Information Extraction Processing Method, Device, Medium and Product

Abstract

This application relates to the field of data processing, and provides an information extraction and processing method, device, medium and product. The method includes: after receiving the retrieval information input by the user, determining the corresponding knowledge type to be retrieved according to the retrieval information, where the knowledge type includes one or more of responsibilities, permissions, processes, and specifications; determining the knowledge results corresponding to the knowledge type to be retrieved in the preset knowledge result library, and outputting the retrieval results to the user according to the knowledge results, where the knowledge results include one or more of knowledge entities, knowledge graphs, knowledge documents, knowledge indexes, and knowledge change history lists. The method of this application improves the accuracy and efficiency of retrieval in the banking and financial systems.

Description

Technical Field

[0001] This application relates to the field of data processing, and particularly to an information extraction processing method, device, medium and product. Background Art

[0002] With the development of Internet technology and the progress of society, data has grown explosively, and the processing of data has become increasingly important. For example, how to accurately obtain the content retrieved by users from a large amount of data.

[0003] In the prior art, generally, the content retrieved by users is extracted from a large amount of data by means of knowledge extraction. The existing knowledge extraction methods are mainly the following two: One is to first perform entity extraction on data based on natural language processing (NLP), and then perform operations such as entity alignment and entity linking on the extracted entities, and construct a knowledge graph according to the relationships between entities. The other is to construct a statistical language model based on deep learning, and then use the model to extract features in the data, and establish relationships or links between the features to obtain the final knowledge. The existing knowledge extraction methods are all general and popular, more reflected in the direction of entity extraction and relationship extraction, and after completing knowledge extraction, a list of relevant documents will be output as the retrieval result to the user. However, on the one hand, there is a large amount of unstructured data in the bank financial system, and due to the unique business fields and application scenarios, these data have data characteristics quite different from other unstructured data. Using the existing knowledge extraction methods to process the data has low accuracy, resulting in low retrieval accuracy in the bank financial system. On the other hand, outputting a list of documents as the retrieval result to the user, the user needs to click on the documents one by one to search, the operation is cumbersome, and the retrieval efficiency is low.

[0004] Therefore, a data processing solution that can improve the retrieval accuracy and efficiency of the bank financial system is needed. Summary of the Invention

[0005] This application provides an information extraction processing method, device, medium and product, which are used to solve the problem of low accuracy and efficiency when the bank financial system uses the existing information extraction processing method for retrieval.

[0006] In the first aspect, this application provides an information extraction processing method, including:

[0007] After receiving the retrieval information input by the user, determining the corresponding knowledge type to be retrieved according to the retrieval information, where the knowledge type includes one or more of business, responsibility, authority, process, and specification;

[0008] Determine the knowledge results corresponding to the knowledge type to be retrieved in the preset knowledge result library, and output the retrieval results to the user according to the knowledge results. The knowledge results include one or more of knowledge entities, knowledge graphs, knowledge documents, knowledge indexes, and knowledge change history lists;

[0009] The knowledge result library is obtained by the following method:

[0010] Determine the correlation degree between the documents to be extracted, and construct a document relationship graph according to the documents to be extracted and the correlation degree between the documents to be extracted;

[0011] Determine the sub-graph to be analyzed in the document relationship graph;

[0012] Extract knowledge from the documents in the sub-graph according to the knowledge type to obtain the knowledge results corresponding to the knowledge type;

[0013] Establish the knowledge result library according to the knowledge results corresponding to the knowledge type.

[0014] In a possible implementation manner, the determination of the correlation degree between the documents to be extracted specifically includes:

[0015] Determine the document type corresponding to each document to be extracted, and determine the document type distribution vector corresponding to each document to be extracted according to the document type. The document types include one or more of system management, training management, business management, operation operation, and business operation;

[0016] Determine the correlation degree between the documents to be extracted according to the document type distribution vector corresponding to each document to be extracted.

[0017] In a possible implementation manner, the determination of the document type corresponding to each document to be extracted, and the determination of the document type distribution vector corresponding to each document to be extracted according to the document type specifically includes:

[0018] Determine the type description word corresponding to each document to be extracted according to the preset type dictionary; determine the document type corresponding to each document to be extracted according to the corresponding relationship between the preset type description word and the document type; perform normalization processing on the document type to determine the document type distribution vector corresponding to each document to be extracted;

[0019] Or,

[0020] Classify each of the to-be-extracted documents using multiple binary classifiers to obtain the classification probability value of each to-be-extracted document; determine the document type of each to-be-extracted document according to the type corresponding to the maximum classification probability value, and determine the document type distribution vector corresponding to each to-be-extracted document according to the classification probability value of each to-be-extracted document.

[0021] In a possible implementation manner, the determining the association degree between the to-be-extracted documents according to the document type distribution vector corresponding to each to-be-extracted document specifically includes:

[0022] Perform word segmentation, syntactic analysis, and dependency analysis on each to-be-extracted document in sequence to determine the vocabulary corresponding to each to-be-extracted document;

[0023] Determine the core vocabulary in the vocabulary corresponding to each to-be-extracted document according to a preset core vocabulary library, and the core vocabulary library is obtained from one or more of a business domain dictionary, a system function set, and a function menu set;

[0024] For any two to-be-extracted documents D1 and D2, use the following formula to determine the association degree between the to-be-extracted documents:

[0025] R = R1 + R2 + R3 + R4 + R5

[0026] R1 = w1·(V1·V2)·2C / (N1 + N2)

[0027] R2 = w2·B1·B2

[0028] R3 = w3·α·A

[0029] R4 = w4·B1·B2

[0030] R5 = w5·M 12 / (M1 + M2)

[0031] Wherein, the R represents the association degree between the to-be-extracted documents D1 and D2, the R1 represents the business concept association degree, the R2 represents the chapter-level citation association degree, the R3 represents the section and paragraph-level citation association degree, the R4 represents the semantic-level association degree, and the R5 represents the word-level association degree; the w1 represents the business concept association weight, the w2 represents the chapter-level citation association weight, the w3 represents the section and paragraph-level citation association weight, the w4 represents the semantic-level association weight, and the w5 represents the word-level association weight;

[0032] Let \(V1\) represent the document type distribution vector of the document \(D1\) to be extracted, and \(V2\) represent the document type distribution vector of the document \(D2\) to be extracted; let \(N1\) represent the number of core vocabulary in the document \(D1\) to be extracted, \(N2\) represent the number of core vocabulary in the document \(D2\) to be extracted, and \(C\) represent the number of identical core vocabulary in the documents \(D1\) and \(D2\) to be extracted; let \(B1\) represent the word vector corresponding to the document \(D1\) to be extracted, and \(B2\) represent the word vector corresponding to the document \(D2\) to be extracted; let \(A\) represent the number of times the document \(D2\) is cited in the chapter paragraphs of the document \(D1\) to be extracted, and \(\alpha\) represent the correlation constant; let \(M1\) represent the frequency of occurrence of the core vocabulary in the document \(D1\) to be extracted, \(M2\) represent the frequency of occurrence of the core vocabulary in the document \(D1\) to be extracted, and \(M\) 12 represents the frequency of occurrence of the identical core vocabulary in the documents \(D1\) and \(D2\) to be extracted.

[0033] In one possible implementation manner, determining the sub-graph to be analyzed in the document relationship graph specifically includes:

[0034] Determining the first core document in the document relationship graph;

[0035] Deleting the documents in the document relationship graph whose correlation degree with the first core document is less than the preset correlation degree threshold;

[0036] Determining the sub-graph to be analyzed according to the remaining first documents in the document relationship graph.

