managing the knowledge base; managing cache knowledge; and controlling questions and answers. According to the method, the pre-training model is constructed through the enterprise management system question input by the user to be questioned and answered and in combination with the corresponding text data, and then the content of the preset management system knowledge base is managed and updated through the pre-training model and in combination with the set access permission to obtain the question and answer data set; performing parameter management on the first parameter information data in the pre-training model according to the question and answer data set, and finally obtaining the first answer data in combination with the first parameter information data after parameter management, so as to improve the association degree between the enterprise management system question and answer and the user feedback. The problem of low association degree between enterprise management system questions and answers and user feedback in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of natural language processing technology, and in particular to a method and system for question-answering control of an enterprise management system based on a large language model. Background Art

[0002] With the rapid development of information technology and the advent of the big data era, enterprises are facing the challenges of massive information processing and knowledge management. At the same time, the rapid development of deep learning technology has brought revolutionary changes to the field of natural language processing. Language Large Model (LLM) is a natural language processing model based on deep learning. It can learn the grammar and semantics of natural language and has broad application prospects in many fields such as document classification retrieval, emotional intent recognition, and machine translation dialogue. However, there are a large number of rules and regulations in enterprise management. Most people in different positions need to learn several or even dozens of management regulations before taking up their posts, while non-management personnel generally have low knowledge reserves. The traditional enterprise management system question-and-answer system stores existing document knowledge in the database, and then uses keywords in the question to query the similarity of the knowledge in the database. This method has poor scalability and weak interpretability, and cannot perform efficient question-and-answer control. Therefore, how to efficiently and accurately answer questions related to enterprise management systems has become an important topic in enterprise informatization construction.

[0003] Existing technologies fully integrate background knowledge by calling large language models and adding prompt templates, and then combine pipeline processing flows to realize automatic and continuous updating of graph database knowledge. Finally, they provide users with more efficient and accurate question-and-answer services through online services.

[0004] For example, the patent application with publication number: CN116932708A discloses an open-domain natural language reasoning question-answering system and method driven by a large language model, which includes: rewriting user questions to obtain rewritten questions; managing the computing and knowledge resources of the large language model and outputting the computing and knowledge resources of the large language model to the question-answering module in combination with the type of rewritten questions; obtaining candidate answers to the rewritten questions by reasoning based on the rewritten questions and the computing and knowledge resources of the large language model and performing analysis; and obtaining the final answer to the rewritten question and explainable information of the final answer based on the results of the analysis.

[0005] For example, the patent application with publication number CN117609436A discloses a university scientific research management question-and-answer system that combines a knowledge graph and a large language model, including: cleaning the collected data, structuring the available data, and storing it in a database in different formats; processing user questions and extracting key feature questions to obtain question classification information; generating diverse and high-quality answers based on the question classification information and combining the query results and the model's own knowledge.

[0006] However, in the process of implementing the technical solution of the present invention, the present invention finds that the above technology has at least the following technical problems:

[0007] In the prior art, since it is impossible to obtain the feedback results of the users to be answered in real time during the question-and-answer process, even if the feedback results are obtained, they are not fully utilized, and there is a problem of low correlation between the questions and answers of the enterprise management system and the user feedback. Summary of the invention

[0008] The embodiments of the present invention solve the problem of low correlation between enterprise management system questions and answers and user feedback in the prior art by providing an enterprise management system question and answer control method and system based on a large language model, thereby improving the correlation between enterprise management system questions and answers and user feedback.

[0009] The embodiment of the present invention provides a method for controlling enterprise management system question and answer based on a language large model, comprising the following steps:

[0010] A pre-trained model is constructed based on the enterprise management system questions input by the user to be answered and combined with the corresponding text data. The pre-trained model is used to parse the input enterprise management system questions in combination with the training data set. The training data set is a collection of reference enterprise management system questions and corresponding reference answers. The content of the preset management system knowledge base is managed and updated through the pre-trained model in combination with the set access rights to obtain a question and answer data set. The question and answer data set represents the enterprise management system questions of the user to be answered and the corresponding reference answers. Parameter management is performed on the first parameter information data in the pre-trained model according to the question and answer data set. The first parameter information data includes parameter information data and non-parameter information data. The parameter management represents updating, storing, querying and recording the first parameter information data. The first answer data is obtained in combination with the first parameter information data after parameter management. The first answer data represents the first parameter information data visualized in the form of natural language.

[0011] Optionally, the specific construction method of the pre-trained model is as follows: collect text data of enterprise management systems and perform data preprocessing, wherein the data preprocessing is used to remove data anomalies; design a model framework based on the results of data preprocessing, and evaluate the designed model framework to obtain a model framework compliance index, wherein the model framework has the function of learning and understanding the text data of the enterprise management system, and the model framework compliance index indicates the degree of match between the designed model framework and the preset model framework; formulate a model training task based on the training data set and in combination with the model framework compliance index that meets preset conditions, wherein the model training task is to perform text classification, question-answer pair classification and sequence labeling on enterprise management mechanism questions input by question-answering users; initialize model parameters using weights and construct a pre-trained model in combination with the training data set and the model training task.

[0012] Optionally, the specific method for obtaining the question and answer data set is as follows: by analyzing the enterprise management system questions input by the user to be answered, feature information is extracted and the feature information is sorted, and the information sorting represents structured processing and visual classification of the feature information; in combination with a security key, the user to be answered is given access rights to the content of a preset management system knowledge base, and the security key consists of a dynamic verification code, biometric identification, and IP address restriction; based on the result of the information sorting, it is queried whether there is corresponding content in the preset management system knowledge base, and if so, a question and answer data pair is generated, otherwise the preset management system knowledge base is expanded and combined with a pre-trained model to generate a question and answer data pair, and the question and answer data pair represents the corresponding answer to the enterprise management system question input by the user to be answered; the question and answer data pair is screened and classified to generate a question and answer data set, and fed back to the user to be answered in the form of a chart.

