Device for automatically supplementing knowledge sustainable learning through semantic understanding based on initial knowledge base

Abstract

The invention provides a device for automatically supplementing knowledge sustainable learning through semantic understanding on the basis of an initial knowledge base, an AI reasoning technology is utilized, extraction of data in an unstructured file is completed, a new learning task is executed according to logic needed by a user, and therefore power needed by the user is continuously expanded for the knowledge base; according to the method, a hierarchical association relationship is formed between the supplemented files in the knowledge base and the original text, and during question answering, connection processing among multiple files can be effectively carried out, so that the connection relationship among the file knowledge bases is greatly improved; on the basis of convergence of application systems continuously produced by users, content is continuously obtained along with production of the application systems of the users and sent to the system, and an ecological environment is provided for sustainable learning.

Description

Technical Field

[0001] The present invention relates to the field of semantic analysis, and in particular to a perpetual learning device that automatically supplements knowledge through semantic understanding based on an initial knowledge base. Background Art

[0002] With the rapid development of natural language processing and artificial intelligence technologies, pre-trained large-scale language models have made breakthrough progress in tasks such as language understanding, question and answer, dialogue, and text generation. A knowledge base is a structured, easy-to-operate, easy-to-use, and fully organized knowledge cluster in knowledge engineering. It is a collection of interconnected knowledge pieces that are stored, organized, managed, and used in computer memory using some (or several) knowledge representation methods to solve problems in a certain (or certain) field.

[0003] In actual use, the content of the knowledge base is limited, and the content accessed to the knowledge base is limited, so it cannot be connected to the business system; it is also impossible to control the scope of content imported into the knowledge base model, and it is impossible to continuously supplement the content, which affects the user experience. Summary of the invention

[0004] The purpose of the present invention is to provide a perpetual learning device that automatically supplements knowledge through semantic understanding based on an initial knowledge base to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a continuous learning device for automatically supplementing knowledge through semantic understanding based on an initial knowledge base, characterized in that it includes a file management module, a knowledge rule management module, a semantic understanding module, a task scheduling module, a knowledge expansion module, a knowledge approval module, a knowledge storage module, and a system display module. The file management module imports a small number of initial or known files into the model and identifies information such as the source of the files; The knowledge rule management module is used to set task rules corresponding to entities and elements, where the entities and elements are inferred from existing documents in the document management module; The semantic understanding module performs semantic analysis on the knowledge files in the model, obtains the content of knowledge reasoning, and performs structured conversion on the reasoning content to obtain corresponding entity or element information; The task scheduling module performs similarity matching between the matched entity attribute information and the rules in the knowledge rule management module, and calls the corresponding rule task; The knowledge expansion module processes the information obtained from the task scheduling module in different ways and performs repeated learning as required when expanding the knowledge base; The knowledge approval module uniformly summarizes and approves the documents formed after the acquired information is converted; The knowledge storage module manages and stores the acquired files in a hierarchical manner according to the process of reasoning and expansion; The system display module is used to display the content and results of each module.

[0006] Preferably, the file management module imports the initial file into the question-answering model, and the initial file import method includes docking with the business system or manually uploading the file; when importing the file, the file needs to be synchronized for classification and processing; various types of files are converted into OFD files or other file formats; the processing methods include OCR recognition, audio and video text recognition, various waveform recognition, and document content batch supplement processing; the module stores the file source in the database, and summarizes the file source of each entry into the database by level to facilitate the display module to call and query.

[0007] Preferably, the knowledge rule management module is used to set the rules for the user to obtain information through the device, and the rules are expressed as elements, which can be a word, a sentence, a paragraph, or a text.

[0008] Preferably, the semantic understanding module performs semantic analysis on the existing files in the knowledge base element by element according to the knowledge base management rules. The semantic understanding module pushes the obtained results to the task scheduling module. These results are output in xml and json formats, and their accuracy is verified synchronously. If there are errors in the output results, the reasoning and question-answering are repeated N times, and the threshold is adjusted to re-semantically understand the original files. Finally, by parsing the correct xml and json, data in the form of Key and value is obtained. The key is used to display the attributes of the entity, and the value represents the value corresponding to the attribute. The rule elements that fail to obtain results are pushed to the task scheduling module with 0 or empty values.