[0037] In one possible implementation manner, extracting knowledge from the documents in the sub-graph according to the knowledge type to obtain the knowledge result corresponding to the knowledge type specifically includes:

[0038] Determining the second core document in the sub-graph to be analyzed, and the second documents other than the second core document, where the first documents include the second core document and the second documents;

[0039] Extracting knowledge from the second core document, and sequentially extracting knowledge from each of the second documents in descending order of correlation degree;

[0040] Among them, the extraction method for each of the first documents is:

[0041] Analyzing the first document to obtain the key features corresponding to the first document and the context information of each key feature, where the key features include one or more of the document title, chapter title, chapter relationship, paragraph, chart, and list;

[0042] Identifying the first document by chapter to determine the chapter corresponding to the knowledge type in the first document;

[0043] Perform knowledge extraction on each of the described chapters, the key features corresponding to the chapters, and the context information corresponding to the key features, so as to obtain the knowledge results corresponding to the first document;

[0044] Fuse the knowledge results of each first document according to the knowledge type to obtain the knowledge results corresponding to the knowledge type.

[0045] In one possible implementation manner, the fusing the knowledge results of each first document according to the knowledge type to obtain the knowledge results corresponding to the knowledge type specifically includes:

[0046] Fuse the knowledge results of each first document according to the knowledge type to obtain the first knowledge results corresponding to the knowledge type;

[0047] Judge whether there is already a second knowledge result corresponding to the knowledge type in the knowledge result library;

[0048] If so, fuse the first knowledge result and the second knowledge result according to the upload time of the document and / or the generation time of the knowledge result to obtain the knowledge result corresponding to the knowledge type.

[0049] In one possible implementation manner, the establishing the knowledge result library according to the knowledge results corresponding to the knowledge type specifically includes:

[0050] Store the knowledge results corresponding to each first document, the first knowledge results, the second knowledge results, and the knowledge results corresponding to the knowledge type in a preset storage form in the knowledge result library, and the storage form includes one or more of knowledge entities, knowledge graphs, knowledge documents, knowledge indexes, and knowledge change history lists;

[0051] Among them, the knowledge entity includes one or more of knowledge definitions, the name of the document where the knowledge is located, and the position in the document where the knowledge is located, and the knowledge change history list includes one or more of the knowledge extraction time, the release time of the relevant document, and the knowledge results corresponding to the extraction time.

[0052] In a second aspect, the present application provides a terminal device, including:

[0053] A receiving module, configured to determine the corresponding knowledge type to be retrieved according to the retrieved information after receiving the retrieved information input by the user, and the knowledge type includes one or more of business, responsibility, authority, process, and specification;

[0054] A processing module, configured to determine knowledge results corresponding to the knowledge type to be retrieved in a preset knowledge result library, and output a retrieval result to the user according to the knowledge results. The knowledge results include one or more of knowledge entities, knowledge graphs, knowledge documents, knowledge indexes, and knowledge change history lists. The knowledge result library is obtained by the following method: determining the association degree between the documents to be extracted, and constructing a document relationship graph according to the documents to be extracted and the association degree between the documents to be extracted; determining a sub-graph to be analyzed in the document relationship graph; performing knowledge extraction on the documents in the sub-graph according to the knowledge type to obtain knowledge results corresponding to the knowledge type; and establishing the knowledge result library according to the knowledge results corresponding to the knowledge type.

[0055] In a third aspect, the present application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0056] The memory stores computer execution instructions;

[0057] The processor executes the computer execution instructions stored in the memory to implement the above method.

[0058] In a fourth aspect, the present application provides a computer-readable storage medium, in which computer execution instructions are stored, and when the computer execution instructions are executed by a processor, they are used to implement the above method.

[0059] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the above method.

[0060] The information extraction processing method, device, medium and product provided by the present application can perform knowledge extraction on the documents to be extracted in advance according to the knowledge type to obtain knowledge results corresponding to different knowledge types, and then establish a knowledge result library according to all the knowledge results. After receiving the retrieval information input by the user, the knowledge results can be obtained from the preset knowledge result library according to the knowledge type corresponding to the retrieval information, and the knowledge results are output to the user as the retrieval results. Among them, the knowledge type can be set according to the data characteristics of the bank financial system, making the knowledge extraction more targeted and improving the accuracy and efficiency of the retrieval. Further, the knowledge results can include one or more of knowledge entities, knowledge graphs, knowledge documents, knowledge indexes, and knowledge change history lists, and are no longer just a document list. Through the knowledge results in various presentation forms, users can obtain the content they want to retrieve more simply and conveniently, without having to click on the document list one by one to search, further improving the retrieval efficiency, and at the same time, also improving the use value of the bank financial system data.

[0061] Furthermore, due to the existence of a large amount of unstructured data in the banking and financial system, such as institutional documents, etc., these unstructured data usually have the characteristics of frequent updates and being systematic, and they will reference and link to each other. Therefore, in order to improve the accuracy and convenience of retrieving these unstructured data, when building a knowledge result library, the correlation degree between the documents to be extracted can be determined, and a document relationship graph can be constructed based on the documents to be extracted and the correlation degree between them. By introducing the correlation degree between documents into the knowledge graph to construct the document relationship graph, the document relationship graph can represent the correlation degree between different documents, improving the accuracy and convenience of knowledge extraction based on the document relationship graph, and further improving the accuracy of retrieval. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0063] Figure 1 It is a flowchart of an information extraction processing method according to an embodiment of the present application;

[0064] Figure 2 It is a flowchart of an information extraction processing method according to another embodiment of the present application;

[0065] Figure 3 It is a schematic diagram of a document relationship graph according to an embodiment of the present application;

[0066] Figure 4 It is a schematic diagram of the structure of a terminal device according to an embodiment of the present application;

[0067] Figure 5 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present application.

[0068] Through the above accompanying drawings, specific embodiments of the present application have been shown, and there will be more detailed descriptions hereinafter. These drawings and textual descriptions are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0069] Here, exemplary embodiments will be described in detail, and their examples are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0070] In the technical solution of this application, the processing of information such as financial data or user data, including collection, storage, use, processing, transmission, provision, and disclosure, complies with the provisions of relevant laws and regulations and does not violate public order and good customs.

[0071] It should be noted that the information extraction processing method, device, medium, and product of this application can be used in the financial field and can also be used in any field other than the financial field. The application fields of the information extraction processing method, device, medium, and product of this application are not limited.

[0072] With the development of Internet technology and social progress, bank financial data has grown explosively, and the processing of this data has become increasingly important. For example, when a bank user wants to retrieve from a large amount of data, how to accurately obtain the content retrieved by the user from a large amount of data.

[0073] In the prior art, generally, the method of knowledge extraction is used to extract the content retrieved by the user from a large amount of data. The existing knowledge extraction methods are mainly the following two: One is to first perform entity extraction on the data based on natural language processing (NLP), and then perform operations such as entity alignment and entity linking on the extracted entities, and construct a knowledge graph based on the relationships between the entities. The other is to construct a statistical language model based on deep learning, and then use the model to extract features in the data and establish relationships or links between the features to obtain the final knowledge. The existing knowledge extraction methods are all general and popular, more reflected in the direction of entity extraction and relationship extraction, and after completing knowledge extraction, a list of relevant documents will be output as the retrieval result to the user.

[0074] However, on the one hand, there is a large amount of unstructured data in the bank financial system, mainly including various policy and regulation documents, institutional documents, office documents, various information, etc., which probably account for more than 80% of the total system data weight. The characteristics of this kind of content data are that there are many types, the content does not have a very fixed pattern, and the usage purposes and analysis requirements of data in different business fields and application scenarios vary greatly. Each business department and business line of the bank will issue various institutional documents to unify business definitions and standardize business operations. Such content documents have the characteristics of being released relatively frequently and the definitions, operations, processes, norms, etc. of the same business being frequently updated as the business develops. Using the existing knowledge extraction methods to process the data has low accuracy, resulting in low retrieval accuracy of the bank financial system.