[0013] Optionally, the specific steps of the parameter management include: receiving the question and answer data set in real time and updating the parameter information data in the question and answer data set; judging whether the non-parameter information data needs to be updated in combination with the parameter update index coefficient, and if necessary, updating and storing it, otherwise storing the question and answer data set after the data update, the parameter update index coefficient indicates the necessity of updating the non-parameter information data; retrieving the corresponding first parameter information data according to the query conditions input by the question and answer user and optimizing the performance of parameter management; locating the abnormal information type that occurs during the optimization process in combination with the abnormal information classification index and performing secondary processing, the abnormal information classification index is used to describe the recognition accuracy of the abnormal information type that occurs.

[0014] Optionally, the specific process of obtaining the first answer data is: performing secondary analysis on the enterprise management system question input by the question-answering user through a pre-trained model, and the secondary analysis is used to analyze and answer the keywords, syntax and semantic structures in the input enterprise management system question; according to the result of the secondary analysis, searching for the corresponding question and answer content in the preset management system knowledge base and combining the first parameter information data after parameter management to obtain the first answer data for feedback to the question-answering user.

[0015] Optionally, the specific method for obtaining the model framework compliance index is as follows: obtain the model framework structure feature data by analyzing the enterprise management system questions input by the question-answering user, and obtain the model structure matching degree according to the model framework structure feature data, wherein the model structure matching degree indicates the degree of fit between the enterprise management system questions input by the question-answering user and the internal structure of the model framework; analyze the relationship between the internal structure connection nodes of the model framework and obtain the model structure complexity in combination with the parameter weights of the model framework, wherein the model structure complexity is used to measure the complexity of the model structure framework; evaluate the enterprise management system questions input by the question-answering user according to the training data set, and compare them with the expected performance of the model framework to obtain the model performance evaluation error rate, wherein the model performance evaluation error rate is used to measure the gap between the performance of the model framework in the actual application process and the expected performance; obtain the corresponding relative deviation by combining the reference values ​​and reference deviations of the model structure matching degree, the model structure complexity and the model performance evaluation error rate, and judge whether the obtained relative deviations are all less than 1, if they are all greater than 1, it indicates that the corresponding model framework does not meet the requirements, otherwise the model framework compliance index is calculated in combination with the average value of the corresponding data, and the model framework compliance index is calculated by the following formula:

[0016]

[0017] Where m is the number of the user to be answered, m = 1, 2, ...., X, X is the total number of users to be answered, mn is the number of questions and answers corresponding to the mth user to be answered, mn = m.1, m.2, ...., mY, mY is the total number of questions and answers corresponding to the mth user to be answered, CI m.n represents the model framework conformity index of the mnth question and answer corresponding to the mth user to be asked and answered, e is a natural constant, g1 represents the weight of the average relative deviation of the model structure matching degree relative to the model framework conformity index, ΔK represents the reference deviation of the model structure matching degree, K m.n represents the model structure matching degree of the mnth question and answer corresponding to the mth user to be asked and answered, K0 represents the model structure reference matching degree, g2 represents the weight of the average relative deviation of the model structure complexity relative to the model framework conformity index, ΔQ represents the model structure complexity reference deviation, Q m.nrepresents the model structure complexity of the mnth question and answer corresponding to the mth user to be asked and answered, Q0 represents the model structure reference complexity, g3 represents the weight of the average relative deviation of the model performance evaluation error rate relative to the model framework compliance index, ΔW represents the model performance evaluation error rate reference deviation, W m.n It represents the model performance evaluation error rate of the mnth question and answer corresponding to the mth user to be asked and answered, and W0 represents the model performance evaluation reference error rate.

[0018] Optionally, the parameter update index coefficient is calculated by the following formula:

[0019]

[0020] In the formula, RI m represents the parameter update index coefficient corresponding to the mth question-answering user, ΔA represents the reference deviation of the data change rate coefficient, and A m represents the average coefficient of data change rate after the mth user asks and answers the question, A0 represents the reference coefficient of data change rate, ΔB represents the reference deviation of configuration consistency coefficient, B m represents the average coefficient of configuration consistency of the mth user after answering the question, B0 represents the reference coefficient of configuration consistency, ΔF represents the reference deviation of the model performance improvement coefficient, and F m It represents the average coefficient of model performance improvement of the mth user after answering the question, F0 represents the reference improvement coefficient of model performance; the data change rate coefficient is used to measure the degree of change of the question and answer data set after the user completes the question and answer; the configuration consistency coefficient is used to measure the degree of match between the pre-trained model configuration and the actual operating environment; the model performance improvement coefficient is used to measure the degree of improvement of the performance of the pre-trained model after the parameters are updated compared with the performance before the parameters are updated.

[0021] Optionally, the secondary processing specifically includes the following steps: extracting key performance data through the optimization process of real-time monitoring parameter management performance and analyzing the causes of abnormal information data while classifying abnormal information; identifying the corresponding abnormal information type based on the results of abnormal information classification and combining the abnormal information classification index; formulating corresponding processing measures based on the abnormal information type and performing visual analysis.

[0022] Optionally, the specific method for obtaining the abnormal information classification index includes: obtaining an abnormal information coefficient based on a question and answer data set and abnormal information data, the abnormal information coefficient including a data quality coefficient, a model overfitting coefficient and a parameter management performance coefficient, and obtaining an abnormal information total coefficient in combination with a reference coefficient and a reference deviation corresponding to the abnormal information coefficient, the data quality coefficient is an accuracy indicator for measuring data quality, the model overfitting coefficient is used to measure whether the pre-trained model has overfitting, and the parameter management performance coefficient is a performance indicator for measuring the pre-trained model in parameter management; obtaining an abnormal information classification index in combination with an abnormal information relative average coefficient, the abnormal information relative average coefficient including a data quality relative average coefficient, a model overfitting relative average coefficient and a parameter management performance relative average coefficient, the data quality relative average coefficient represents the ratio of the data quality average coefficient to the data quality reference coefficient, the model overfitting relative average coefficient represents the ratio of the model overfitting average coefficient to the model overfitting reference coefficient, and the parameter management performance relative average coefficient represents the ratio of the parameter management performance average coefficient to the parameter management quality reference coefficient.

