A method and system for archival knowledge management based on large language model

Abstract

The present invention provides an archival knowledge management method and system based on a large language model in the field of archival knowledge management technology. The method comprises: step S1, creating an archival knowledge base, an archival knowledge extraction rule base and a question template base; step S2, acquiring historical archival files to construct a data set; step S3, creating a large language model and training it; step S4, identifying the archival files to be managed by OCR technology to obtain the archival content to be managed, and extracting the corresponding archival knowledge extraction rules and question templates; step S5, inputting the generated knowledge extraction questions and the archival content to be managed into the large language model to obtain archival knowledge; step S6, storing the archival knowledge in a neo4j graphical knowledge base and a mysql relational knowledge base; step S7, acquiring knowledge triples from the graphical knowledge base and expressing them in the form of a knowledge graph; step S8, managing the addition, deletion, update and query of the archival knowledge. The advantages of the present invention are: improving the efficiency and quality of archival knowledge management.

Description

Technical Field

[0001] The present invention relates to the technical field of archive knowledge management, and in particular to an archive knowledge management method and system based on a large language model. Background Art

[0002] Archival knowledge management is the management of archive maintenance, archiving, and review. In the current field of archival knowledge management technology, archives are scattered, disorganized, of varying quality, large in quantity, and isolated. Industry pain points such as information islands, low retrieval efficiency, and difficulty in deep mining and utilization of knowledge are becoming increasingly prominent.

[0003] Regarding the management of archival knowledge, some methods are managed through manual maintenance, manual review and release, and the knowledge is manually associated with archives, which has the disadvantages of huge maintenance workload and the inability to update archival knowledge in a timely manner; some methods build an archival knowledge base and extract archival knowledge through word segmentation technology, but if archival knowledge is extracted only through word segmentation technology, the scope of archival knowledge will be limited to the word segmentation base, and the correlation between archival knowledge cannot be obtained, and thus the value of archival knowledge cannot be fully utilized. In other words, there are certain limitations in the current archival knowledge management.

[0004] Therefore, how to provide an archival knowledge management method and system based on a large language model to improve the efficiency and quality of archival knowledge management has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide an archive knowledge management method based on a large language model, comprising the following steps:

[0006] Step S1, creating an archive knowledge base, an archive knowledge extraction rule base and a question template base;

[0007] Step S2, obtaining a large number of historical archive files, identifying the historical archive content of each of the historical archive files by using OCR technology, and constructing a data set after marking the content of each of the historical archive files;

[0008] Step S3: creating a large language model for extracting archival knowledge, and training the large language model using the data set;

[0009] Step S4, obtaining the archive file to be managed, identifying the archive file to be managed by OCR technology to obtain the archive content to be managed, and matching the corresponding archive knowledge extraction rules and question templates from the archive knowledge extraction rule library and the question template library based on the field of the archive file to be managed;

[0010] Step S5: Generate a knowledge extraction question based on the matched archival knowledge extraction rule and question template, input the knowledge extraction question and the archival content to be managed into a large language model, obtain archival knowledge including knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information, and store the archival knowledge in an archival knowledge base;

[0011] Step S6, writing the knowledge entities, entity relationships, and knowledge attributes into a pre-created neo4j graphical knowledge base in the form of knowledge triples, and writing the knowledge ancillary information into a pre-created mysql relational knowledge base;

[0012] Step S7: Based on the neo4j graphical knowledge base and the mysql relational knowledge base, the knowledge entities, entity relationships, knowledge attributes or knowledge ancillary information in the archival knowledge are added, deleted, updated and queried.

[0013] Furthermore, in step S1, the archive knowledge base is used to classify and store the archive knowledge by field;

[0014] The archival knowledge extraction rule base is used to store archival knowledge extraction rules in different fields;

[0015] The question template library is used to store question templates in different fields, and the question templates are used to generate knowledge extraction questions according to archive knowledge extraction rules;

[0016] The step S2 is specifically as follows:

[0017] A large number of historical archive files are obtained, and the historical archive content of each of the historical archive files is identified through OCR technology. Knowledge extraction problems and archival knowledge annotation are performed on each of the historical archive contents, and a data set is constructed based on the annotated historical archive contents.