[0075] On the other hand, the existing knowledge extraction methods generally output a list of documents as the retrieval result to the user, and the user needs to click on the documents one by one to search, which is cumbersome and has low retrieval efficiency.

[0076] Based on this technical problem, the inventive concept of the present application lies in: how to provide an information extraction and processing method with higher retrieval accuracy and efficiency in the bank financial system.

[0077] The information extraction and processing method provided by the present application aims to solve the above technical problems in the prior art. This method can pre-extract knowledge from the documents to be extracted according to the knowledge types to obtain knowledge results corresponding to different knowledge types, and then establish a knowledge result library based on all the knowledge results. After receiving the retrieval information input by the user, the knowledge result can be obtained from the preset knowledge result library according to the knowledge type to be retrieved corresponding to the retrieval information, and the knowledge result is output to the user as the retrieval result. Among them, the knowledge type can be set according to the data characteristics of the bank financial system, making the knowledge extraction more targeted and improving the retrieval accuracy and efficiency. Further, the knowledge result can include one or more of knowledge entities, knowledge graphs, knowledge documents, knowledge indexes, and knowledge change history lists, not just a document list. Through the knowledge results in various display forms, users can obtain the content they want to retrieve more simply and conveniently, without having to click on the document list one by one to search, further improving the retrieval efficiency. At the same time, it also improves the utilization value of the data in the bank financial system.

[0078] Further, since there is a large amount of unstructured data in the bank financial system, such as institutional documents, etc., these unstructured data usually have the characteristics of frequent updates and systematization, and they will reference and link to each other. Therefore, in order to improve the accuracy and convenience of retrieving these unstructured data, when establishing the knowledge result library, the correlation degree between the documents to be extracted can be determined, and a document relationship graph can be constructed according to the documents to be extracted and the correlation degree between them. By introducing the correlation degree between each document into the knowledge graph to construct the document relationship graph, the document relationship graph can represent the correlation degree between different documents, improving the accuracy and convenience of knowledge extraction based on the document relationship graph, and further improving the retrieval accuracy.

[0079] The technical solution of the present application and how the technical solution of the present application solves the above technical problems will be described in detail below with specific embodiments. These specific embodiments below can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below with reference to the accompanying drawings.

[0080] Embodiment 1

[0081] Figure 1It is a flowchart of an information extraction processing method provided by an embodiment of the present application. The execution subject of the information extraction processing method provided by the embodiment of the present application can be a data processing device or a terminal device integrated with a data processing device (abbreviation: terminal device). In this embodiment, the information extraction processing method is described by taking the execution subject as the terminal device. As Figure 1 shown, the information extraction processing method may include the following steps:

[0082] S101: After receiving the retrieval information input by the user, determine the corresponding knowledge type to be retrieved according to the retrieval information. The knowledge type may include one or more of business, responsibility, authority, process, and specification.

[0083] It should be noted that most of the unstructured data in the bank financial system is institutional documents, and the institutional documents also fully reflect the characteristics of the data in the bank financial system. Therefore, in this embodiment, the institutional documents are used as the object of data processing for exemplary elaboration.

[0084] In this embodiment, the above-mentioned user may refer to bank staff, such as customer service staff, training staff, etc. When customer service staff answer questions, they may not know the answers to all questions and may need to retrieve information according to the questions. When training staff conduct business or institutional training, they may need to systematically summarize the business or institution and also need to retrieve information. Of course, the above is only an exemplary description of the retrieval application scenario and is not limited thereto.

[0085] In this embodiment, the retrieval information may be a sentence or a word, and no specific limitation is imposed on the specific form of the retrieval information. Exemplarily, the user may input "credit card application process", "counterfeit currency detection", etc. as the retrieval information.

[0086] In this embodiment, the knowledge type may be determined according to the characteristics of the data in the bank financial system, such as business, responsibility, authority, process, specification, etc. Of course, the knowledge type is not limited thereto, and those skilled in the art can set it flexibly.

[0087] Exemplarily, when the user inputs "credit card application process" as the retrieval information, the knowledge type may be "process" and "credit card".

[0088] S102: Determine the knowledge results corresponding to the knowledge type to be retrieved in the preset knowledge result library, and output the retrieval results to the user according to the knowledge results. The knowledge results include one or more of knowledge entities, knowledge graphs, knowledge documents, knowledge indexes, and knowledge change history lists. The knowledge result library can be obtained by the following methods: Determine the association degree between the documents to be extracted, and construct a document relationship graph according to the documents to be extracted and the association degree between them; Determine the sub-graphs to be analyzed in the document relationship graph; Perform knowledge extraction on the documents in the sub-graph according to the knowledge type to obtain the knowledge results corresponding to the knowledge type; Establish a knowledge result library according to the knowledge results corresponding to the knowledge type.

[0089] In this embodiment, for the specific implementation manner of obtaining the knowledge result library in step S102 above, please refer to Embodiment 2 for details.

[0090] In this embodiment, those skilled in the art can flexibly set the presentation form of the knowledge results. Preferably, in order to facilitate the user to view the retrieval results, the knowledge results can be presented in the form of knowledge entities. The knowledge entities can include attributes such as knowledge definitions and operation processes, so that the user can have a clear understanding of the questions asked and does not need to search for the retrieval results multiple times. Further, when a customer asks questions, they generally do not ask only one question, but usually ask multiple related questions. Therefore, in order to improve the service quality and the efficiency of the reply, the knowledge results can also be presentation forms presented in a knowledge type system such as knowledge graphs, knowledge indexes, and knowledge change history lists. If the customer asks multiple related questions, the user does not need to perform multiple retrievals. Further, the knowledge results can also be presented in the form of knowledge documents. By presenting the original documents, it can provide a knowledge support basis for the user, not only making the user understand that other knowledge presentation forms are well-grounded, but also facilitating the user to view.

[0091] The method of this application can pre-perform knowledge extraction on the documents to be extracted according to the knowledge type to obtain the knowledge results corresponding to different knowledge types, and then establish a knowledge result library according to all the knowledge results. After receiving the retrieval information input by the user, the knowledge results can be obtained from the preset knowledge result library according to the knowledge type to be retrieved corresponding to the retrieval information, and the knowledge results are output to the user as the retrieval results. Among them, the knowledge type can be set according to the data characteristics of the bank financial system, making the knowledge extraction more targeted and improving the accuracy and efficiency of the retrieval. Further, the knowledge results can include one or more of knowledge entities, knowledge graphs, knowledge documents, knowledge indexes, and knowledge change history lists, and are no longer just a document list. Through the knowledge results in various presentation forms, the user can obtain the content to be retrieved more simply and conveniently, without having to click on the document list one by one to search, further improving the retrieval efficiency. At the same time, it also improves the utilization value of the bank financial system data.

[0092] Furthermore, due to the existence of a large amount of unstructured data in the banking and financial system, such as institutional documents, etc., these unstructured data usually have the characteristics of frequent updates and systematization, and they will reference and link to each other. Therefore, in order to improve the accuracy and convenience of retrieving these unstructured data, when building a knowledge result library, the correlation degree between the documents to be extracted can be determined, and a document relationship graph can be constructed based on the documents to be extracted and the correlation degree between the documents to be extracted. By introducing the correlation degree between each document into the knowledge graph to construct the document relationship graph, the document relationship graph can represent the correlation degree between different documents, improving the accuracy and convenience of knowledge extraction based on the document relationship graph, and further improving the accuracy of retrieval.