[0023] The embodiment of the present invention provides an enterprise management system question-and-answer control system based on a language big model, comprising: a pre-trained model construction module, a knowledge base management module, a cache knowledge management module and a question-and-answer control module; wherein the pre-trained model construction module is used to construct a pre-trained model according to the enterprise management system question input by the user to be answered and in combination with the corresponding text data, and the pre-trained model is used to parse the input enterprise management system question in combination with a training data set, and the training data set is a collection of reference enterprise management system questions and corresponding reference answers; the knowledge base management module is used to manage and update the content of a preset management system knowledge base through the pre-trained model and in combination with set access rights to obtain a question-and-answer data set, and the question-and-answer data set represents the enterprise management system question of the user to be answered and the corresponding reference answer; the cache knowledge management module is used to perform parameter management on the first parameter information data in the pre-trained model according to the question-and-answer data set, and the first parameter information data includes parameter information data and non-parameter information data, and the parameter management represents updating, storing, querying and recording the first parameter information data; the question-and-answer control module is used to obtain first answer data in combination with the first parameter information data after parameter management, and the first answer data represents the first parameter information data visualized in the form of natural language.

[0024] The above technical solution provided by the present invention has at least the following technical effects compared with the prior art:

[0025] In the above scheme, a pre-trained model is constructed by combining the enterprise management system questions input by the question-and-answer user with the corresponding text data, and then the content of the preset management system knowledge base is managed and updated in combination with the set access rights to obtain a question and answer data set, and then parameter management is performed on the first parameter information data in the pre-trained model according to the question and answer data set, and finally the first answer data is obtained in combination with the first parameter information data after parameter management, thereby realizing the combination of user feedback and enterprise system management, and then realizing the improvement of the correlation between enterprise management system questions and answers and user feedback, effectively solving the problem of low correlation between enterprise management system questions and answers and user feedback in the prior art.

[0026] By preprocessing the collected text data of enterprise management systems and designing a model framework, the designed model framework is evaluated to obtain the model framework compliance index, and then a model training task is formulated based on the training data set and the model framework compliance index that meets the preset conditions. Finally, the model parameters are initialized with weights and a pre-trained model is constructed in combination with the training data set and the model training task, thereby achieving the precise construction of the pre-trained model and improving the accuracy of the pre-trained model construction.

[0027] By receiving the question and answer data set in real time and updating the parameter information data in the question and answer data set, and then combining the parameter update index coefficient to determine whether the non-parameter information data needs to be updated, and then retrieving the corresponding first parameter information data according to the query conditions entered by the question and answer user to optimize the parameter management performance, and finally combining the abnormal information classification index to locate the abnormal information type that occurs during the optimization process and perform secondary processing, thereby achieving more accurate processing of the abnormal information type, and then achieving more accurate optimization of parameter management performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A flow chart of a method for question-answering control of enterprise management system based on a language large model provided by an embodiment of the present invention;

[0029] Figure 2 A flowchart for constructing a pre-training model provided in an embodiment of the present invention;

[0030] Figure 3 A flowchart of obtaining a question-answering dataset provided by an embodiment of the present invention;

[0031] Figure 4 A schematic diagram of the structure of an enterprise management system question-answering control system based on a language large model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The embodiment of the present invention solves the problem of low correlation between enterprise management system questions and answers and user feedback in the prior art by providing an enterprise management system question-answering control method and system based on a language large model. The enterprise management system questions to be input by the question-answering user are combined with a set of corresponding reference answers, and a model framework with the function of learning and understanding enterprise management system text data is designed. At the same time, the designed model framework is evaluated to obtain the degree of matching between the designed model framework and the preset model framework. Then, according to the training data set and combined with the model framework compliance index that meets the preset conditions, the enterprise management mechanism questions input by the question-answering user are subjected to text classification, question-answer pair classification and sequence labeling. Then, the model parameters are initialized using weights and combined with the training data set and the model framework. The training task builds a pre-trained model, and then uses the security key to set the user to be asked questions and answers' access permissions to the contents of the preset management system knowledge base and query whether there is corresponding content in the preset management system knowledge base. If so, a question and answer data pair is generated, otherwise the preset management system knowledge base is expanded and combined with the pre-trained model to generate a question and answer data pair. At the same time, the question and answer data pairs are screened and classified to generate a question and answer data set and fed back to the user to be asked questions and answers in the form of a chart. Finally, the parameter information data and non-parameter information data in the pre-trained model are updated, stored, queried and recorded according to the question and answer data set, and the first answer data is obtained through the first parameter information data visualized in natural language form, thereby improving the correlation between enterprise management system questions and user feedback.

[0033] The technical solution in the embodiment of the present invention is to solve the problem that the correlation between the above-mentioned enterprise management system questions and answers and user feedback is low. The overall idea is as follows:

[0034] A pre-trained model is constructed based on the enterprise management system questions input by the user to be answered and the corresponding text data, and then the content of the preset management system knowledge base is managed and updated in combination with the set access rights to obtain the question and answer data set and feedback to the user to be answered in the form of a chart, and then the first parameter information data in the pre-trained model is parameter managed according to the question and answer data set, and finally the first answer data is obtained in combination with the first parameter information data after parameter management, thereby achieving the effect of improving the correlation between enterprise management system questions and answers and user feedback.

[0035] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0036] like Figure 1As shown, it is a flow chart of the enterprise management system question and answer control method based on the language big model provided by an embodiment of the present invention. The enterprise management system question and answer control method based on the language big model provided by an embodiment of the present invention includes the following steps: constructing a pre-trained model according to the enterprise management system question input by the user to be answered and in combination with the corresponding text data, the pre-trained model is used to parse the input enterprise management system question in combination with the training data set, and the training data set is a collection of reference enterprise management system questions and corresponding reference answers; managing and updating the content of the preset management system knowledge base through the pre-trained model and in combination with the set access rights to obtain a question and answer data set, the question and answer data set represents the enterprise management system question of the user to be answered and the corresponding reference answer; performing parameter management on the first parameter information data in the pre-trained model according to the question and answer data set, the first parameter information data includes parameter information data and non-parameter information data, and the parameter management represents updating, storing, querying and recording the first parameter information data; obtaining first answer data in combination with the first parameter information data after parameter management, and the first answer data represents the first parameter information data visualized in the form of natural language.

