A knowledge graph-based question-answering method, system, device and medium for land use approval

Abstract

The present application relates to the technical field of knowledge graph construction, and in particular to a knowledge graph-based question-and-answer method, system, device and medium for land approval, which obtains existing policy documents, establishes a relationship network between policy documents based on the existing policy documents, and obtains a land approval knowledge ontology; obtains policy documents to be extracted, extracts transaction topics from the extracted policy documents based on the land approval knowledge ontology, and obtains an approval process knowledge subgraph; extracts special information from the extracted policy documents based on the land approval knowledge ontology, and obtains a policy document knowledge subgraph; obtains approval problem information, analyzes the approval problem information based on the approval process knowledge subgraph and the policy document knowledge subgraph, obtains approval answers, constructs a high-quality land approval knowledge graph and intelligent question-and-answer, and improves the completion efficiency of land approval business.

Description

Technical Field

[0001] The present application relates to the technical field of knowledge graph construction, and in particular to a knowledge graph-based question-answering method, system, device and medium for land use approval. Background Art

[0002] Driven by the trend of digitalization, it has become inevitable to comprehensively reshape natural resource data and business. As a core task of natural resource management and a key link in information construction, land approval has accumulated a large amount of business data in recent years. In order to improve the digital level of approval business, the information integration of land data is being promoted, and a large number of land approval information management systems and platforms have been developed.

[0003] At present, the land use approval business still generally has many problems such as poor interactivity, weak system practicality, and difficulty in data sharing and integration, which have created obstacles for its collaborative management. Specifically, the existing approval business still adopts the strategy of "layered classification" expression for massive multimodal land data, and the business is managed in a fragmented manner, resulting in high difficulty and low efficiency in data and business integration; at the same time, the land use approval entity and its related business information are constantly developing and changing. These information fragments cannot run through their entire life cycle in time and space, resulting in difficulty in backtracking, and the relevant management knowledge and experience cannot be well inherited. Therefore, there is an urgent need to comprehensively sort out the land use approval business process to ensure the integrity and continuity of land use approval in space and time.

[0004] However, due to the problems of diverse data modalities and complex logical relationships in land use approval business, a single and one-sided information extraction method can no longer meet the needs of diversified information extraction in land use approval business. In addition, existing research has a shallow understanding of the complexity of natural resource management business, resulting in only shallow service forms such as storage, retrieval, and visual analysis in practical applications, which affects the quality and completion efficiency of land use approval business. The above problems need to be solved. Summary of the invention

[0005] In order to build a high-quality knowledge graph and intelligent question-answering system for land use approval and improve the efficiency of land use approval business, this application provides a knowledge graph-based question-answering method, system, device and medium for land use approval, using the following technical solutions:

[0006] In the first aspect, the present application provides a knowledge graph-based question-answering method for land use approval, including:

[0007] Obtain existing policy documents, establish a relationship network between policy documents based on the existing policy documents, and obtain the knowledge ontology of land use approval;

[0008] Obtain the policy documents to be extracted, extract the transaction topics of the policy documents to be extracted based on the land use approval knowledge ontology, and obtain the approval process knowledge subgraph;

[0009] Based on the knowledge ontology of land use approval, the thematic information of the extracted policy documents is extracted to obtain the knowledge subgraph of the policy documents;

[0010] Obtain approval problem information, analyze the approval problem information according to the approval process knowledge subgraph and the policy document knowledge subgraph, and obtain the approval answer.

[0011] Preferably, the transaction subject extraction includes:

[0012] The text data of the policy document to be extracted is semantically embedded and the text data is converted into a vector with semantic information.

[0013] Preferably, it also includes:

[0014] The vector generated by semantic embedding is subjected to dimensionality reduction and semantic clustering to obtain group text data containing topics and elements respectively.

[0015] Preferably, it also includes:

[0016] The group text data is divided into sentences and word frequency mining is performed to obtain the approval process knowledge subgraph.

[0017] Preferably, the thematic information extraction includes:

[0018] Convert policy documents into image data and extract framework information from the image data.