[0093] Through the full deconstruction of institutional documents in this application, knowledge entries are extracted and a certain degree of integration is performed. Then, users can directly retrieve knowledge points and can also return knowledge, entities, documents, etc. related to the knowledge points. What is presented is a knowledge context and a structured and organized knowledge system, rather than the original text document, which can effectively improve the user experience and increase the utilization rate of institutional documents.

[0094] Next, a specific Example 2 will be used to elaborate in detail on the specific implementation manner of obtaining the knowledge result library in step S102 of the above Example 1.

[0095] Example 2

[0096] Figure 2 is a flowchart of an information extraction processing method provided by another embodiment of this application. The execution subject of the information extraction processing method provided by the embodiment of this application can be a data processing device or a terminal device integrated with a data processing device (abbreviation: terminal device). In this embodiment, the execution subject is the terminal device to illustrate the information extraction processing method. As Figure 2 shown, the information extraction processing method may include the following steps:

[0097] S201: Determine the correlation degree between the documents to be extracted, and construct a document relationship graph based on the documents to be extracted and the correlation degree between the documents to be extracted.

[0098] In this embodiment, the knowledge result library may include all institutional documents of the banking and financial system, and the knowledge result library may include a data access interface through which users can upload relevant documents. When a new document is uploaded to the knowledge result library, the document can be used as a document to be extracted and corresponding knowledge results can be obtained through knowledge extraction.

[0099] Exemplarily, the knowledge result library may include 1000 documents. Knowledge extraction has been performed on these documents to obtain corresponding knowledge results, and both these documents and the knowledge results are classified and stored in the knowledge result library according to the knowledge type. A certain user uploads 10 documents through the data interface of the knowledge result library. These 10 documents can be used as the documents to be extracted. After knowledge extraction, they will be classified and stored in the knowledge result library together with the knowledge results.

[0100] Of course, the documents and the knowledge results can also be stored in different areas of the knowledge result library respectively. When the documents and the knowledge results are stored separately, corresponding classification storage can also be performed on the documents. For example, the documents can be classified and stored according to the business department, business field, document type, target reader, etc.

[0101] In this embodiment, the knowledge result library may further include a data service interface. Users can download the knowledge results through the data service interface for other systems or other services. Of course, this data service interface can also be used for document screening, unloading or pushing, and there is no limitation here.

[0102] Exemplarily, a certain user wants to obtain a knowledge result in a specific form for display, but this form does not exist in the knowledge result library. Then the user can download the corresponding knowledge results and then perform conversion to obtain the desired form.

[0103] In a possible implementation manner, determining the correlation degree between the documents to be extracted in the above step S201 may include: determining the document type corresponding to each document to be extracted, and determining the document type distribution vector corresponding to each document to be extracted according to the document type. The document type includes one or more of system management, training management, business management, operation operation, business operation; determining the correlation degree between the documents to be extracted according to the document type distribution vector corresponding to each document to be extracted.

[0104] In this implementation manner, those skilled in the art can flexibly set the document type according to the characteristics of the institutional documents of the banking and financial system. For example, system management, training management, business management, operation operation, business operation, etc. Of course, the document type is not limited to this.

[0105] In this implementation manner, when the document is uploaded to the knowledge result library, the document can be classified and sorted according to the document type. And bank institutional documents generally have several fixed types, including but not limited to system management methods, training management methods, business management methods, operation operation processes, business operation procedures, etc. For example, the system management method mainly emphasizes the responsibilities, requirements and specifications of system construction; the training management method is mainly for business training; the business operation procedure mainly emphasizes operation permissions, processes and specifications, etc.

[0106] In this embodiment, the institutional documents of the banking and financial system are basically related to its business. Even for the same business, the correlation degrees of documents of different document types are different. For example, the related businesses of the "Training Management Measures for Anti-counterfeiting Currency", the "Operation Rules for Anti-counterfeiting Currency", and the "Work Management Measures for Anti-counterfeiting Currency" are all anti-counterfeiting currency, but the document types of these three documents are different, and the correlation degrees between any two documents are also different. Therefore, the document type distribution vector corresponding to each document to be extracted can be determined according to the document type, and the correlation degree between the documents to be extracted can be determined according to the document type distribution vector corresponding to each document to be extracted. Through such a setting, the accuracy of the correlation degree between the documents to be extracted can be improved. In addition, by setting the document type according to the characteristics of the institutional documents of the banking and financial system, the accuracy of the correlation degree between the documents to be extracted can be further improved.

[0107] In a possible embodiment, determining the document type corresponding to each document to be extracted and determining the document type distribution vector corresponding to each document to be extracted according to the document type may include: determining the type description word corresponding to each document to be extracted according to a preset type dictionary; determining the document type corresponding to each document to be extracted according to the corresponding relationship between the preset type description word and the document type; performing normalization processing on the document type to determine the document type distribution vector corresponding to each document to be extracted.

[0108] In this embodiment, the title of the institutional document of the banking and financial system is generally set as "<Bank Name><Business Concept><Document Type>", where there may occasionally be words such as "Regarding" in front of the bank name; the business concept is generally a specific banking business, such as the foreign currency exchange business. Therefore, by identifying and extracting the title of the institutional document or the first few paragraphs of the document text, the type description word representing the document type can be obtained. For example, "Process", "Operation", "Operation", etc.

[0109] In this embodiment, the type dictionary can be enumerated and integrated by those skilled in the art in advance for the relevant description words of the document type. It can be a dictionary template including all type description words. Of course, the type dictionary can continuously update and organize the type description words therein.

[0110] In this embodiment, determining the type description word corresponding to each document to be extracted according to the preset type dictionary can be performed by using the exact match or fuzzy match method. The fuzzy match method can be based on the edit distance. Here, the existing exact match or fuzzy match method can be used for matching, which will not be elaborated here.

[0111] In this embodiment, since the institutional documents of the banking and financial system have obvious characteristics, that is, the general document type is reflected in the document title or the text. Therefore, a type dictionary including all types of vocabulary can be preset, and the type description words of the document to be extracted can be determined according to the type dictionary. According to the type description words, the document type corresponding to the document to be extracted can be determined simply and accurately. After obtaining the document type, by performing normalization processing on the document type, the document type distribution vector corresponding to each document to be extracted can be determined simply and conveniently.

[0112] Alternatively, to determine the document type corresponding to each document to be extracted and determine the document type distribution vector corresponding to each document to be extracted according to the document type, it may further include: using multiple binary classifiers to classify each document to be extracted to obtain the classification probability value of each document to be extracted; determining the document type of each document to be extracted according to the type corresponding to the maximum classification probability value, and determining the document type distribution vector corresponding to each document to be extracted according to the classification probability value of each document to be extracted.

[0113] In this embodiment, existing binary classifier technologies can be used to classify documents. Specifically, the target type can be used as the positive type, and all other types can be used as the negative type to train N (the number of document types) binary classifiers. When a new document type needs to be added, positive type documents can be screened in a timely manner, and the existing other document data can be used as negative type documents to construct a classifier. The specific structure of the binary classifier can refer to existing technologies and will not be elaborated here.

[0114] Exemplarily, the document to be extracted can be input into N classifiers to obtain N classification results. The value output by the classifier can be regarded as the probability that the document belongs to this classification. The type corresponding to the classifier with the largest value is the most likely document type of this document. All probability values can be retained during this process. For example, if there are 10 document types arranged in a certain fixed order such as C1, C2…C9, C10. Then the prediction result of each document is used as a vector with a length of 10. For example, [0.82, 0.53, 0.12, 0.61, …, 0.23], and this vector can be used as the document type distribution vector.