[0037] In this embodiment, in the process of building a pre-trained model, deep learning technology, such as Transformer architecture, is usually used to perform unsupervised learning on text information and pre-trained data to capture the contextual information and semantic relationship of the language; in the parameter management process, when a new question and answer data set is added, key information related to the enterprise management system can be quickly identified and extracted, and then the parameters in the pre-trained model can be adjusted and optimized in a targeted manner. In this process, parameter management strategies, such as caching mechanism and parameter optimization, are also required to ensure that the pre-trained model always maintains the latest understanding of the enterprise management system, thereby improving the performance of the pre-trained model and providing accurate and timely answers. At the same time, through the storage and query functions, the response speed of the pre-trained model and the ability to handle complex problems are also improved; advanced natural language production technology is used to convert complex parameter information and answers into fluent and easy-to-understand natural language text. In addition, the module will flexibly select the corresponding form to answer according to the specific needs and preferences of the user to be answered, so as to meet the usage needs in different scenarios, enhance the interactive experience of the user to be answered and the enterprise management system question and answer control system, and achieve an increase in the correlation between the enterprise management system question and answer and user feedback.

[0038] like Figure 2As shown, it is a flow chart for constructing a pre-training model provided by an embodiment of the present invention. The specific construction method of the pre-training model is as follows: collect text data of enterprise management system and perform data preprocessing, the data preprocessing is used to remove data anomalies; design a model framework according to the result of data preprocessing, and evaluate the designed model framework to obtain the model framework compliance index, the model framework has the function of learning and understanding the text data of enterprise management system, and the model framework compliance index indicates the degree of match between the designed model framework and the preset model framework; formulate a model training task according to the training data set and the model framework that meets the preset conditions in combination with the model framework compliance index, the model training task is to perform text classification, question and answer pair classification and sequence labeling on the enterprise management mechanism questions input by the question and answer user; use weights to initialize model parameters and construct a pre-training model in combination with the training data set and the model training task.

[0039] In this embodiment, according to the preprocessed data, it is also necessary to design a pre-training model framework in combination with the pre-training model completion index, wherein the pre-training model completion target refers to constructing a model that can learn and understand the text data of the enterprise management system through natural language processing. In the process of designing the model framework, the three main factors of the complexity of the pre-training model, the demand for computing resources and the expected performance are taken into consideration to ensure that the constructed pre-training model can meet the actual needs and has sufficient generalization ability; the model parameters are initialized by weights and combined with the training data set to construct the pre-training model, wherein the weight initialization optimizes the performance of the pre-training model by setting hyperparameters, and deep learning algorithms and techniques, such as gradient descent and directional propagation, are used to repeatedly iterate the pre-training model, thereby achieving a more accurate construction of the pre-training model.

[0040] like Figure 3 As shown, it is a flow chart for obtaining a question and answer data set provided by an embodiment of the present invention. The specific method for obtaining the question and answer data set is as follows: extract feature information by analyzing the enterprise management system questions input by the user to be answered and sort the feature information, and the information sorting means structured processing and visual classification of the feature information; set the access rights of the user to be answered to the content of the preset management system knowledge base in combination with the security key, and the security key is composed of a dynamic verification code, biometric identification and IP address restriction; query whether there is corresponding content in the preset management system knowledge base according to the result of the information sorting, if so, generate a question and answer data pair, otherwise expand the preset management system knowledge base and generate a question and answer data pair in combination with the pre-trained model, the question and answer data pair represents the corresponding answer to the enterprise management system question input by the user to be answered; screen and classify the question and answer data pair to generate a question and answer data set, and feed it back to the user to be answered in the form of a chart.

[0041] In this embodiment, according to the needs of the users to be asked questions, the organized knowledge base content is expanded, including adding new professional knowledge, case study information and industry trend information, and as the feedback results are continuously accumulated, question-answer data pairs are generated based on the expanded knowledge base content combined with the pre-trained model, and the expanded knowledge base is used as input. The model generates answers corresponding to the questions input by the users to be asked questions, ensuring that the question-answer data pairs have a high relevance, and then the question-answer data pairs are screened and classified, wherein the screening is used to remove duplicate, irrelevant and low-quality data pairs, and the classification is used to organize the data pairs according to themes, fields and dimensions. The filtered and classified data are displayed in the form of charts, including the distribution of the data set and the key information of the data set. At the same time, the pre-trained model is used to generate a high-quality question and answer data set, which not only increases the data foundation of the enterprise management system question and answer control system, but also improves the ability to handle complex problems and the accuracy of questions and answers. The question and answer data set is then continuously optimized, and the errors in the question and answer data set are repaired by collecting feedback from users waiting for questions and answers, adding new question and answer content and adjusting the way the answers are expressed, so that the question and answer control system can better meet the needs and expectations of users waiting for questions and answers, and achieve more accurate acquisition of question and answer data sets.

[0042] Optionally, the specific steps of parameter management include: receiving the question and answer data set in real time and updating the parameter information data in the question and answer data set; judging whether the non-parameter information data needs to be updated in combination with the parameter update index coefficient, and if necessary, updating and storing it, otherwise storing the question and answer data set after the data update, and the parameter update index coefficient indicates the necessity of updating the non-parameter information data; retrieving the corresponding first parameter information data according to the query conditions input by the question and answer user and optimizing the performance of parameter management; locating the abnormal information type that appears in the optimization process in combination with the abnormal information classification index and performing secondary processing, and the abnormal information classification index is used to describe the recognition accuracy of the abnormal information type that appears.

[0043] In this embodiment, by receiving the question and answer data set in real time and updating the parameter information data in the question and answer data set, this real-time update mechanism ensures that the pre-trained model can quickly adapt to the new question and answer data. At the same time, it is also responsible for storing and backing up the updated parameter information to ensure the integrity and security of the data. According to the actual needs of the user to be asked questions, the parameter update index coefficient and the updated question and answer data set, it is determined whether the non-parameter information data needs to be updated. This flexible data update mechanism ensures the accuracy and consistency of the non-parameter information data, and realizes the efficient management of the parameter information data and the non-parameter information data in the pre-trained model.