[0009] Preferably, the task scheduling module refers to the user configuring the corresponding task tools element by element while setting the knowledge base management rules. These task tools include interfaces, data sharing and exchange tools, and rpa data robots; when scheduling tasks, different tasks are called according to different task types; the task scheduling module supports multi-threaded concurrent acquisition, and if the value passed is 0 or empty, the task is not executed; the file data reports, text files, and knowledge content obtained after these tasks are pushed to the knowledge base expansion module.

[0010] Preferably, the knowledge expansion module processes the information obtained in task scheduling in different ways according to different types. When the acquired information is data, parse the information into unstructured content and write the converted data into a file. When converting, the json and XML self-describing data formats are used to describe and display the content in the form of Key and Value; When the acquired information is a file, the file is processed according to different rules for different file formats. If it is a scanned document, OCR is performed on the file to generate a double-layer file with (text); When the acquired information is a link, the acquired HTML link is automatically saved as a local HTML file; The knowledge expansion module imports the above-mentioned files into the file library and marks the specific sources of the files; When expanding knowledge, you can set the number of questions and answers for the same question. When the set number is reached, the question body of multiple questions and answers is obtained, and the file content is automatically supplemented to supplement the knowledge base.

[0011] Preferably, the knowledge approval module is the management module of the device, which not only supports setting different storage relationships for supplementary file information, setting automatic addition to the knowledge base, or setting simultaneous addition to the knowledge base after review, but also has the identity information, digital certificate, authority management, filtering strategy, and encryption method functions required for the operation of the device.

[0012] Preferably, when the knowledge storage module aggregates the files, it sets the association relationship between the files written by different source reasoning as a source tree relationship to automatically supplement the file library.

[0013] Preferably, the system display module is a display of the contents and results of each module of the device, and the displayed files support hierarchical display according to the reasoning process of the files, and the displayed files indicate the source reference for traceability verification.

[0014] A continuous learning device for automatically supplementing knowledge through semantic understanding based on an initial knowledge base is characterized by comprising the following steps: Step S1: Import the initialization or known file into the model, and the knowledge rule management module extracts the entity information or relationship information from the file and marks the corresponding information content on the file; Step S2: import the files of the knowledge base into the model and perform vector analysis on the files; Step S3: The task prefabrication scheduling module performs similarity matching between the entity attribute information matched in the file question and answer module and the rules in the knowledge rule management module, and processes the content in the task scheduling and the acquired information in different ways; Step S4: hierarchically manage and store the acquired files according to the process of reasoning and expansion, and uniformly summarize the files formed after the acquired information is converted; Step S5: Mark the question and answer content with various file source references. The various contents of the inference display can be traced back to the original text by clicking the answer content. When tracing back, the files involved are displayed in the form of a bound volume and summarized in a guided manner. Switch and click on each source to jump to read. Among them, if the file is a scan or a picture, jump to the OFD file after OCR for accurate matching and tracing. When there is a hierarchical relationship between files, it supports multi-layer display of traceable content.

[0015] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention creates a machine learning logic and generates a device for continuous learning; (2) The present invention makes full use of AI reasoning technology to complete the extraction of data from unstructured files and execute new learning tasks according to the logic required by users, thereby continuously expanding the knowledge base to provide the power required by users; (3) The present invention is based on the convergence of application systems that are continuously produced by users. As the user's application systems produce continuously acquired content, it is sent to the system, providing an ecological environment for sustainable learning.