[0018] Furthermore, the step S3 is specifically as follows:

[0019] Creating a large language model for extracting archival knowledge, dividing the data set into a training set, a validation set, and a test set based on a ratio of 7:2:1;

[0020] The large language model is trained by the training set until a preset convergence condition is met, and the hyperparameters of the large language model including at least the number of iterations, the learning rate, and the training batch are continuously optimized during the training process; the trained large language model is verified by the verification set to determine whether the extraction accuracy of the large language model is greater than a preset accuracy threshold. If not, the verification fails, and the training set is expanded to continue training; if yes, the verification passes, and:

[0021] The large language model is tested by the test set to determine whether the confidence of the large language model is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training; if yes, the test passes and the training ends.

[0022] Furthermore, in step S5, the knowledge attachment information at least includes archive description information and source archive ID;

[0023] In step S6, there exists a mapping relationship between knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information between the neo4j graphical knowledge base and the mysql relational knowledge base; when the neo4j graphical knowledge base or the mysql relational knowledge base is updated, the mysql relational knowledge base or the neo4j graphical knowledge base is updated synchronously.

[0024] Furthermore, the step S7 is specifically as follows:

[0025] The input archive management instruction is obtained, and the archive management instruction is parsed to obtain archive management items and search conditions. Based on the search conditions, the archive knowledge including knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information is matched from the neo4j graphical knowledge base and the mysql relational knowledge base. Based on the archive management items, the archive knowledge is added, deleted, updated and queried, and the archive knowledge is displayed in the form of a knowledge graph.

[0026] In a second aspect, the present invention provides an archive knowledge management system based on a large language model, comprising:

[0027] A library creation module is used to create an archive knowledge library, an archive knowledge extraction rule library and a question template library;

[0028] A data set construction module is used to obtain a large number of historical archive files, identify the historical archive content of each of the historical archive files through OCR technology, and construct a data set after marking the historical archive content;

[0029] A large language model creation module, used to create a large language model for extracting archive knowledge, and train the large language model using the data set;

[0030] An archive file identification module is used to obtain archive files to be managed, identify the archive files to be managed by OCR technology to obtain archive content to be managed, and match corresponding archive knowledge extraction rules and question templates from an archive knowledge extraction rule library and a question template library based on the fields of the archive files to be managed;

[0031] An archive knowledge extraction module is used to generate a knowledge extraction question based on the matched archive knowledge extraction rules and question templates, input the knowledge extraction question and the archive content to be managed into a large language model, obtain archive knowledge including knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information, and store the archive knowledge in an archive knowledge base;

[0032] The archive knowledge storage module is used to write the knowledge entities, entity relationships, and knowledge attributes into a pre-created neo4j graphical knowledge base in the form of knowledge triples, and write the knowledge ancillary information into a pre-created mysql relational knowledge base;

[0033] The archive knowledge management module is used to add, delete, update and query the knowledge entities, entity relationships, knowledge attributes or knowledge ancillary information in the archive knowledge based on the neo4j graphical knowledge base and the mysql relational knowledge base.

[0034] Furthermore, in the library creation module, the archive knowledge base is used to classify and store archive knowledge by field;

[0035] The archival knowledge extraction rule base is used to store archival knowledge extraction rules in different fields;

[0036] The question template library is used to store question templates in different fields, and the question templates are used to generate knowledge extraction questions according to archive knowledge extraction rules;

[0037] The data set construction module is specifically used for:

[0038] A large number of historical archive files are obtained, and the historical archive content of each of the historical archive files is identified through OCR technology. Knowledge extraction problems and archival knowledge annotation are performed on each of the historical archive contents, and a data set is constructed based on the annotated historical archive contents.

[0039] Furthermore, the large language model creation module is specifically used for:

[0040] Creating a large language model for extracting archival knowledge, dividing the data set into a training set, a validation set, and a test set based on a ratio of 7:2:1;

[0041] The large language model is trained by the training set until a preset convergence condition is met, and the hyperparameters of the large language model including at least the number of iterations, the learning rate, and the training batch are continuously optimized during the training process; the trained large language model is verified by the verification set to determine whether the extraction accuracy of the large language model is greater than a preset accuracy threshold. If not, the verification fails, and the training set is expanded to continue training; if yes, the verification passes, and:

[0042] The large language model is tested by the test set to determine whether the confidence of the large language model is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training; if yes, the test passes and the training ends.