[0044] Optionally, the specific process of obtaining the first answer data is: performing secondary analysis on the enterprise management system question input by the question-and-answer user through a pre-trained model, and the secondary analysis is used to analyze and answer the keywords, syntax and semantic structures in the input enterprise management system question; according to the result of the secondary analysis, searching for the corresponding question and answer content in the preset management system knowledge base and combining the first parameter information data after parameter management to obtain the first answer data for feedback to the question-and-answer user.

[0045] In this embodiment, by deeply analyzing the keywords, syntax and semantic structure of the questions raised by the users to be answered, the pre-trained model can more accurately understand the actual needs of the users to be answered. At the same time, the secondary analysis not only focuses on the text information on the surface of the question, but also deeply explores the intentions and contextual relationships behind it, so as to ensure the accuracy of the answer. After the secondary analysis is completed, the content related to the question is quickly located through the intelligent matching algorithm, and the answer to the question is presented in the form of a chart in combination with the first parameter information data after parameter management. This process ensures the real-time nature of the natural language response and realizes more accurate acquisition of the first answer data.

[0046] Optionally, in addition to being obtained by constructing an additional evaluation model and combining the characteristic data and performance data of the current model framework, the model framework conformity index can also be obtained by the following method: obtaining the model framework structure characteristic data by analyzing the enterprise management system questions input by the question-answering user, and obtaining the model structure matching degree based on the model framework structure characteristic data, the model structure matching degree indicates the degree of fit between the enterprise management system questions input by the question-answering user and the internal structure of the model framework; analyzing the relationship between the internal structure connection nodes of the model framework and obtaining the model structure complexity in combination with the parameter weights of the model framework, the model structure complexity is used to measure the complexity of the model structure framework; evaluating the enterprise management system questions input by the question-answering user based on the training data set, and comparing them with the expected performance of the model framework to obtain the model performance evaluation error rate, the model performance evaluation error rate is used to measure the gap between the performance of the model framework in the actual application process and the expected performance; combining the reference values ​​and reference deviations of the model structure matching degree, the model structure complexity and the model performance evaluation error rate to obtain the corresponding relative deviation, and judging whether the obtained relative deviations are all less than 1, if they are all greater than 1, it indicates that the corresponding model framework does not meet the requirements, otherwise the model framework conformity index is calculated in combination with the average value of the corresponding data, and the model framework conformity index is calculated by the following formula:

[0047]

[0048] Where m is the number of the user to be answered, m = 1, 2, ...., X, X is the total number of users to be answered, mn is the number of questions and answers corresponding to the mth user to be answered, mn = m.1, m.2, ...., mY, mY is the total number of questions and answers corresponding to the mth user to be answered, CI m.n represents the model framework conformity index of the mnth question and answer corresponding to the mth user to be asked and answered, e is a natural constant, g1 represents the weight of the average relative deviation of the model structure matching degree relative to the model framework conformity index, ΔK represents the reference deviation of the model structure matching degree, K m.n represents the model structure matching degree of the mnth question and answer corresponding to the mth user to be asked and answered, K0 represents the model structure reference matching degree, g2 represents the weight of the average relative deviation of the model structure complexity relative to the model framework conformity index, ΔQ represents the model structure complexity reference deviation, Q m.n represents the model structure complexity of the mnth question and answer corresponding to the mth user to be asked and answered, Q0 represents the model structure reference complexity, g3 represents the weight of the average relative deviation of the model performance evaluation error rate relative to the model framework compliance index, ΔW represents the model performance evaluation error rate reference deviation, W m.n It represents the model performance evaluation error rate of the mnth question and answer corresponding to the mth user to be asked and answered, and W0 represents the model performance evaluation reference error rate.

[0049] In this embodiment, the matching degree and performance of the model framework can be quantitatively evaluated by calculating the model framework conformity index, thereby guiding the optimization and improvement of the model. In the actual calculation process, the calculation formula of the model framework conformity index can be used to output a loss function through a machine learning algorithm. When the loss function tends to a stable value, it indicates that the model framework is more matched with the actual needs. Otherwise, the model framework needs to be adjusted and optimized, thereby improving the accuracy and reliability of the model framework conformity index.

[0050] Optionally, the parameter update index coefficient is automatically adjusted by monitoring the performance index of the pre-trained model, specifically including: when the performance of the pre-trained model decreases or fluctuates, the parameter update mechanism can be automatically triggered, and the parameter update index coefficient is automatically adjusted according to the current performance data, the training data set and the feedback data of the user to be asked and answered. In addition to the above method, the parameter update index coefficient can also be calculated by the following formula:

[0051]

[0052] In the formula, RI m represents the parameter update index coefficient corresponding to the mth question-answering user, ΔA represents the reference deviation of the data change rate coefficient, and A mrepresents the average coefficient of data change rate after the mth user asks and answers the question, A0 represents the reference coefficient of data change rate, ΔB represents the reference deviation of configuration consistency coefficient, B m represents the average coefficient of configuration consistency of the mth user after answering the question, B0 represents the reference coefficient of configuration consistency, ΔF represents the reference deviation of the model performance improvement coefficient, and F m It represents the average coefficient of model performance improvement of the mth user after answering the question, F0 represents the reference improvement coefficient of model performance; the data change rate coefficient is used to measure the degree of change of the question and answer data set after the user completes the question and answer; the configuration consistency coefficient is used to measure the degree of match between the pre-trained model configuration and the actual operating environment; the model performance improvement coefficient is used to measure the degree of improvement of the performance of the pre-trained model after the parameters are updated compared with the performance before the parameters are updated.

[0053] In this embodiment, the number and total number of users to be answered reflect the breadth and diversity of the question-and-answer activity, the number and total number of questions and answers asked by the users to be answered reveal the activity level and user participation of the question-and-answer activity. During the question-and-answer process, with the addition of new enterprise management system questions and answers, the question-and-answer data set will continue to change. The data change rate coefficient can accurately capture this change and adjust the pre-trained model parameters in time to adapt to the new data set environment. In actual applications, the operating environment of the pre-trained model may change for various reasons, such as hardware upgrades and software updates. The configuration consistency coefficient can accurately evaluate the impact of these changes on the performance of the pre-trained model, ensure that the pre-trained model can remain stable and efficient in the actual operating environment, and achieve improved accuracy and stability of the pre-trained model.