[0115] In this embodiment, specific classifiers can use classifiers based on statistical models such as support vector machines and naive Bayes, and the way of ensemble learning can be considered to improve the classification accuracy according to the actual document data volume and other situations. Deep learning methods can also be used to reduce the processing of feature engineering, but in the case of a small number of documents, methods based on statistical learning models are more often used.

[0116] In this embodiment, the method for determining the document type according to the preset type dictionary mainly relies on whether the document title conforms to the general pattern of institutional documents. The method of using the type dictionary can solve part of the problem of determining the document type, but for documents whose document types do not appear in the document title, effective classification cannot be achieved. Therefore, the document type can be determined by classifying each document to be extracted through multiple binary classifiers, so that the determination of the document type is more comprehensive and effective, and the accuracy is higher. Further, the binary classifier will output the classification probability value of each document to be extracted for each type. According to the classification probability value, the document type distribution vector corresponding to the document to be extracted can be simply and directly determined without further normalization processing, and the operation is simple and convenient.

[0117] In a possible implementation manner, according to the document type distribution vector corresponding to each document to be extracted, determining the correlation degree between the documents to be extracted may include:

[0118] S1: Perform word segmentation, syntactic analysis, and dependency analysis on each document to be extracted in turn to determine the vocabulary corresponding to each document to be extracted.

[0119] S2: According to the preset core vocabulary library, determine the core vocabulary in the vocabulary corresponding to each document to be extracted. The core vocabulary library is obtained from one or more of the business domain dictionary, system function set, and function menu set.

[0120] S3: For any two documents to be extracted D1 and D2, use the following formulas (1)-(6) to determine the correlation degree between the documents to be extracted:

[0121] R = R1 + R2 + R3 + R4 + R5 (1)

[0122] R1 = w1·(V1·V2)·2C / (N1 + N2) (2)

[0123] R2 = w2·B1·B2 (3)

[0124] R3 = w3·α·A (4)

[0125] R4 = w4·B1·B2 (5)

[0126] R5 = w5·M 12 / (M1 + M2) (6)

[0127] Among them, R represents the correlation degree between the to-be-extracted documents D1 and D2, R1 represents the business concept correlation degree, R2 represents the chapter-level citation correlation degree, R3 represents the section and paragraph-level citation correlation degree, R4 represents the semantic-level correlation degree, and R5 represents the word-level correlation degree; w1 represents the business concept correlation weight, w2 represents the chapter-level citation correlation weight, w3 represents the section and paragraph-level citation correlation weight, w4 represents the semantic-level correlation weight, and w5 represents the word-level correlation weight;

[0128] V1 represents the document type distribution vector of the to-be-extracted document D1, and V2 represents the document type distribution vector of the to-be-extracted document D2; N1 represents the number of core words in the to-be-extracted document D1, N2 represents the number of core words in the to-be-extracted document D2, and C represents the number of identical core words in the to-be-extracted documents D1 and D2; B1 represents the word vector corresponding to the to-be-extracted document D1, and B2 represents the word vector corresponding to the to-be-extracted document D2; A represents the number of times the to-be-extracted document D2 is cited in the sections and paragraphs of the to-be-extracted document D1, and α represents the correlation degree constant; M1 represents the frequency of occurrence of the core word in the to-be-extracted document D1, M2 represents the frequency of occurrence of the core word in the to-be-extracted document D1, and M 12 represents the frequency of occurrence of the identical core words in the to-be-extracted documents D1 and D2.

[0129] In this embodiment, step S1 can first perform word segmentation on the to-be-extracted document, that is, split the sentences in the document into words, and then perform syntactic analysis and dependency analysis on the split words. Through such a setting, the split words can be expanded to make the vocabulary corresponding to the to-be-extracted document richer.

[0130] In this embodiment, the core word refers to the business core word, that is, the word associated with the business. The business domain dictionary, the system function set, and the function menu set can be preset by those skilled in the art according to experience or through enumeration, and are the core word library including all business core words.

[0131] In this embodiment, for the weights w1, w2, w3, w4, and w5, those skilled in the art can flexibly set them according to the applicable application scenario. w1, w2, w3, w4, and w5 can be the same or different, and no restrictions are made here.

[0132] In this embodiment, business concept association means that multiple documents are targeted at the same business concept or core entity. For example, "Training Management Measures for Anti-counterfeiting Currency", "Operation Procedures for Anti-counterfeiting Currency", "Work Management Measures for Anti-counterfeiting Currency", etc. are all institutional documents centered around anti-counterfeiting currency. Such documents belong to business concept association and are documents under the same business entity. Business concept association analysis is first processed through the matching recognition of the business core words in the document titles. If the business core concept words are included in the titles of both judged documents, it can be determined that the two documents belong to the same business concept. For documents that cannot match the business core words in the titles, the association degree between the two can be calculated. The overlap of the two documents in business concepts will be reflected in the core words in the documents. Here, the word frequency is not the key concern, mainly looking at the matching degree of the core words. And considering that even if two documents belong to the same business concept, their document types may be different. Therefore, a document type distribution vector can also be introduced based on the matching degree of the core words to make the business concept association degree more accurate.

[0133] In this embodiment, the characteristic of chapter-level reference association is that it is in the document body but not under a certain chapter of the document. For example, it appears at the beginning of the document or directly in the reference document list in the document. That is, chapter-level reference association means that a document references another document. If there is no chapter-level reference association, R2 can be counted as 0.

[0134] In this embodiment, the word vectors corresponding to the document to be extracted can be constructed by the bag-of-words model or the Embedding method, or with the help of an external word vector library. The specific construction method of the word vectors is not restricted here. B1·B2 is to calculate the cosine similarity between the document to be extracted D1 and D2.

[0135] In this embodiment, section-paragraph-level reference association means that other institutional documents are cited in certain sections and paragraphs of the target document. The association degree of such references can be accumulated according to the number of occurrences. Each time a reference appears, the association degree of the corresponding document increases by an association degree constant. Exemplarily, the association degree constant can be 0.1.

[0136] In this embodiment, since institutional documents present the characteristics of being systematic and continuously updated, it is necessary to focus on considering the association relationship between documents. When calculating the association degree between documents, by introducing business concept association, chapter-level reference association, section-paragraph-level reference association, semantic-level association and word-level association, the association degree between the documents to be extracted calculated accordingly is more accurate.

[0137] S202: Determine the sub-graph to be analyzed in the document relationship graph.

[0138] In a possible implementation, the above step S202 of determining the sub-graph to be analyzed in the document relationship graph may include: determining the first core document in the document relationship graph; deleting the documents in the document relationship graph whose association degree with the first core document is less than a preset association degree threshold; and determining the sub-graph to be analyzed according to the remaining first documents in the document relationship graph.

[0139] In this implementation, the importance of each document in the document relationship graph can be determined according to a sorting algorithm, such as the PageRank algorithm, and the document with the highest importance is the first core document.

[0140] In this implementation, considering that when a certain document is used as input for query, its importance is not high, and the association degrees of its 1-degree documents (connected by only one relationship line) may be very low and cannot reach the set threshold, so it is default to query documents within the range of no less than 2-degree relationships (connected by two or more relationship lines) each time.

[0141] In this implementation, the document relationship graph may include all documents with association degrees, but in fact, some documents have no association and only use the same vocabulary. Therefore, the document relationship graph can be denoised first, that is, deleting the documents with association degrees less than the threshold. Through such a setting, not only can the workload of subsequent knowledge extraction be reduced, but the accuracy of the knowledge results can also be guaranteed.