[0016] The present invention forms a hierarchical association relationship between the files supplemented in the knowledge base and the original text, and can effectively perform connection processing between multiple files during question and answer, greatly improving the connection relationship between file knowledge bases. DETAILED DESCRIPTION

[0017] The technical solution of the present invention will be clearly and completely described below through embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Example

[0018] The perpetual learning device for automatically supplementing knowledge through semantic understanding based on the initial knowledge base is composed of a file management module, a knowledge rule management module, a semantic understanding module, a task scheduling module, a knowledge expansion module, a knowledge approval module, a knowledge storage module, and a system display module. The working steps of the perpetual learning device for automatically supplementing knowledge through semantic understanding based on the initial knowledge base are as follows: Step S1: The file management module imports a small number of initialized or known files into the question-answering model, and identifies information such as the source of the files. The initial file import method includes connecting to the business system or manually uploading files. When importing files, it is necessary to synchronize the files for classification and processing, and convert various types of files into OFD files or other file formats. The processing methods include OCR recognition, audio and video text recognition, various waveform recognition, and document content batch supplement processing. The file management module stores the file source in the database, and summarizes the source of each file into the database by level to facilitate the display module to call and query.

[0019] Step S2: The knowledge rule management module sets the rules for the user to obtain information through the device. The rules are expressed as elements, which can be a word, a sentence, a paragraph, or a text. The entity information or relationship information is extracted from the file, and the corresponding information content is marked on the file; Although the user knows that the existing knowledge base cannot satisfy the user's desired information temporarily when setting the required information elements, the user can gradually achieve the desired information by continuously supplementing the knowledge base through the device to expand the content of the knowledge base.

[0020] The knowledge base rule management module can add, delete, modify and check rule elements to meet the user's ever-changing information needs, and ultimately obtain the information the user needs by continuously increasing and expanding the knowledge base content.

[0021] The task rule content corresponding to entities and elements, each entity attribute or element supports the setting of one or more docking models or corresponding user-controlled application systems. For example: using RPA to capture data, full-text retrieval, AI retrieval, and interface docking to obtain the desired knowledge content to generate text files, form a knowledge source group related to the knowledge base, and continuously obtain knowledge through the operation of related application systems.

[0022] Step S3: The semantic understanding module performs semantic analysis on the existing files in the knowledge base element by element according to the knowledge base management rules. The semantic understanding module pushes the results that have been obtained to the task scheduling module. These results are output in xml and json formats, and their accuracy is verified synchronously. If there are errors in the output results, the module re-reasons and asks questions N times, adjusts the threshold to re-semantically understand the original files, and finally obtains data in the form of Key and value by parsing the correct xml and json. The key is used to display the attributes of the entity, and the value represents the value corresponding to the attribute. The rule elements that fail to obtain results are pushed to the task scheduling module with 0 or empty values.

[0023] Step S4: The task scheduling module configures the corresponding task tools element by element while setting the knowledge base management rules. These task tools include interfaces, data sharing and exchange tools, and RPA data robots; When scheduling tasks, different tasks are called according to different task types. The task scheduling module supports multi-threaded concurrent acquisition. If the value passed is 0 or empty, the task will not be executed. The file data reports, text files, and knowledge content obtained after these tasks will be pushed to the knowledge base expansion module.

[0024] Step S5: The knowledge expansion module processes the information obtained from task scheduling in different ways according to different types. When the acquired information is data, parse the information into unstructured content and write the converted data into a file. When converting, the json and XML self-describing data formats are used to describe and display the content in the form of Key and Value; When the acquired information is a file, the file is processed according to different rules for different file formats. If it is a scanned document, OCR is performed on the file to generate a double-layer file with (text); When the acquired information is a link, the acquired HTML link is automatically saved as a local HTML file; The knowledge expansion module adds the above-mentioned newly added data reports, text files, and knowledge content to the knowledge base, and marks the specific source of the file. The knowledge base expansion module should set regular learning tasks. If there is an emergency, the learning task can be manually triggered to enable the newly stored knowledge content to be vectorized in time. After the module completes the vectorization, it will automatically trigger the semantic understanding module. The semantic understanding module generates new learning results and pushes the results to the task scheduling module again. The task scheduling module pushes the newly added data reports, text files, and knowledge content to the knowledge expansion module. The knowledge base expansion module completes the vectorization of the newly stored knowledge content in time, triggers the semantic understanding module again, and repeats the above steps continuously, so that the device realizes the function of automatically supplementing knowledge through semantic understanding for continuous learning until the user's purpose of obtaining the required information is met; When expanding knowledge, you can set the number of questions and answers for the same question. When the set number is reached, the question body of multiple questions and answers is obtained, and the file content is automatically supplemented to supplement the knowledge base.