[0054] Optionally, the secondary processing specifically includes the following steps: extracting key performance data through the optimization process of real-time monitoring parameter management performance, analyzing the causes of abnormal information data and classifying the abnormal information; identifying the corresponding abnormal information type based on the results of abnormal information classification and combining the abnormal information classification index; formulating corresponding processing measures based on the abnormal information type and performing visual analysis.

[0055] In this embodiment, real-time monitoring will be carried out at the beginning of the secondary processing. The key to this step is to extract key performance data in real time. These data can directly reflect the operating status and performance of the system. By analyzing these data, abnormal information data can be discovered in time and the reasons for their occurrence can be deeply explored. Then, the abnormal information is classified. The importance of this step lies in that by automatically classifying the abnormal data, the type and nature of the abnormal information can be more clearly understood. This classification not only helps managers to better understand the operating status of the enterprise management mechanism question and answer control system, but also provides strong support for the subsequent formulation of treatment measures, so as to more accurately locate the problem. On the basis of the classification of abnormal information, positioning will be performed according to the classification results and the abnormal information classification index. The key to this step is that by comprehensively using the classification results and the classification index, the specific cause of the abnormality or the specific location of the abnormality can be accurately identified. This positioning capability enables the enterprise management mechanism question and answer control system to find the root cause of the problem more accurately, provides a clearer direction for the subsequent formulation of treatment measures, and realizes the efficiency of the feedback results of the question and answer users.

[0056] Optionally, a specific method for obtaining the abnormal information classification index includes: obtaining an abnormal information coefficient based on a question-and-answer data set and abnormal information data, the abnormal information coefficient including a data quality coefficient, a model overfitting coefficient and a parameter management performance coefficient, and obtaining an abnormal information total coefficient in combination with a reference coefficient and a reference deviation corresponding to the abnormal information coefficient, the data quality coefficient is an accuracy indicator for measuring data quality, the model overfitting coefficient is used to measure whether the pre-trained model has overfitting, and the parameter management performance coefficient is a performance indicator for measuring the pre-trained model in parameter management; obtaining the abnormal information classification index in combination with the abnormal information relative average coefficient, the abnormal information relative average coefficient including a data quality relative average coefficient, a model overfitting relative average coefficient and a parameter management performance relative average coefficient, the data quality relative average coefficient represents the ratio of the data quality average coefficient to the data quality reference coefficient, the model overfitting relative average coefficient represents the ratio of the model overfitting average coefficient to the model overfitting reference coefficient, and the parameter management performance relative average coefficient represents the ratio of the parameter management performance average coefficient to the parameter management quality reference coefficient.

[0057] In this embodiment, the total data quality coefficient represents the ratio of the absolute value of the difference between the data quality coefficient and the data quality reference coefficient to the reference deviation of the data quality coefficient, the total model overfitting coefficient represents the ratio of the absolute value of the difference between the model overfitting coefficient and the model overfitting reference coefficient, and the total parameter management performance coefficient represents the ratio of the absolute value of the difference between the parameter management performance coefficient and the parameter management performance reference coefficient to the reference deviation of the parameter management performance coefficient; since the classification and determination of abnormal information often have a certain fuzziness, the fuzzy evaluation method in fuzzy numbers can be used for calculation: first determine the fuzzy membership of each type of abnormal information, and then perform fuzzy synthesis based on the membership and the corresponding weight coefficient to obtain the abnormal information classification index; wherein, the fuzzy membership refers to the degree to which an element belongs to a fuzzy set, usually represented by a value between 0 and 1, wherein 0 means not belonging at all, 1 means belonging completely, and the value between 0 and 1 represents the degree of belonging. In addition to obtaining the abnormal information classification index by the fuzzy evaluation method, it can also be calculated by the following formula:

[0058]

[0059] In the formula, HI m represents the abnormal information classification index corresponding to the mth user to be asked questions, Δα represents the reference deviation of the pre-training data quality coefficient, α m.n represents the pre-training data quality coefficient of the mnth question and answer corresponding to the mth user to be asked and answered, α0 represents the pre-training data quality reference coefficient, α” represents the pre-training data quality average coefficient, Δβ represents the pre-training model overfitting coefficient reference deviation, β m.n represents the overfitting coefficient of the pre-trained model corresponding to the mnth question and answer of the mth user to be asked and answered, β0 represents the overfitting reference coefficient of the pre-trained model, β” represents the average overfitting coefficient of the pre-trained model, Δλ represents the reference deviation of the parameter management performance coefficient, λ m.n represents the parameter management performance coefficient of the mnth question and answer corresponding to the mth user to be answered, λ0 represents the parameter management performance reference coefficient, and λ” represents the average parameter management performance coefficient.

[0060] like Figure 4As shown, it is a structural schematic diagram of the enterprise management system question-and-answer control system based on the language big model provided by an embodiment of the present invention. The enterprise management system question-and-answer control system based on the language big model provided by an embodiment of the present invention includes: a pre-trained model construction module, a knowledge base management module, a cache knowledge management module and a question-and-answer control module; wherein the pre-trained model construction module is used to construct a pre-trained model according to the enterprise management system questions input by the user to be answered and in combination with the corresponding text data, and the pre-trained model is used to parse the input enterprise management system questions in combination with the training data set, and the training data set is a collection of reference enterprise management system questions and corresponding reference answers; the knowledge base management module is used to manage and update the content of the preset management system knowledge base through the pre-trained model and in combination with the set access rights to obtain a question-and-answer data set, and the question-and-answer data set represents the enterprise management system questions of the user to be answered and the corresponding reference answers; the cache knowledge management module is used to perform parameter management on the first parameter information data in the pre-trained model according to the question-and-answer data set, and the first parameter information data includes parameter information data and non-parameter information data, and parameter management represents updating, storing, querying and recording the first parameter information data; the question-and-answer control module is used to obtain first answer data in combination with the first parameter information data after parameter management, and the first answer data represents the first parameter information data visualized in the form of natural language.