[0142] S203: Extract knowledge from the documents in the sub-graph according to the knowledge type to obtain the knowledge results corresponding to the knowledge type.

[0143] In a possible implementation, the above step S203 of extracting knowledge from the documents in the sub-graph according to the knowledge type to obtain the knowledge results corresponding to the knowledge type may include:

[0144] S2031: Determine the second core document in the sub-graph to be analyzed, and the second documents other than the second core document. The first documents include the second core document and the second documents.

[0145] S2032: Extract knowledge from the second core document, and extract knowledge from each second document in turn in descending order of the association degree;

[0146] Among them, the extraction method for each first document is as follows: parse the first document to obtain the key features corresponding to the first document and the context information of each key feature. The key features include one or more of the document title, chapter title, chapter relationship, paragraph, chart, and list; identify the first document by chapter to determine the chapter corresponding to the knowledge type in the first document; sequentially perform knowledge extraction on each chapter, the key features corresponding to the chapter, and the context information corresponding to the key features to obtain the knowledge result corresponding to the first document.

[0147] S2033: Integrate the knowledge results of each first document according to the knowledge type to obtain the knowledge result corresponding to the knowledge type.

[0148] In this embodiment, when integrating the knowledge results of each first document, the integration can be based on the knowledge result of the second core document.

[0149] In this embodiment, the importance of each first document in the sub-graph to be analyzed can be determined according to a sorting algorithm, such as the PageRank algorithm. The document with the highest importance is the second core document. The core document will be used as the preferred document for knowledge extraction of institutional documents. Because the standard definition of business concepts is often given in the core document. Other institutional documents may be derived from the core document. For example, when the management institution issues new requirements and notices for a certain business, the corresponding business department of the bank will formulate internal corresponding institutional documents based on the notice. The notice is generally cited in the institutional document. So, one possible situation here is that the source document may not be the core document (such as the notice document issued by the management institution), and the first institutional document issued within the bank will become the core document. However, the relationships between these documents can be found in the document relationship graph.

[0150] In this embodiment, the knowledge extraction of this application is different from general entity extraction and relationship extraction. Instead, it extracts useful information from institutional documents, that is, it fragments the free text and then organizes it to form understandable, intuitive, and comprehensive knowledge information. The content of institutional documents generally includes key information elements such as "definitions", "business explanations", "processes", "principles", "responsibilities", "operation procedures", etc. These key information elements constitute institutional knowledge entries, which may be expressed in the form of multiple paragraphs, tables, lists, etc. in the document. Therefore, it is necessary to consider the key features corresponding to different types of element information in the document. These key features mainly include "titles", "texts", "lists", "tables", "references", etc. Different types of key features are mainly used to divide boundaries and determine extraction rules. For example, when extracting operation procedures, general operation procedures are usually presented in the form of tables or lists. After identifying the tables and lists in the document and then identifying whether they are operation procedures, a large number of useless operations can be excluded.

[0151] In this embodiment, after identifying the knowledge type in units of chapters, the chapters corresponding to the knowledge type can be identified. Then, when extracting, traditional knowledge extraction tasks can be performed on the chapters corresponding to the knowledge type, including entity recognition and relationship recognition. After identifying the entities and relationships, entity linking and alignment operations can be performed through constructed thesauruses such as business thesauruses, system functions, menu functions, etc. and the knowledge that has been extracted.

[0152] Exemplarily, Figure 3 is a schematic diagram of the document relationship graph of an embodiment of this application, as Figure 3 shown. The core document is d1. Since there are only two-level connection lines with d1 in this document relationship graph, the document is not deleted. When performing knowledge extraction, first extract d1, and then starting from d1 based on BFS and in accordance with the degree of association, the next ones to be analyzed are d5, d2, d4, d3, d8, d6, d7.

[0153] In this embodiment, after deleting the documents in the document relationship graph whose association degree with the first core document is less than the threshold, the core document of the remaining first document may change, that is, the first core document may no longer be the core document. Therefore, it is necessary to use a sorting algorithm to correct the core document to obtain the second core document. Further, parse the first document to obtain the key features corresponding to the first document, and then perform knowledge extraction according to the key features, making the knowledge extraction more targeted, which can improve the efficiency and accuracy of knowledge extraction. Further, identify the knowledge type in units of chapters, which can remove the chapters irrelevant to the knowledge type, thereby reducing the workload of knowledge extraction and further improving the efficiency and accuracy of knowledge extraction. Further, after extracting the knowledge results of the chapters corresponding to the knowledge type, the knowledge results related to the knowledge type in all documents can be fused to make the final knowledge results more comprehensive and accurate.

[0154] In a possible implementation manner, fusing the knowledge results of each first document according to the knowledge type to obtain the knowledge results corresponding to the knowledge type may include: fusing the knowledge results of each first document according to the knowledge type to obtain the first knowledge results corresponding to the knowledge type; determining whether there are already second knowledge results corresponding to the knowledge type in the knowledge result library; if so, fusing the first knowledge results and the second knowledge results according to the upload time of the document and / or the generation time of the knowledge results to obtain the knowledge results corresponding to the knowledge type.

[0155] In this embodiment, with the update of the documents, there may be multiple documents defining the business responsibilities of each department. When performing fusion, the business, organization, and authority responsibility points can be identified in sequence. All the original information of the knowledge on both sides will be retained during fusion. In addition, the time factor of the documents needs to be considered. In principle, the knowledge defined later is more accurate than the documents with older time. The problem of knowledge point coverage needs to be considered during fusion, and the differences need to be identified and stored in the knowledge result library for subsequent analysis and tracking of the differences.

[0156] In this embodiment, the above embodiment has described the knowledge extraction process of the documents related to a new business. However, due to the characteristics of institutional documents being systematic and continuously updated, the results obtained in the above embodiment may be the knowledge extraction results of the updated documents of the existing business. Therefore, after obtaining the first knowledge result, it is also possible to determine whether there is a second knowledge result corresponding to the knowledge type in the knowledge result library. If so, the first knowledge result and the second knowledge result are fused to obtain the knowledge result corresponding to the knowledge type. By such a setting, the accuracy of the knowledge result can be improved. Further, when fusing, it can be fused according to the upload time of the document and / or the generation time of the knowledge result. In principle, the knowledge defined later is more accurate than the documents with older time. By such a setting, the accuracy of the knowledge result can be further improved.

[0157] S204: Establish a knowledge result library according to the knowledge result corresponding to the knowledge type.

[0158] In a possible implementation manner, the above step S204 of establishing a knowledge result library according to the knowledge result corresponding to the knowledge type may include: storing the knowledge result corresponding to each first document, the first knowledge result, the second knowledge result, and the knowledge result corresponding to the knowledge type in a preset storage form in the knowledge result library. The storage form includes one or more of knowledge entities, knowledge graphs, knowledge documents, knowledge indexes, and knowledge change history lists; among them, the knowledge entity includes one or more of knowledge definitions, the name of the document where the knowledge is located, and the position in the document where the knowledge is located. The knowledge change history list includes one or more of the knowledge extraction time, the release time of the relevant document, and the knowledge result corresponding to the extraction time.

[0159] In this embodiment, the knowledge graph can store, reorganize, and express the content of institutional documents in the form of a knowledge graph. The responsibilities and authorities extracted from the document can be filled into the corresponding entity and relationship data, and the specific authority descriptions are stored as relationship attributes. The entity attributes can include core concepts such as business, system, special work, organization, account, and operation management. Among them, the special work includes temporary special work items such as special training and anti-money laundering. The established relationship system can include the responsibility and authority relationship between the organization and the business or system.