[0019] Preferably, it also includes:

[0020] Extract content information from image data;

[0021] Based on the framework information extraction and content information, the policy document knowledge subgraph is obtained.

[0022] Preferably, the specific steps of analyzing the approval problem information according to the approval process knowledge subgraph and the policy document knowledge subgraph are:

[0023] Identify keywords related to the approval business in the approval problem information;

[0024] The text search engine associated with the approval process knowledge subgraph and the policy document knowledge subgraph;

[0025] Enter keywords into the text search engine to search and get the approval answer.

[0026] In the second aspect, the present application provides a knowledge graph question-answering system for land use approval, including:

[0027] The existing document acquisition module is used to acquire existing policy documents, establish a relationship network between policy documents based on the existing policy documents, and obtain the knowledge ontology of land use approval;

[0028] The first knowledge graph construction module is used to obtain the policy documents to be extracted, extract the transaction topics of the policy documents to be extracted based on the land use approval knowledge ontology, and obtain the approval process knowledge subgraph;

[0029] The second knowledge graph construction module is used to extract thematic information from the extracted policy documents based on the land use approval knowledge ontology to obtain the policy document knowledge subgraph;

[0030] The question-and-answer module is used to obtain approval question information, analyze the approval question information according to the approval process knowledge subgraph and the policy document knowledge subgraph, and obtain the approval answer.

[0031] In the third aspect, the present application provides a knowledge graph-based question and answer device for land use approval, comprising a memory and a processor, the memory storing a computer program, the processor being configured to run the computer program to execute the knowledge graph-based question and answer method for land use approval as described above.

[0032] In a fourth aspect, the present application provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the knowledge graph-based question-and-answer method for land use reporting as described above when running.

[0033] In summary, compared with the prior art, the technical solution provided by this application has at least the following beneficial effects:

[0034] This application develops a land approval knowledge ontology through existing policy documents, and obtains the approval process knowledge subgraph and policy document knowledge subgraph based on the land approval knowledge ontology. The approval problem information is analyzed through the approval process knowledge subgraph and the policy document knowledge subgraph to obtain the approval answer. It comprehensively considers the inherent complex characteristics of business data, provides unified modeling and generative information extraction technology, improves the quality of land approval knowledge graph construction, broadens the response capabilities to business-related issues, and improves the completion efficiency of land approval business. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a flow chart of a knowledge graph-based question-answering method for land use approval described in an embodiment of the present application.

[0036] Figure 2 It is a design diagram of the knowledge ontology model layer of land use approval described in the embodiment of this application.

[0037] Figure 3It is a technical roadmap for constructing a policy document knowledge subgraph as described in the embodiment of this application.

[0038] Figure 4 It is a question-answering flowchart of the knowledge graph and large language model described in the embodiment of the present application.

[0039] Figure 5 It is a structural diagram of a knowledge graph-based question-answering system for land use reporting described in an embodiment of the present application.

[0040] Description of reference numerals:

[0041] 1. Existing file acquisition module; 2. First knowledge graph construction module; 3. Second knowledge graph construction module; 4. Question and answer module. DETAILED DESCRIPTION

[0042] The following combination Figure 1-Figure 5 The present application is described in further detail. The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments and are not intended to be limiting.

[0043] Reference Figure 1 , a knowledge graph question-answering method for land use approval involved in this application specifically includes:

[0044] Step S1: Obtain existing policy documents, establish a relationship network between policy documents based on the existing policy documents, and obtain the knowledge ontology of land use approval;

[0045] Step S2: Obtain the policy document to be extracted, extract the transaction subject of the policy document to be extracted based on the land use approval knowledge ontology, and obtain the approval process knowledge subgraph;

[0046] Step S3: extracting thematic information from the policy documents to be extracted based on the land use approval knowledge ontology to obtain a policy document knowledge subgraph;

[0047] Step S4: Acquire approval problem information, analyze the approval problem information according to the approval process knowledge subgraph and the policy document knowledge subgraph, and obtain the approval answer.