[0043] Furthermore, in the archive knowledge extraction module, the knowledge attachment information at least includes archive description information and source archive ID;

[0044] In the archive knowledge storage module, the neo4j graphical knowledge base and the mysql relational knowledge base have a mapping relationship between knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information; when the neo4j graphical knowledge base or the mysql relational knowledge base is updated, the mysql relational knowledge base or the neo4j graphical knowledge base is updated synchronously.

[0045] Furthermore, the archive knowledge management module is specifically used for:

[0046] The input archive management instruction is obtained, and the archive management instruction is parsed to obtain archive management items and search conditions. Based on the search conditions, the archive knowledge including knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information is matched from the neo4j graphical knowledge base and the mysql relational knowledge base. Based on the archive management items, the archive knowledge is added, deleted, updated and queried, and the archive knowledge is displayed in the form of a knowledge graph.

[0047] The advantages of the present invention are:

[0048] 1. By creating an archive knowledge base, an archive knowledge extraction rule base, and a question template base, a large amount of historical archive content is identified through OCR technology to construct a data set, and the created large language model is trained through the constructed data set; then the archive content to be managed is identified through OCR technology, and the archive knowledge extraction rules and question templates are obtained from the archive knowledge extraction rule base and the question template base respectively based on the field of the archive files to be managed, and the knowledge extraction questions generated based on the archive knowledge extraction rules and question templates and the archive content to be managed are input into the large language model to obtain the archive knowledge; then the archive knowledge is stored in the neo4j graphical knowledge base and the mysql relational knowledge base, and the archive knowledge is added, deleted, updated, and queried based on the neo4j graphical knowledge base and the mysql relational knowledge base; that is, the archive knowledge is automatically extracted by combining OCR technology and the large language model without manual operation, and the archive knowledge is stored through the neo4j graphical knowledge base and the mysql relational knowledge base. Combined with the knowledge graph, the relationship between the archive knowledge can be quickly found later, and the value of the archive knowledge is fully utilized, which ultimately greatly improves the efficiency and quality of archive knowledge management.

[0049] 2. By setting up an archival knowledge base, archival knowledge is classified and stored by field. The archival knowledge extraction rule base is used to store archival knowledge extraction rules in different fields, and the question template base is used to store question templates in different fields. Through the classified storage of various archival files, archival knowledge extraction rules, and question templates, archival files can be managed by field, which helps to improve the pertinence and professionalism of archival management. Building and maintaining independent knowledge extraction rule sets and question rule bases will help the large language model understand the specific application scenarios of rules in various fields, effectively improve the professionalism and processing efficiency of archival knowledge management, and more accurately meet the characteristics and needs of file management in various fields.

[0050] 3. By continuously optimizing the number of iterations, learning rate, and training batch hyperparameters during the training of the large language model, verifying the trained large language model through the validation set to determine whether the extraction accuracy of the large language model is higher than the preset threshold, and testing the large language model through the test set to determine whether the confidence of the large language model is greater than the preset confidence threshold, multiple training, testing, and verification are performed, which greatly improves the accuracy of the large language model in extracting archival knowledge.

[0051] 4. By regularly updating and maintaining the archival knowledge base, the archival knowledge in the MySQL relational knowledge base and the neo4j graphical knowledge base can be synchronized. When the archival knowledge in the MySQL relational knowledge base or the neo4j graphical knowledge base is modified, the modified archival knowledge will be synchronously updated to the neo4j graphical knowledge base or the MySQL relational knowledge base, which is conducive to improving the maintainability of the archival knowledge and ensuring the timeliness of archival knowledge search.

[0052] 5. Through the search conditions carried by the archive management instructions, the archival knowledge including knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information is matched from the neo4j graphical knowledge base and the mysql relational knowledge base. At the same time, the archival knowledge is managed by adding, deleting, updating and querying. Finally, the relevant archival knowledge is displayed through the knowledge graph, so that users can intuitively understand the relationship and structure between archival knowledge, which is convenient for users to find archival knowledge, analyze the relationship between archival knowledge, and quickly obtain valuable information. BRIEF DESCRIPTION OF THE DRAWINGS

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

[0054] Figure 1 The present invention is a flowchart of an archive knowledge management method based on a large language model.