[0025] Step S6: The knowledge approval module is the management module of this device. It not only supports setting different storage relationships for supplementary file information, setting automatic addition to the knowledge base, or setting it to be added to the knowledge base synchronously after review, but also has the identity information, digital certificate, authority management, filtering strategy, and encryption method functions required for the operation of the device.

[0026] Step S7: When the knowledge storage module aggregates the files, it sets the association relationship between the files written by different sources through reasoning as a source tree relationship, and automatically supplements the file library.

[0027] Step S8: The system display module displays the content and results of each module. The displayed files support hierarchical display according to the reasoning process of the files. The displayed files indicate the source reference for traceability verification.

[0028] When displaying, the contents of the knowledge base of the file management module can be classified, divided into different libraries, and displayed and counted in different formats. The file sources in the file management module are displayed at the required level. The knowledge rule management module can display the existing rule elements and conduct statistical queries by category life cycle; The semantic understanding module can display the inferred information in various analysis methods such as list, timeline, comparison, relationship analysis, parallel condition analysis, etc., and trace the source of each inference of the semantic understanding module. This information is the main query module for users to obtain work goals. Users can obtain the operation results of the entire device through this to achieve their work goals. The task scheduling module displays the usage and efficiency of the scheduling module according to various rules; The knowledge expansion module conducts statistical analysis to query new results and display the expanded content; The knowledge approval module displays approval logs, function usage, function updates, and system operation logs; The above-mentioned display module supports various displays and outputs of various charts, mind maps, streaming files, layout files, layout data stream files, and html dynamic files.

[0029] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be considered exemplary and non-restrictive in all respects, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims be included in the present invention.

Claims

1. A continuous learning device for automatically supplementing knowledge through semantic understanding based on an initial knowledge base, characterized by: It includes file management module, knowledge rule management module, semantic understanding module, task scheduling module, knowledge expansion module, knowledge approval module, knowledge storage module and system display module. The file management module imports a small number of initial or known files into the model and identifies information such as the source of the files; The knowledge rule management module is used to set task rules corresponding to entities and elements, where the entities and elements are inferred from existing documents in the document management module; The semantic understanding module performs semantic analysis on the knowledge files in the model, obtains the content of knowledge reasoning, and performs structured conversion on the reasoning content to obtain corresponding entity or element information; The task scheduling module performs similarity matching between the matched entity attribute information and the rules in the knowledge rule management module, and calls the corresponding rule task; The knowledge expansion module processes the information obtained from the task scheduling module in different ways and performs repeated learning as required when expanding the knowledge base; The knowledge approval module uniformly summarizes and approves the documents formed after the acquired information is converted; The knowledge storage module manages and stores the acquired files in a hierarchical manner according to the process of reasoning and expansion; The system display module is used to display the content and results of each module.

2. The device for continuous learning based on the initial knowledge base and automatically supplementing knowledge through semantic understanding according to claim 1 is characterized by: The file management module imports the initial file into the question-answering model. The initial file import method includes connecting to the business system or manually uploading the file; when importing the file, the file needs to be synchronized for classification and processing; various types of files are converted into OFD files or other file formats; the processing methods include OCR recognition, audio and video text recognition, various waveform recognition, and document content batch supplement processing; the module stores the file source in the database, and summarizes the file source of each entry into the database by level to facilitate the display module to call and query.

3. The device for continuous learning based on the initial knowledge base and automatically supplementing knowledge through semantic understanding according to claim 1 is characterized in that: The knowledge rule management module is used to set the rules that the user needs to obtain information through the device. The rule is expressed as an element, which can be a word, a sentence, a paragraph, or a text.