[0048] Specifically, the knowledge graph is a graph-based data structure, which is essentially a semantic network that reveals the relationship between entities. Knowledge graph technology can be used to formally describe the concepts, entities, attributes and their relationships in land management work, so that concepts and entities can be interconnected to form a network knowledge structure, which is convenient for the development of land management work, thereby realizing multi-source and multi-factor business collaborative governance. The "one code to manage land" association model based on the construction of knowledge graph technology provides a new way to realize the association of the entire life cycle of land. In addition, incorporating knowledge graph technology into the construction land approval process to automatically generate auxiliary approval opinions shows potential value in improving approval efficiency and management effectiveness. Knowledge graph technology can play a relatively important role in the intelligence and efficiency improvement of land approval.

[0049] This application uses the actual approval process and existing policy documents as references to comprehensively sort out the land approval process, the departments and powers involved, the policies, etc., establish a relationship network between it and the functional departments, land approval business, and related policies, and use existing policy documents to formulate a land approval knowledge ontology. Based on the land approval knowledge ontology, the approval process knowledge subgraph and the policy document knowledge subgraph are constructed. The approval process knowledge subgraph and the policy document knowledge subgraph are knowledge graphs. Through the analysis of the knowledge graph, the approval answer is obtained. It comprehensively considers the inherent complex characteristics of business data, provides unified modeling and generative information extraction technology, improves the quality of land approval knowledge graph construction, and broadens the response capabilities to business-related issues, and improves the completion efficiency of land approval business.

[0050] Among them, referring to the figure, the knowledge expression model of land use approval event is converted into a graph structure to construct the land use approval knowledge graph. The concept layer can be represented as the ontology in the land use approval knowledge graph. The top-level knowledge ontology is "land use approval event", which contains two categories, "policy documents" and "approval process", and the land use approval knowledge ontology model layer is obtained.

[0051] As one implementation method, the transaction subject extraction includes:

[0052] The text data of the policy document to be extracted is semantically embedded and the text data is converted into a vector with semantic information.

[0053] Specifically, under the guidance of the land use approval knowledge ontology, the BertTopic topic modeling technology is used to extract transaction topics using the process of "semantic embedding-dimensionality reduction-semantic clustering-word segmentation-word frequency mining" to construct a knowledge subgraph of the approval process.

[0054] Based on the collected land use approval policy document information, the PaddleOCR model is used to obtain the text data corresponding to scanned materials and other materials. Semantic embedding is performed using semantic vector representation methods such as SBert, which converts the text data into a vector representation with semantic information, which is convenient for subsequent input into the computer for analysis and processing.

[0055] As one implementation method, the vector generated by semantic embedding is subjected to dimensionality reduction processing and semantic clustering to obtain group text data containing topics and elements respectively. The group text data is divided into sentences and word frequency mining is performed to obtain the approval process knowledge subgraph.

[0056] Specifically, the high-dimensional vector generated by semantic embedding is subjected to dimensionality reduction processing, and semantic clustering is performed using the text vector similarity measurement model and the HDBSCAN hierarchical clustering model to obtain text data consisting of different groups containing related topics and elements. Then, the jieba word segmentation tool is used to divide the sentences for subsequent word frequency mining.

[0057] Based on semantic clustering and word segmentation, the C-TF-IDF word frequency index is used to mine the subject information of the land approval process, and the terms related to the land approval business are mined. They are further screened and organized into appropriate process elements through manual work, and finally refined and filled into knowledge related to the approval process.

[0058] As one implementation method, the topic information extraction includes:

[0059] Convert policy documents into image data and extract framework information from the image data.

[0060] Specifically, referring to the figure, under the guidance of the land use approval knowledge ontology, multimodal information extraction model methods such as UIE, UIE-X and Adobe PDFExtract API are used to construct an automatic extraction pipeline for policy document thematic information, related information and clause content information to form a policy document knowledge subgraph.