[0055] Figure 2 The present invention is a schematic diagram of the structure of an archive knowledge management system based on a large language model. DETAILED DESCRIPTION

[0056] The technical solution in the embodiments of the present application has the following overall idea: combining OCR technology and a large language model to automatically extract archival knowledge without manual operation, storing archival knowledge through a neo4j graphical knowledge base and a mysql relational knowledge base, and combining with a knowledge graph to quickly find the association between archival knowledge in the future, giving full play to the value of archival knowledge and improving the efficiency and quality of archival knowledge management.

[0057] Please refer to Figure 1 to Figure 2 As shown, a preferred embodiment of the archival knowledge management method based on a large language model of the present invention comprises the following steps:

[0058] Step S1, creating an archive knowledge base, an archive knowledge extraction rule base and a question template base;

[0059] Step S2, obtaining a large number of historical archive files, identifying the historical archive content of each of the historical archive files by using OCR technology, and constructing a data set after marking the content of each of the historical archive files;

[0060] Step S3, creating a large language model for extracting archival knowledge, and training the large language model using the data set; the trained large language model supports analyzing the relationship between archival knowledge based on context information, and identifying and extracting the relationship information between archival knowledge;

[0061] Step S4, obtaining the archive file to be managed, identifying the archive file to be managed by OCR technology to obtain the archive content to be managed, and matching the corresponding archive knowledge extraction rules and question templates from the archive knowledge extraction rule library and the question template library based on the field of the archive file to be managed;

[0062] Step S5: Generate a knowledge extraction question based on the matched archive knowledge extraction rule and question template, input the knowledge extraction question and the archive content to be managed into the large language model, obtain archive knowledge including knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information, and store the archive knowledge in the archive knowledge base; the knowledge extraction question is generated by replacing variables in the question template;

[0063] Step S6, writing the knowledge entities, entity relationships, and knowledge attributes into a pre-created neo4j graphical knowledge base in the form of knowledge triples, and writing the knowledge ancillary information into a pre-created mysql relational knowledge base for storing different types of archival knowledge;

[0064] Step S7: Based on the neo4j graphical knowledge base and the mysql relational knowledge base, the knowledge entities, entity relationships, knowledge attributes or knowledge ancillary information in the archival knowledge are added, deleted, updated and queried.

[0065] In the step S1, the archive knowledge base is used to classify and store archive knowledge by field;

[0066] The archival knowledge extraction rule base is used to store archival knowledge extraction rules in different fields. The archival knowledge extraction rule base extracts archival knowledge including at least knowledge entities, entity relationships and knowledge attributes according to different fields of archival knowledge, and is used to build and maintain knowledge extraction rule collections in different fields; field management can build and maintain independent knowledge extraction rule sets according to different fields, which helps to improve the pertinence and professionalism of knowledge extraction, so that the knowledge extraction rules in each field can more accurately reflect the characteristics and needs of the field, that is, clearly and accurately define the knowledge ontology and the relationship between knowledge ontologies, guide the extraction process of archival knowledge, and ensure that the extracted archival knowledge is both rich and accurate; the knowledge extraction rule base also provides examples of knowledge named entities that are compatible with field prompt words, which helps the large language model to understand the specific application scenarios of rules in each field, and more flexibly deal with the knowledge extraction of archives in different fields;

[0067] The question template library is used to store question templates in different fields, and the question templates are used to generate knowledge extraction questions according to archive knowledge extraction rules; by storing the question templates in different fields, the large language model can adapt to the professional knowledge extraction in different fields, thereby improving professionalism and processing efficiency;

[0068] The step S2 is specifically as follows:

[0069] A large number of historical archive files are obtained, and the historical archive content of each of the historical archive files is identified through OCR technology. Knowledge extraction problems and archival knowledge annotation are performed on each of the historical archive contents, and a data set is constructed based on the annotated historical archive contents.

[0070] The step S3 is specifically as follows:

[0071] Creating a large language model for extracting archival knowledge, dividing the data set into a training set, a validation set, and a test set based on a ratio of 7:2:1;

[0072] The large language model is trained by the training set until a preset convergence condition is met, and the hyperparameters of the large language model including at least the number of iterations, the learning rate, and the training batch are continuously optimized during the training process; the trained large language model is verified by the verification set to determine whether the extraction accuracy of the large language model is greater than a preset accuracy threshold. If not, the verification fails, and the training set is expanded to continue training; if yes, the verification passes, and:

[0073] The large language model is tested by the test set to determine whether the confidence of the large language model is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training; if yes, the test passes and the training ends.