4. The device for continuous learning based on the initial knowledge base and automatically supplementing knowledge through semantic understanding according to claim 1 is characterized in that: The semantic understanding module performs semantic analysis on the existing files in the knowledge base element by element according to the knowledge base management rules. The semantic understanding module pushes the results that have been obtained to the task scheduling module. These results are output in xml and json formats, and their accuracy is verified synchronously. If there are errors in the output results, the reasoning and question-answering are repeated N times, and the threshold is adjusted to re-semantically understand the original files. Finally, by parsing the correct xml and json, data in the form of Key and value is obtained. The key is used to display the attributes of the entity, and the value represents the value corresponding to the attribute. The rule elements that fail to obtain results are pushed to the task scheduling module with 0 or empty values.

5. The device for continuous learning based on the initial knowledge base and automatically supplementing knowledge through semantic understanding according to claim 1 is characterized in that: The task scheduling module refers to the user configuring the corresponding task tools element by element while setting the knowledge base management rules. These task tools include interfaces, data sharing and exchange tools, and RPA data robots. When scheduling tasks, different tasks are called according to different task types. The task scheduling module supports multi-threaded concurrent acquisition. If the value passed is 0 or empty, the task will not be executed. The file data reports, text files, and knowledge content obtained after these tasks will be pushed to the knowledge base expansion module.

6. The device for continuous learning based on the initial knowledge base and automatically supplementing knowledge through semantic understanding according to claim 1, characterized in that: The knowledge expansion module processes the information obtained from task scheduling in different ways according to different types. When the acquired information is data, parse the information into unstructured content and write the converted data into a file. When converting, the json and XML self-describing data formats are used to describe and display the content in the form of Key and Value; When the acquired information is a file, the file is processed according to different rules for different file formats. If it is a scanned document, OCR is performed on the file to generate a double-layer file with text; When the acquired information is a link, the acquired HTML link is automatically saved as a local HTML file; The knowledge expansion module imports the above-mentioned files into the file library and marks the specific sources of the files; When expanding knowledge, you can set the number of questions and answers for the same question. When the set number is reached, the question body of multiple questions and answers is obtained, and the file content is automatically supplemented to supplement the knowledge base.

7. The device for continuous learning based on the initial knowledge base and automatically supplementing knowledge through semantic understanding according to claim 1 is characterized in that: The knowledge approval module is the management module of this device. It not only supports setting different storage relationships for supplementary file information, setting automatic addition to the knowledge base, or setting it to be added to the knowledge base synchronously after review, but also has the identity information, digital certificate, authority management, filtering strategy, and encryption method functions required for the operation of the device.

8. The device for continuous learning based on the initial knowledge base and automatically supplementing knowledge through semantic understanding according to claim 1 is characterized by: When the knowledge storage module aggregates the files, it sets the association relationship between the files written by different source reasoning as a source tree relationship to automatically supplement the file library.

9. The device for continuous learning based on the initial knowledge base and automatically supplementing knowledge through semantic understanding according to claim 1, characterized in that: The system display module is a display of the contents and results of each module of the device. The displayed files support hierarchical display according to the reasoning process of the files. The displayed files are marked with source references for traceability and verification.

10. A continuous learning device for automatically supplementing knowledge through semantic understanding based on an initial knowledge base, characterized in that: The following steps are involved: Step S1: Import the initialization or known file into the model, and the knowledge rule management module extracts the entity information or relationship information from the file and marks the corresponding information content on the file; Step S2: import the files of the knowledge base into the model and perform vector analysis on the files; Step S3: The task prefabrication scheduling module performs similarity matching between the entity attribute information matched in the file question and answer module and the rules in the knowledge rule management module, and processes the content in the task scheduling and the acquired information in different ways; Step S4: hierarchically manage and store the acquired files according to the process of reasoning and expansion, and uniformly summarize the files formed after the acquired information is converted; Step S5: Mark the question and answer content with various file source references. The various contents of the inference display can be traced back to the original text by clicking the answer content. When tracing back, the files involved are displayed in the form of a bound volume and summarized in a guided manner. Switch and click on each source to jump to read. Among them, if the file is a scan or a picture, jump to the OFD file after OCR for accurate matching and tracing. When there is a hierarchical relationship between files, it supports multi-layer display of traceable content.