[0061] In this embodiment, in the enterprise management question and answer control system based on the language big model, there is a close connection and collaborative relationship between the various modules, which jointly realize the complete process from the enterprise management system questions input by the user to the generation of answers. Among them, the question and answer control module is the output end of the system, and it also has the ability of intelligent error correction and adaptive learning. When encountering uncertain or ambiguous questions, the module can automatically adjust and optimize according to the question and answer data set and the feedback results of the users to be answered, thereby improving the availability and user satisfaction of the enterprise management question and answer control system, and achieving the accuracy and reliability of the feedback from the users to be answered.

[0062] The technical solution in the above-mentioned embodiment of the present invention has at least the following technical effects or advantages: relative to the open domain natural language reasoning question answering system and method driven by a large language model disclosed in the patent application with publication number: CN116932708A, the embodiment of the present invention performs data preprocessing on the collected text data of the enterprise management system and designs a model framework, and at the same time evaluates the designed model framework to obtain the model framework compliance index, and then formulates a model training task according to the training data set and the model framework that meets the preset conditions in combination with the model framework compliance index, and finally uses the weights to initialize the model parameters and constructs a pre-trained model in combination with the training data set and the model training task, thereby realizing the precise construction of the pre-trained model, and then realizing the pre-trained model. The accuracy of model construction is improved; relative to a university scientific research management question and answer system combining a knowledge graph and a large language model disclosed in a patent application with publication number: CN117609436A, the embodiment of the present invention receives a question and answer data set in real time and updates the parameter information data in the question and answer data set, and then determines whether the non-parameter information data needs to be updated in combination with the parameter update index coefficient, and then retrieves the corresponding first parameter information data according to the query condition input by the question and answer user to optimize the performance of parameter management, and finally locates the abnormal information type that appears in the optimization process in combination with the abnormal information classification index and performs secondary processing, thereby achieving more accurate processing of the abnormal information type, and then achieving more accurate optimization of the parameter management performance.

[0063] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0064] The present invention is described with reference to flowcharts and / or block diagrams of systems, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0065] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0066] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0067] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0068] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for question-answering control of enterprise management system based on language big model, characterized in that: The following steps are involved: A pre-training model is constructed based on the enterprise management system questions input by the user to be answered and in combination with the corresponding text data. The pre-training model is used to parse the input enterprise management system questions in combination with the training data set. The training data set is a collection of reference enterprise management system questions and corresponding reference answers; The content of the preset management system knowledge base is managed and updated through the pre-trained model and in combination with the set access rights to obtain a question and answer dataset, wherein the question and answer dataset represents the enterprise management system questions to be asked and answered by the user and the corresponding reference answers; Performing parameter management on first parameter information data in the pre-trained model according to the question-answering data set, where the first parameter information data includes parameter information data and non-parameter information data, and the parameter management represents updating, storing, querying and recording the first parameter information data; The first answer data is obtained in combination with the first parameter information data after parameter management, and the first answer data represents the first parameter information data visualized in a natural language form.

2. The enterprise management system question-answering control method based on the language big model as claimed in claim 1 is characterized in that: The specific construction method of the pre-training model is as follows: Collecting text data of enterprise management systems and performing data preprocessing, wherein the data preprocessing is used to remove data anomalies; Design a model framework according to the result of data preprocessing, and evaluate the designed model framework to obtain a model framework compliance index, wherein the model framework has the function of learning and understanding the text data of the enterprise management system, and the model framework compliance index indicates the degree of matching between the designed model framework and the preset model framework; Formulate a model training task based on the training data set and in combination with a model framework that meets the preset conditions in the model framework compliance index, wherein the model training task is to perform text classification, question-answer pair classification, and sequence labeling on the enterprise management mechanism questions input by the question-answering user; Initialize the model parameters using weights and build a pre-trained model by combining the training dataset and model training tasks.

3. The enterprise management system question-answering control method based on the language big model as claimed in claim 1 is characterized in that: The specific method for obtaining the question-answering dataset is as follows: By analyzing the enterprise management system questions input by the question-answering user, feature information is extracted and the feature information is sorted, wherein the feature information is structured and visually classified; The permission for the user to be answered to access the content of the preset management system knowledge base is set in combination with the security key, which is composed of a dynamic verification code, biometric identification and IP address restriction; According to the result of information collation, it is queried whether there is corresponding content in the preset management system knowledge base. If so, a question-answer data pair is generated. Otherwise, the preset management system knowledge base is expanded and combined with the pre-trained model to generate a question-answer data pair, wherein the question-answer data pair represents the answer corresponding to the enterprise management system question to be input by the question-answering user; The question-answer data pairs are screened and classified to generate question-answer data sets, which are then fed back to the users in the form of charts.

4. The enterprise management system question-answering control method based on language big model as claimed in claim 2 is characterized in that: The specific steps of parameter management include: Receive the question and answer data set in real time and update the parameter information data in the question and answer data set; Determine whether the non-parameter information data needs to be updated in combination with the parameter update index coefficient. If necessary, update and store it. Otherwise, store the updated question and answer data set. The parameter update index coefficient indicates the necessity of updating the non-parameter information data. Retrieving the corresponding first parameter information data according to the query condition input by the question-answering user and optimizing the performance of parameter management; The abnormal information type appearing in the optimization process is located and processed again in combination with the abnormal information classification index, where the abnormal information classification index is used to describe the recognition accuracy of the abnormal information type appearing.

5. The enterprise management system question-answering control method based on language big model as claimed in claim 1 is characterized in that: The specific process of obtaining the first answer data is as follows: The pre-trained model is used to perform secondary analysis on the enterprise management system questions input by the question-answering user, and the secondary analysis is used to analyze and answer the keywords, syntax and semantic structures in the input enterprise management system questions; According to the result of the secondary analysis, the corresponding question and answer content is searched in the preset management system knowledge base and the first parameter information data after parameter management is combined to obtain the first answer data to be fed back to the user to be answered.