[0074] In the step S5, the knowledge attachment information includes at least file description information and source file ID;

[0075] In step S6, there is a mapping relationship between knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information between the neo4j graphical knowledge base and the mysql relational knowledge base; when the neo4j graphical knowledge base or the mysql relational knowledge base is updated, the mysql relational knowledge base or the neo4j graphical knowledge base is updated synchronously, and the archival knowledge in the mysql relational knowledge base is modified. The modified archival knowledge will be automatically converted into a Cypherr script and synchronously updated to the neo4j graphical knowledge base.

[0076] The step S7 is specifically as follows:

[0077] The input archive management instruction is obtained, and the archive management instruction is parsed to obtain archive management items and search conditions. Based on the search conditions, the archive knowledge including knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information is matched from the neo4j graphical knowledge base and the mysql relational knowledge base. Based on the archive management items, the archive knowledge is added, deleted, updated and queried, and the archive knowledge is displayed in the form of a knowledge graph.

[0078] When querying, the corresponding knowledge triples are obtained from the neo4j graphical knowledge base through visualization technology, and the knowledge triples are displayed in the form of knowledge graphs. The archive source ID is extracted from the mysql relational knowledge base, and the archive file corresponding to the archive source ID is displayed. In specific implementation, fuzzy search, screening and filtering functions are supported, allowing users to enter partial keywords to filter out the required archive knowledge.

[0079] A preferred embodiment of the archive knowledge management system based on a large language model of the present invention includes the following modules:

[0080] A library creation module is used to create an archive knowledge library, an archive knowledge extraction rule library and a question template library;

[0081] A data set construction module is used to obtain a large number of historical archive files, identify the historical archive content of each of the historical archive files through OCR technology, and construct a data set after marking the historical archive content;

[0082] A large language model creation module is used to create a large language model for extracting archival knowledge, and train the large language model through the data set; the trained large language model supports analyzing the relationship between archival knowledge according to context information, and identifies and extracts the relationship information between archival knowledge;

[0083] An archive file identification module is used to obtain archive files to be managed, identify the archive files to be managed by OCR technology to obtain archive content to be managed, and match corresponding archive knowledge extraction rules and question templates from an archive knowledge extraction rule library and a question template library based on the fields of the archive files to be managed;

[0084] The archive knowledge extraction module is used to generate a knowledge extraction question based on the matched archive knowledge extraction rules and question template, input the knowledge extraction question and the archive content to be managed into the large language model, obtain archive knowledge including knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information, and store the archive knowledge in the archive knowledge base; the knowledge extraction question is generated by replacing variables in the question template;

[0085] The archival knowledge storage module is used to write the knowledge entities, entity relationships, and knowledge attributes into a pre-created neo4j graphical knowledge base in the form of knowledge triples, and write the knowledge ancillary information into a pre-created mysql relational knowledge base, which is used to store different types of archival knowledge;

[0086] The archive knowledge management module is used to add, delete, update and query the knowledge entities, entity relationships, knowledge attributes or knowledge ancillary information in the archive knowledge based on the neo4j graphical knowledge base and the mysql relational knowledge base.

[0087] In the library creation module, the archive knowledge base is used to classify and store archive knowledge by field;

[0088] The archival knowledge extraction rule base is used to store archival knowledge extraction rules in different fields. The archival knowledge extraction rule base extracts archival knowledge including at least knowledge entities, entity relationships and knowledge attributes according to different fields of archival knowledge, and is used to build and maintain knowledge extraction rule collections in different fields; field management can build and maintain independent knowledge extraction rule sets according to different fields, which helps to improve the pertinence and professionalism of knowledge extraction, so that the knowledge extraction rules in each field can more accurately reflect the characteristics and needs of the field, that is, clearly and accurately define the knowledge ontology and the relationship between knowledge ontologies, guide the extraction process of archival knowledge, and ensure that the extracted archival knowledge is both rich and accurate; the knowledge extraction rule base also provides examples of knowledge named entities that are compatible with field prompt words, which helps the large language model to understand the specific application scenarios of rules in each field, and more flexibly deal with the knowledge extraction of archives in different fields;

[0089] The question template library is used to store question templates in different fields, identify key information of archival knowledge from archival knowledge, and extract the main content of archival knowledge. The question template is used to generate knowledge extraction questions according to archival knowledge extraction rules. By storing the question templates in different fields, the large language model can adapt to the extraction of professional knowledge in different fields, thereby improving professionalism and processing efficiency.