[0160] In this embodiment, the knowledge change history list may include the time of knowledge result extraction, the release time of relevant documents, and the result after knowledge result fusion at that time point. In this way, a timeline of knowledge results is constructed, which facilitates historical backtracking of knowledge results. For documents of the same business concept, they can also be stored and displayed in the form of a timeline. For different versions of documents of the same business concept and the same document type (i.e., the same document in the general sense), a differential history line is constructed, which can intuitively display the evolution of a certain operation process or the change of the responsibilities of an organization in a certain business.

[0161] In this embodiment, when constructing the knowledge index, the extracted knowledge results can be stored in Elasticsearch, and indexes can be established for its related business concepts, system functions, function menus, etc.; indexes can be established for its related documents. In this way, an all-round index system structure from business concepts, business entities to knowledge points and then to specific documents is formed.

[0162] In this embodiment, all the results extracted in the knowledge extraction process can be stored in the knowledge result library for easy analysis and tracking. Further, by storing the knowledge results in various forms of expression, the use value of the knowledge results can be improved, and the knowledge results can also be made more comprehensive, so that users can obtain the content they want to retrieve more simply and conveniently.

[0163] In this embodiment, since there is a large amount of unstructured data in the bank financial system, such as institutional documents, etc., these unstructured data usually have the characteristics of frequent updates and systematization, and they will reference and link to each other. Therefore, in order to improve the accuracy and convenience of retrieving these unstructured data, when establishing the knowledge result library, the correlation degree between the documents to be extracted can be determined, and a document relationship graph can be constructed according to the documents to be extracted and the correlation degree between the documents to be extracted. By introducing the correlation degree between each document into the knowledge graph to construct the document relationship graph, the document relationship graph can represent the correlation degree between different documents, improving the accuracy and convenience of knowledge extraction based on the document relationship graph, and further improving the accuracy of retrieval. Further, by performing noise reduction processing on the document relationship graph, not only can the workload of subsequent knowledge extraction be reduced, but the accuracy of knowledge results can also be ensured. Further, by storing the knowledge results in various forms of expression, the use value of the knowledge results can be improved, and the knowledge results can also be made more comprehensive, so that users can obtain the content they want to retrieve more simply and conveniently.

[0164] Next, a specific embodiment is used to elaborate on the information extraction processing method of this application.

[0165] Embodiment III

[0166] In a specific embodiment, a customer service representative of a certain bank receives a question from a customer. The customer asks about the application process for a credit card, but the customer service representative is not clear about the application process. The customer service representative enters "application process for credit card" in the search bar on the knowledge result database page of the bank's terminal device for retrieval. The specific retrieval process is as follows:

[0167] First step, after the terminal device receives the retrieval information "application process for credit card" input by the customer service representative, it determines the corresponding knowledge types to be retrieved as "credit card" and "process".

[0168] Second step, the terminal device searches the knowledge result database to obtain knowledge results related to "credit card" and "process". The knowledge results include knowledge entities, knowledge graphs, and knowledge documents, and outputs the knowledge results as retrieval results to the customer service representative. The customer service representative provides relevant answers to the customer based on the application process of the credit card presented by the knowledge entity.

[0169] Figure 4 The structural schematic diagram of the terminal device according to an embodiment of the present application is as follows Figure 4 As shown, the terminal device includes: a receiving module 41, configured to determine the corresponding knowledge types to be retrieved according to the retrieval information after receiving the retrieval information input by the user. The knowledge types may include one or more of business, responsibility, authority, process, and specification; a processing module 42, configured to determine the knowledge results corresponding to the knowledge types to be retrieved in the preset knowledge result database, and output retrieval results to the user according to the knowledge results. The knowledge results include one or more of knowledge entities, knowledge graphs, knowledge documents, knowledge indexes, and knowledge change history lists. The knowledge result database is obtained by the following method: determining the correlation degree between the documents to be extracted, and constructing a document relationship graph according to the documents to be extracted and the correlation degree between the documents to be extracted; determining the sub-graphs to be analyzed in the document relationship graph; performing knowledge extraction on the documents in the sub-graphs according to the knowledge types to obtain the knowledge results corresponding to the knowledge types; establishing a knowledge result database according to the knowledge results corresponding to the knowledge types. In one implementation manner, the description of the specific functions implemented by the terminal device can refer to steps S101 - S102 in Embodiment 1 and steps S201 - S204 in Embodiment 2, which will not be elaborated here.

[0170] Figure 5 The structural schematic diagram of the electronic device according to an embodiment of the present application is as follows Figure 5 As shown, the electronic device includes: a processor 101, and a memory 102 communicatively connected to the processor 101; the memory 102 stores computer execution instructions; the processor 101 executes the computer execution instructions stored in the memory 102 to implement the steps of the information extraction processing method in each of the above method embodiments.

[0171] The electronic device can be independent or part of a terminal device. The processor 101 and the memory 102 can use the existing hardware of the terminal device.

[0172] In the above-mentioned electronic device, the memory 102 and the processor 101 are directly or indirectly electrically connected to achieve data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses or signal lines, such as through a bus connection. The memory 102 stores computer execution instructions for implementing the data access control method, including at least one software function module stored in the memory 102 in the form of software or firmware. The processor 101 executes various functional applications and data processing by running the software programs and modules stored in the memory 102.

[0173] The memory 102 can be, but is not limited to, a random access memory (Random Access Memory, abbreviated as RAM), a read-only memory (Read Only Memory, abbreviated as ROM), a programmable read-only memory (Programmable Read-Only Memory, abbreviated as PROM), an erasable programmable read-only memory (Erasable Programmable Read-Only Memory, abbreviated as EPROM), an electrically erasable programmable read-only memory (Electric Erasable Programmable Read-Only Memory, abbreviated as EEPROM), etc. Among them, the memory 102 is used to store programs, and the processor 101 executes the programs after receiving the execution instructions. Further, the software programs and modules in the above-mentioned memory 102 may also include an operating system, which may include various software components and / or drivers for managing system tasks (such as memory management, storage device control, power management, etc.), and may communicate with various hardware or software components to provide a running environment for other software components.

[0174] The processor 101 can be an integrated circuit chip with signal processing capabilities. The above-mentioned processor 101 can be a general-purpose processor, including a central processing unit (Central Processing Unit, abbreviated as CPU), a network processor (Network Processor, abbreviated as NP), etc. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0175] An embodiment of the present application further provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the steps of the method embodiments of the present application.

[0176] An embodiment of the present application further provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the steps of the method embodiments of the present application.

[0177] Those skilled in the art will readily conceive of other implementations of the present application after considering the specification and practicing the invention disclosed herein. The present application aims to cover any variations, uses, or adaptations of the present application, which follow the general principles of the present application and include well-known knowledge or conventional technical means in the technical field not disclosed in the present application. The specification and embodiments are only regarded as exemplary, and the true scope and spirit of the present application are pointed out by the appended claims.

[0178] It should be understood that the present application is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present application is only limited by the appended claims.