6. The enterprise management system question-answering control method based on language big model as claimed in claim 4 is characterized in that: The specific method for obtaining the model framework compliance index is as follows: The model framework structure characteristic data is obtained by analyzing the enterprise management system questions input by the question-answering user, and the model structure matching degree is obtained according to the model framework structure characteristic data, wherein the model structure matching degree indicates the degree of fit between the enterprise management system questions input by the question-answering user and the internal structure of the model framework; Analyze the relationship between the internal structural connection nodes of the model framework and obtain the model structure complexity in combination with the parameter weights of the model framework. The model structure complexity is used to measure the complexity of the model structure framework. Evaluate the enterprise management system questions input by the question-answering user based on the training data set, and compare them with the expected performance of the model framework to obtain the model performance evaluation error rate, which is used to measure the gap between the performance of the model framework in actual application and the expected performance; The corresponding relative deviation is obtained by combining the reference values ​​and reference deviations of the model structure matching degree, model structure complexity and model performance evaluation error rate, and it is determined whether the obtained relative deviations are all less than 1. If they are all greater than 1, it indicates that the corresponding model framework does not meet the requirements. Otherwise, the model framework compliance index is calculated by combining the average value of the corresponding data. The model framework compliance index is calculated by the following formula: Where m is the number of the user to be answered, m = 1, 2, ...., X, X is the total number of users to be answered, mn is the number of questions and answers corresponding to the mth user to be answered, mn = m.1, m.2, ...., mY, mY is the total number of questions and answers corresponding to the mth user to be answered, CI m.n represents the model framework conformity index of the mnth question and answer corresponding to the mth user to be asked and answered, e is a natural constant, g1 represents the weight of the average relative deviation of the model structure matching degree relative to the model framework conformity index, ΔK represents the reference deviation of the model structure matching degree, K m.n represents the model structure matching degree of the mnth question and answer corresponding to the mth user to be asked and answered, K0 represents the model structure reference matching degree, g2 represents the weight of the average relative deviation of the model structure complexity relative to the model framework conformity index, ΔQ represents the model structure complexity reference deviation, Q m.n represents the model structure complexity of the mnth question and answer corresponding to the mth user to be asked and answered, Q0 represents the model structure reference complexity, g3 represents the weight of the average relative deviation of the model performance evaluation error rate relative to the model framework compliance index, ΔW represents the model performance evaluation error rate reference deviation, W m.n It represents the model performance evaluation error rate of the mnth question and answer corresponding to the mth user to be asked and answered, and W0 represents the model performance evaluation reference error rate.

7. The enterprise management system question-answering control method based on language big model as claimed in claim 4 is characterized in that: The parameter update index coefficient is calculated by the following formula: In the formula, RI m represents the parameter update index coefficient corresponding to the mth user to be answered, ΔA represents the reference deviation of the data change rate coefficient, and A m represents the average coefficient of data change rate after the mth question-answering user asks and answers the question, A0 represents the reference coefficient of data change rate, ΔB represents the reference deviation of configuration consistency coefficient, B m represents the average coefficient of configuration consistency of the mth user after answering the question, B0 represents the reference coefficient of configuration consistency, ΔF represents the reference deviation of the model performance improvement coefficient, and F m represents the average coefficient of model performance improvement after answering the question for the mth user, and F0 represents the reference improvement coefficient of model performance; The data change rate coefficient is used to measure the degree of change in the question-answering data set after the question-answering user completes the question-answering; The configuration consistency coefficient is used to measure the matching degree between the pre-trained model configuration and the actual operating environment; The model performance improvement coefficient is a measure of the degree of improvement in the performance of the pre-trained model after the parameters are updated compared to before the parameters were updated.

8. The enterprise management system question-answering control method based on language big model as claimed in claim 4 is characterized in that: The secondary processing specifically comprises the following steps: Through the optimization process of real-time monitoring parameter management performance, key performance data is extracted and the causes of abnormal information data are analyzed while classifying abnormal information; According to the abnormal information classification result and in combination with the abnormal information classification index, the corresponding abnormal information type is identified; Formulate corresponding processing measures according to the type of abnormal information and perform visual analysis.

9. The enterprise management system question-answering control method based on language big model as claimed in claim 8 is characterized in that: The specific method for obtaining the abnormal information classification index includes: Obtain an abnormal information coefficient according to the question-answer data set and the abnormal information data, wherein the abnormal information coefficient includes a data quality coefficient, a model overfitting coefficient and a parameter management performance coefficient, and obtain the abnormal information total coefficient in combination with a reference coefficient and a reference deviation corresponding to the abnormal information coefficient, wherein the data quality coefficient is an accuracy index for measuring data quality, the model overfitting coefficient is used to measure whether the pre-trained model has an overfitting phenomenon, and the parameter management performance coefficient is a performance index for measuring the pre-trained model in terms of parameter management; An abnormal information classification index is obtained in combination with the relative average coefficient of abnormal information. The relative average coefficient of abnormal information includes the relative average coefficient of data quality, the relative average coefficient of model overfitting and the relative average coefficient of parameter management performance. The relative average coefficient of data quality represents the ratio of the data quality average coefficient to the data quality reference coefficient, the relative average coefficient of model overfitting represents the ratio of the model overfitting average coefficient to the model overfitting reference coefficient, and the relative average coefficient of parameter management performance represents the ratio of the parameter management performance average coefficient to the parameter management quality reference coefficient.

10. A question-answering control system for enterprise management system based on a large language model, characterized in that: It includes a pre-training model building module, a knowledge base management module, a cache knowledge management module and a question-answering control module; The pre-training model building module is used to build a pre-training model based on the enterprise management system questions input by the user to be answered and in combination with the corresponding text data. The pre-training model is used to parse the input enterprise management system questions in combination with the training data set. The training data set is a collection of reference enterprise management system questions and corresponding reference answers. The knowledge base management module is used to manage and update the content of the preset management system knowledge base through the pre-trained model and in combination with the set access rights to obtain a question and answer dataset, wherein the question and answer dataset represents the enterprise management system questions to be asked and answered by the user and the corresponding reference answers; The cache knowledge management module is used to perform parameter management on the first parameter information data in the pre-trained model according to the question-answer data set, wherein the first parameter information data includes parameter information data and non-parameter information data, and the parameter management represents updating, storing, querying and recording the first parameter information data; The question and answer control module is used to obtain first answer data in combination with the first parameter information data after parameter management, and the first answer data represents the first parameter information data visually presented in the form of natural language.

Images