[0090] The data set construction module is specifically used for:

[0091] A large number of historical archive files are obtained, and the historical archive content of each of the historical archive files is identified through OCR technology. Knowledge extraction problems and archival knowledge annotation are performed on each of the historical archive contents, and a data set is constructed based on the annotated historical archive contents.

[0092] The large language model creation module is specifically used for:

[0093] Creating a large language model for extracting archival knowledge, dividing the data set into a training set, a validation set, and a test set based on a ratio of 7:2:1;

[0094] The large language model is trained by the training set until a preset convergence condition is met, and the hyperparameters of the large language model including at least the number of iterations, the learning rate, and the training batch are continuously optimized during the training process; the trained large language model is verified by the verification set to determine whether the extraction accuracy of the large language model is greater than a preset accuracy threshold. If not, the verification fails, and the training set is expanded to continue training; if yes, the verification passes, and:

[0095] The large language model is tested by the test set to determine whether the confidence of the large language model is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training; if yes, the test passes and the training ends.

[0096] In the archive knowledge extraction module, the knowledge attachment information includes at least archive description information and source archive ID;

[0097] In the archival knowledge storage module, the neo4j graphical knowledge base and the mysql relational knowledge base have a mapping relationship between knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information; when the neo4j graphical knowledge base or the mysql relational knowledge base is updated, the mysql relational knowledge base or the neo4j graphical knowledge base is updated synchronously, and the archival knowledge in the mysql relational knowledge base is modified. The modified archival knowledge will be automatically converted into a Cypherr script and synchronously updated to the neo4j graphical knowledge base.

[0098] The archive knowledge management module is specifically used for:

[0099] The input archive management instruction is obtained, and the archive management instruction is parsed to obtain archive management items and search conditions. Based on the search conditions, the archive knowledge including knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information is matched from the neo4j graphical knowledge base and the mysql relational knowledge base. Based on the archive management items, the archive knowledge is added, deleted, updated and queried, and the archive knowledge is displayed in the form of a knowledge graph.

[0100] When querying, the corresponding knowledge triples are obtained from the neo4j graphical knowledge base through visualization technology, and the knowledge triples are displayed in the form of knowledge graphs. The archive source ID is extracted from the mysql relational knowledge base, and the archive file corresponding to the archive source ID is displayed. In specific implementation, fuzzy search, screening and filtering functions are supported, allowing users to enter partial keywords to filter out the required archive knowledge.

[0101] In summary, the advantages of the present invention are:

[0102] 1. By creating an archive knowledge base, an archive knowledge extraction rule base, and a question template base, a large amount of historical archive content is identified through OCR technology to construct a data set, and the created large language model is trained through the constructed data set; then the archive content to be managed is identified through OCR technology, and the archive knowledge extraction rules and question templates are obtained from the archive knowledge extraction rule base and the question template base respectively based on the field of the archive files to be managed, and the knowledge extraction questions generated based on the archive knowledge extraction rules and question templates and the archive content to be managed are input into the large language model to obtain the archive knowledge; then the archive knowledge is stored in the neo4j graphical knowledge base and the mysql relational knowledge base, and the archive knowledge is added, deleted, updated, and queried based on the neo4j graphical knowledge base and the mysql relational knowledge base; that is, the archive knowledge is automatically extracted by combining OCR technology and the large language model without manual operation, and the archive knowledge is stored through the neo4j graphical knowledge base and the mysql relational knowledge base. Combined with the knowledge graph, the relationship between the archive knowledge can be quickly found later, and the value of the archive knowledge is fully utilized, which ultimately greatly improves the efficiency and quality of archive knowledge management.

[0103] 2. By setting up an archival knowledge base, archival knowledge is classified and stored by field. The archival knowledge extraction rule base is used to store archival knowledge extraction rules in different fields, and the question template base is used to store question templates in different fields. Through the classified storage of various archival files, archival knowledge extraction rules, and question templates, archival files can be managed by field, which helps to improve the pertinence and professionalism of archival management. Building and maintaining independent knowledge extraction rule sets and question rule bases will help the large language model understand the specific application scenarios of rules in various fields, effectively improve the professionalism and processing efficiency of archival knowledge management, and more accurately meet the characteristics and needs of file management in various fields.