[0061] The specific steps are to use tool libraries such as PyMuPDF to loop through documents in Word, PDF and other formats and automatically convert them into multiple single-page jpg images so that the UIE-X multimodal document extraction model can read the data and extract relevant information. The UIE-X multimodal document extraction model is trained based on the "small amount of annotation-fine-tuning" method to extract the framework information from the policy documents that have been converted into images, including five categories: document title, document number, issuing department, publishing time, and clause subtitles.

[0062] As one implementation mode, it also includes:

[0063] Extract content information from image data;

[0064] Based on the framework information extraction and content information, the policy document knowledge subgraph is obtained.

[0065] Specifically, the structure of the policy document PDF is analyzed based on Adobe PDF Extract API, the policy document PDF is divided into different regions, and each region and its content are organized and extracted according to the designed regular extraction rules. On this basis, the document content information is reorganized and the content positioning is achieved by comparing the relative positions of the extracted document framework information and the document content information.

[0066] Based on the "small amount of annotation-fine-tuning" method, the UIE text extraction model is trained to obtain the land approval related words in the subtitles and corresponding contents of the policy documents, including departments and institutions, policies and regulations, and approval matters. The policy clause subgraphs in different documents are linked according to the semantic matching relationship of the related words to form a policy document knowledge subgraph. The constructed policy document knowledge subgraph will also be associated with the approval process subgraph through the same semantic matching method, thereby completing the connection of knowledge triples and the construction of the land auxiliary approval knowledge graph.

[0067] As one implementation method, the specific steps of analyzing the approval problem information according to the approval process knowledge subgraph and the policy document knowledge subgraph are as follows:

[0068] Identify keywords related to the approval business in the approval problem information;

[0069] The text search engine associated with the approval process knowledge subgraph and the policy document knowledge subgraph;

[0070] Enter keywords into the text search engine to search and get the approval answer.

[0071] Specifically, refer to Figure 4 Based on the constructed knowledge graph, an intelligent question-answering framework for auxiliary land approval is designed. Through the steps of "question analysis-related knowledge retrieval-prompt and answer generation", auxiliary approval answers that conform to the terminology habits and accuracy requirements in the approval field are obtained.

[0072] Accept user input and identify keywords that are highly relevant to the approval business. The parsed keywords will be used to construct Query DSL query statements and finally input into the Elasticsearch text search engine to retrieve semantically related entities.

[0073] By associating entities stored in the knowledge graph with the corresponding documents in the Elasticsearch engine based on the same id, the module can return relevant information about entities similar to the top-k keywords and their one-hop subgraphs after semantic retrieval, and input them into the large language model module as a prompt component.

[0074] The original question information and entity retrieval information provided in the first two steps are integrated and used as prompts to input into the large language model to generate auxiliary approval answers that are both accurate and versatile. The large language model uses the domestic open source ChatGLM-6B series model as its basis.

[0075] Reference Figure 5 , a knowledge graph question-answering system for land use approval is provided for the embodiment of the present application, the system comprising:

[0076] Existing document acquisition module 1 is used to acquire existing policy documents, establish a relationship network between policy documents based on the existing policy documents, and obtain the knowledge ontology of land use approval;

[0077] The first knowledge graph construction module 2 is used to obtain the policy documents to be extracted, extract the transaction topics of the policy documents to be extracted based on the land use approval knowledge ontology, and obtain the approval process knowledge subgraph;

[0078] The second knowledge graph construction module 3 is used to extract thematic information from the extracted policy documents based on the land use approval knowledge ontology to obtain the policy document knowledge subgraph;

[0079] The question-and-answer module 4 is used to obtain approval question information, analyze the approval question information according to the approval process knowledge subgraph and the policy document knowledge subgraph, and obtain the approval answer.

[0080] An embodiment of the present application provides a knowledge graph-based question and answer device for land use approval reports, including a memory and a processor, the memory storing a computer program, and the processor being configured to run the computer program to execute the knowledge graph-based question and answer method for land use approval reports as described above.

[0081] An embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the knowledge graph-based question-answering method for land use reporting as described above when running.