Claims

1. An information extraction processing method, characterized in that: include: After receiving the search information input by the user, determining the corresponding knowledge type to be searched according to the search information, the knowledge type includes one or more of business, responsibility, authority, process, and specification; Determine the knowledge results corresponding to the knowledge type to be retrieved in a preset knowledge result library, and output the search results to the user based on the knowledge results, wherein the knowledge results include one or more of knowledge entities, knowledge graphs, knowledge documents, knowledge indexes, and knowledge change history lists; The knowledge result database is obtained by the following method: Determining the correlation between the documents to be extracted, and constructing a document relationship graph based on the documents to be extracted and the correlation between the documents to be extracted; Determining a subgraph to be analyzed in the document relationship graph; Performing knowledge extraction on the documents in the sub-graph according to the knowledge type to obtain knowledge results corresponding to the knowledge type; Establishing the knowledge result database according to the knowledge results corresponding to the knowledge type; For any two documents to be extracted and , use the following formula to determine the correlation between the documents to be extracted: in, Indicates the document to be extracted and The correlation between Indicates the relevance of business concepts, represents the chapter-level citation relevance, Indicates the chapter and paragraph level citation relevance, Indicates the semantic level relevance, Indicates word-level relevance; Represents the business concept association weight, represents the chapter-level citation association weight, Indicates the chapter and paragraph level citation association weight, represents the semantic level association weight, represents the word-level association weight; Indicates the document to be extracted The document type distribution vector, Indicates the document to be extracted The document type distribution vector of ; Indicates the document to be extracted The number of core words in Indicates the document to be extracted The number of core words in Indicates the document to be extracted and The number of identical core words in Indicates the document to be extracted The corresponding word vector, Indicates the document to be extracted The corresponding word vector; Indicates the document to be extracted Reference the document to be extracted in the chapter paragraph The number of times, represents the correlation constant; Indicates that the core vocabulary is in the document to be extracted The frequency of occurrence in Indicates that the core vocabulary is in the document to be extracted The frequency of occurrence in Indicates the document to be extracted and The frequency of the same core words in the text.

2. The method according to claim 1, characterized in that The document type distribution vector of the document to be extracted is obtained in the following way: Determine the document type corresponding to each of the documents to be extracted, and determine the document type distribution vector corresponding to each of the documents to be extracted based on the document type, wherein the document type includes one or more of system management, training management, business management, operational operations, business operations, and rules and regulations.

3. The method according to claim 2, characterized in that The determining of the document type corresponding to each of the to-be-extracted documents, and determining the document type distribution vector corresponding to each of the to-be-extracted documents according to the document type, specifically includes: Determining a type descriptor corresponding to each of the documents to be extracted based on a preset type dictionary; determining a document type corresponding to each of the documents to be extracted based on a correspondence between the preset type descriptors and the document types; and performing normalization processing on the document types to determine a document type distribution vector corresponding to each of the documents to be extracted; or, Each of the documents to be extracted is classified using multiple binary classifiers to obtain a classification probability value for each of the documents to be extracted; the document type of each of the documents to be extracted is determined based on the type corresponding to the maximum classification probability value, and the document type distribution vector corresponding to each of the documents to be extracted is determined based on the classification probability value of each of the documents to be extracted.

4. The method according to claim 3, characterized in that The vocabulary and core vocabulary of the document to be extracted are obtained in the following way: Performing word segmentation, syntactic analysis, and dependency analysis on each of the documents to be extracted in order to determine the vocabulary corresponding to each of the documents to be extracted; According to a preset core word library, core words in the vocabulary corresponding to each of the documents to be extracted are determined, wherein the core word library is obtained according to one or more of a business domain dictionary, a system function set, and a function menu set.

5. The method according to any one of claims 1 to 4, characterized in that Determining the subgraph to be analyzed in the document relationship graph specifically includes: Determining a first core document in the document relationship graph; Deleting documents in the document relationship graph whose relevance to the first core document is less than a preset relevance threshold; The sub-graph to be analyzed is determined according to the remaining first document in the document relationship graph.

6. The method according to claim 5, characterized in that The step of extracting knowledge from the documents in the sub-graph according to the knowledge type to obtain knowledge results corresponding to the knowledge type specifically includes: Determining a second core document in the sub-graph to be analyzed, and a second document other than the second core document, wherein the first document includes the second core document and the second document; Performing knowledge extraction on the second core documents, and performing knowledge extraction on each of the second documents in descending order of relevance; The extraction method for each first document is: Parsing the first document to obtain key features corresponding to the first document and context information of each key feature, wherein the key features include one or more of a document title, a section title, a section relationship, a paragraph, a chart, and a list; Identifying the first document in chapter units to determine the chapters in the first document that correspond to the knowledge type; performing knowledge extraction on each of the chapters, the key features corresponding to the chapters, and the context information corresponding to the key features in sequence to obtain a knowledge result corresponding to the first document; The knowledge results of each of the first documents are fused according to the knowledge type to obtain a knowledge result corresponding to the knowledge type.

7. The method according to claim 6, characterized in that The fusing of the knowledge results of each of the first documents according to the knowledge type to obtain the knowledge results corresponding to the knowledge type specifically includes: fusing the knowledge results of each of the first documents according to the knowledge type to obtain a first knowledge result corresponding to the knowledge type; Determining whether a second knowledge result corresponding to the knowledge type already exists in the knowledge result database; If so, the first knowledge result and the second knowledge result are merged according to the upload time of the document and / or the generation time of the knowledge result to obtain the knowledge result corresponding to the knowledge type.

8. The method according to claim 7, characterized in that The establishing of the knowledge result database according to the knowledge result corresponding to the knowledge type specifically includes: Storing the knowledge result corresponding to each of the first documents, the first knowledge result, the second knowledge result, and the knowledge result corresponding to the knowledge type in the knowledge result library in a preset storage format, wherein the storage format includes one or more of a knowledge entity, a knowledge graph, a knowledge document, a knowledge index, and a knowledge change history list; Among them, the knowledge entity includes one or more of the knowledge definition, the name of the document where the knowledge is located, and the location of the knowledge in the document; the knowledge change history list includes one or more of the knowledge extraction time, the related document release time, and the knowledge results corresponding to the extraction time.

9. A terminal device comprising: A receiving module is configured to, after receiving search information input by a user, determine a corresponding knowledge type to be searched based on the search information, wherein the knowledge type includes one or more of business, responsibility, authority, process, and specification; A processing module is used to determine the knowledge results corresponding to the knowledge type to be retrieved in a preset knowledge result library, and output the search results to the user based on the knowledge results, wherein the knowledge results include one or more of knowledge entities, knowledge graphs, knowledge documents, knowledge indexes, and knowledge change history lists; the knowledge result library is obtained by: determining the association between documents to be extracted, and constructing a document relationship graph based on the association between the documents to be extracted and the association between the documents to be extracted; and determining a subgraph to be analyzed in the document relationship graph; Performing knowledge extraction on the documents in the sub-graph according to the knowledge type to obtain knowledge results corresponding to the knowledge type; Establishing the knowledge result database according to the knowledge results corresponding to the knowledge type; For any two documents to be extracted and , use the following formula to determine the correlation between the documents to be extracted: in, Indicates the document to be extracted and The correlation between Indicates the relevance of business concepts, represents the chapter-level citation relevance, Indicates the chapter and paragraph level citation relevance, Indicates the semantic level relevance, Indicates word-level relevance; Represents the business concept association weight, represents the chapter-level citation association weight, Indicates the chapter and paragraph level citation association weight, represents the semantic level association weight, represents the word-level association weight; Indicates the document to be extracted The document type distribution vector, Indicates the document to be extracted The document type distribution vector of ; Indicates the document to be extracted The number of core words in Indicates the document to be extracted The number of core words in Indicates the document to be extracted and The number of identical core words in Indicates the document to be extracted The corresponding word vector, Indicates the document to be extracted The corresponding word vector; Indicates the document to be extracted Reference the document to be extracted in the chapter paragraph The number of times, represents the correlation constant; Indicates that the core vocabulary is in the document to be extracted The frequency of occurrence in Indicates that the core vocabulary is in the document to be extracted The frequency of occurrence in Indicates the document to be extracted and The frequency of the same core words in the text.

10. An electronic device comprising a processor and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 8 when executed by a processor.

12. A computer program product comprising a computer program, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images