[0104] 3. By continuously optimizing the number of iterations, learning rate, and training batch hyperparameters during the training of the large language model, verifying the trained large language model through the validation set to determine whether the extraction accuracy of the large language model is higher than the preset threshold, and testing the large language model through the test set to determine whether the confidence of the large language model is greater than the preset confidence threshold, multiple training, testing, and verification are performed, which greatly improves the accuracy of the large language model in extracting archival knowledge.

[0105] 4. By regularly updating and maintaining the archival knowledge base, the archival knowledge in the MySQL relational knowledge base and the neo4j graphical knowledge base can be synchronized. When the archival knowledge in the MySQL relational knowledge base or the neo4j graphical knowledge base is modified, the modified archival knowledge will be synchronously updated to the neo4j graphical knowledge base or the MySQL relational knowledge base, which is conducive to improving the maintainability of the archival knowledge and ensuring the timeliness of archival knowledge search.

[0106] 5. Through the search conditions carried by the archive management instructions, the archival knowledge including knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information is matched from the neo4j graphical knowledge base and the mysql relational knowledge base. At the same time, the archival knowledge is managed by adding, deleting, updating and querying. Finally, the relevant archival knowledge is displayed through the knowledge graph, so that users can intuitively understand the relationship and structure between archival knowledge, which is convenient for users to find archival knowledge, analyze the relationship between archival knowledge, and quickly obtain valuable information.

[0107] Although the specific implementation modes of the present invention are described above, those skilled in the art should understand that the specific implementation modes described are only illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A method for archival knowledge management based on a large language model, characterized in that: The steps include: Step S1, creating an archive knowledge base, an archive knowledge extraction rule base and a question template base; Step S2, obtaining a large number of historical archive files, identifying the historical archive content of each of the historical archive files by using OCR technology, performing knowledge extraction problems and archival knowledge annotation on each of the historical archive contents, and constructing a data set based on the annotated historical archive contents; Step S3: creating a large language model for extracting archival knowledge, and training the large language model using the data set; Step S4, obtaining the archive file to be managed, identifying the archive file to be managed by OCR technology to obtain the archive content to be managed, and matching the corresponding archive knowledge extraction rules and question templates from the archive knowledge extraction rule library and the question template library based on the field of the archive file to be managed; Step S5: Generate a knowledge extraction question based on the matched archival knowledge extraction rule and question template, input the knowledge extraction question and the archival content to be managed into a large language model, obtain archival knowledge including knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information, and store the archival knowledge in an archival knowledge base; Step S6, writing the knowledge entities, entity relationships, and knowledge attributes into a pre-created neo4j graphical knowledge base in the form of knowledge triples, and writing the knowledge ancillary information into a pre-created mysql relational knowledge base; Step S7: Based on the neo4j graphical knowledge base and the mysql relational knowledge base, add, delete, update and query the knowledge entities, entity relationships, knowledge attributes or knowledge ancillary information in the archive knowledge; The step S3 is specifically as follows: Creating a large language model for extracting archival knowledge, dividing the data set into a training set, a validation set, and a test set based on a ratio of 7:2:1; The large language model is trained by the training set until a preset convergence condition is met, and the hyperparameters of the large language model including at least the number of iterations, the learning rate, and the training batch are continuously optimized during the training process; the trained large language model is verified by the verification set to determine whether the extraction accuracy of the large language model is greater than a preset accuracy threshold. If not, the verification fails, and the training set is expanded to continue training; if yes, the verification passes, and: The large language model is tested by the test set to determine whether the confidence of the large language model is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training; if yes, the test passes and the training ends.

2. The archival knowledge management method based on a large language model according to claim 1, characterized in that: In the step S1, the archive knowledge base is used to classify and store archive knowledge by field; The archival knowledge extraction rule base is used to store archival knowledge extraction rules in different fields; The question template library is used to store question templates in different fields, and the question templates are used to generate knowledge extraction questions according to archive knowledge extraction rules.

3. The archival knowledge management method based on a large language model according to claim 1 is characterized in that: In the step S5, the knowledge attachment information includes at least file description information and source file ID; In step S6, there exists a mapping relationship between knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information between the neo4j graphical knowledge base and the mysql relational knowledge base; when the neo4j graphical knowledge base or the mysql relational knowledge base is updated, the mysql relational knowledge base or the neo4j graphical knowledge base is updated synchronously.

4. The archival knowledge management method based on a large language model according to claim 1 is characterized in that: The step S7 is specifically as follows: The input archive management instruction is obtained, and the archive management instruction is parsed to obtain archive management items and search conditions. Based on the search conditions, the archive knowledge including knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information is matched from the neo4j graphical knowledge base and the mysql relational knowledge base. Based on the archive management items, the archive knowledge is added, deleted, updated and queried, and the archive knowledge is displayed in the form of a knowledge graph.

5. An archive knowledge management system based on a large language model, characterized by: Includes the following modules: A library creation module is used to create an archive knowledge library, an archive knowledge extraction rule library and a question template library; A data set construction module is used to obtain a large number of historical archive files, identify the historical archive content of each of the historical archive files through OCR technology, perform knowledge extraction problems and annotate archival knowledge on the content of each of the historical archive files, and construct a data set based on the annotated content of each of the historical archive files; A large language model creation module, used to create a large language model for extracting archive knowledge, and train the large language model using the data set; An archive file identification module is used to obtain archive files to be managed, identify the archive files to be managed by OCR technology to obtain archive content to be managed, and match corresponding archive knowledge extraction rules and question templates from an archive knowledge extraction rule library and a question template library based on the fields of the archive files to be managed; An archive knowledge extraction module is used to generate a knowledge extraction question based on the matched archive knowledge extraction rules and question templates, input the knowledge extraction question and the archive content to be managed into a large language model, obtain archive knowledge including knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information, and store the archive knowledge in an archive knowledge base; The archive knowledge storage module is used to write the knowledge entities, entity relationships, and knowledge attributes into a pre-created neo4j graphical knowledge base in the form of knowledge triples, and write the knowledge ancillary information into a pre-created mysql relational knowledge base; The archive knowledge management module is used to manage the addition, deletion, update and query of knowledge entities, entity relationships, knowledge attributes or knowledge ancillary information in the archive knowledge based on the neo4j graphical knowledge base and the mysql relational knowledge base; The large language model creation module is specifically used for: Creating a large language model for extracting archival knowledge, dividing the data set into a training set, a validation set, and a test set based on a ratio of 7:2:1; The large language model is trained by the training set until a preset convergence condition is met, and the hyperparameters of the large language model including at least the number of iterations, the learning rate, and the training batch are continuously optimized during the training process; the trained large language model is verified by the verification set to determine whether the extraction accuracy of the large language model is greater than a preset accuracy threshold. If not, the verification fails, and the training set is expanded to continue training; if yes, the verification passes, and: The large language model is tested by the test set to determine whether the confidence of the large language model is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training; if yes, the test passes and the training ends.

6. The archive knowledge management system based on a large language model according to claim 5, characterized in that: In the library creation module, the archive knowledge base is used to classify and store archive knowledge by field; The archival knowledge extraction rule base is used to store archival knowledge extraction rules in different fields; The question template library is used to store question templates in different fields, and the question templates are used to generate knowledge extraction questions according to archive knowledge extraction rules.

7. The archive knowledge management system based on a large language model according to claim 5, characterized in that: In the archive knowledge extraction module, the knowledge attachment information includes at least archive description information and source archive ID; In the archive knowledge storage module, the neo4j graphical knowledge base and the mysql relational knowledge base have a mapping relationship between knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information; when the neo4j graphical knowledge base or the mysql relational knowledge base is updated, the mysql relational knowledge base or the neo4j graphical knowledge base is updated synchronously.

8. The archive knowledge management system based on a large language model according to claim 5, characterized in that: The archive knowledge management module is specifically used for: The input archive management instruction is obtained, and the archive management instruction is parsed to obtain archive management items and search conditions. Based on the search conditions, the archive knowledge including knowledge entities, entity relationships, knowledge attributes and knowledge ancillary information is matched from the neo4j graphical knowledge base and the mysql relational knowledge base. Based on the archive management items, the archive knowledge is added, deleted, updated and queried, and the archive knowledge is displayed in the form of a knowledge graph.

Images