[0082] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device and product can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0083] In several embodiments provided in this application, it should be understood that the disclosed methods, systems, devices and program products may be implemented in other ways.

[0084] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0085] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A knowledge graph question answering method for land use approval, characterized in that: include: Obtain existing policy documents, establish a relationship network between policy documents based on the existing policy documents, and obtain the knowledge ontology of land use approval; Obtain the policy documents to be extracted, extract the transaction topics of the policy documents to be extracted based on the land use approval knowledge ontology, and obtain the approval process knowledge subgraph; Based on the knowledge ontology of land use approval, the extracted policy documents are subjected to thematic information extraction, wherein the thematic information extraction includes converting the policy documents into image data, extracting framework information from the image data, and using the PyMuPDF tool library to loop through Word and PDF format documents and automatically convert them into multiple single-page jpg images, and extract framework information from the policy documents that have been converted into images based on the UIE-X multimodal document extraction model, including file title, file number, issuing department, issuing time and clause subtitles; extract content information from the image data; obtain the policy document knowledge subgraph based on the framework information extraction and content information, and obtain the policy document knowledge subgraph, specifically by structurally analyzing the policy document PDF, dividing the policy document PDF into different regions, and organizing and extracting each region and its content based on the designed regular extraction rules, obtaining the land use approval related words in the policy document subtitle and the corresponding content, and associating the policy clause subgraphs in different documents based on the semantic matching relationship of the related words to form the policy document knowledge subgraph; Obtain approval problem information, analyze the approval problem information according to the approval process knowledge subgraph and the policy document knowledge subgraph, and obtain the approval answer.

2. The knowledge graph question-answering method for land use approval according to claim 1 is characterized in that: The transaction subject extraction includes: The text data of the policy document to be extracted is semantically embedded and the text data is converted into a vector with semantic information.

3. The knowledge graph question-answering method for land use approval according to claim 2 is characterized in that: Also includes: The vector generated by semantic embedding is subjected to dimensionality reduction and semantic clustering to obtain group text data containing topics and elements respectively.

4. The knowledge graph question answering method for land use approval according to claim 3 is characterized in that: Also includes: The group text data is divided into sentences and word frequency mining is performed to obtain the approval process knowledge subgraph.

5. The knowledge graph question-answering method for land use approval according to claim 1 is characterized in that: The specific steps of analyzing the approval problem information according to the approval process knowledge subgraph and the policy document knowledge subgraph are as follows: Identify keywords related to the approval business in the approval problem information; The text search engine associated with the approval process knowledge subgraph and the policy document knowledge subgraph; Enter keywords into the text search engine to search and get the approval answer.

6. A knowledge graph question answering system for land use approval, characterized in that: include: The existing document acquisition module is used to acquire existing policy documents, establish a relationship network between policy documents based on the existing policy documents, and obtain the knowledge ontology of land use approval; The first knowledge graph construction module is used to obtain the policy documents to be extracted, extract the transaction topics of the policy documents to be extracted based on the land use approval knowledge ontology, and obtain the approval process knowledge subgraph; The second knowledge graph construction module is used to extract thematic information from the extracted policy documents based on the land use approval knowledge ontology, wherein the thematic information extraction includes converting the policy documents into image data and extracting framework information from the image data. The specific steps are to use the PyMuPDF tool library to loop through Word and PDF format documents and automatically convert them into multiple single-page jpg images, and extract the framework information from the policy documents that have been converted into images based on the UIE-X multimodal document extraction model, including the file title, file number, issuing department, issuing time and clause subtitles; extract content information from the image data; obtain the policy document knowledge subgraph based on the framework information extraction and content information, and obtain the policy document knowledge subgraph, specifically by structurally analyzing the policy document PDF, dividing the policy document PDF into different regions, and organizing and extracting each region and its content based on the designed regular extraction rules, obtaining the land use approval related words in the policy document subtitle and the corresponding content, and associating the policy clause subgraphs in different documents based on the semantic matching relationship of the related words to form a policy document knowledge subgraph; The question-and-answer module is used to obtain approval question information, analyze the approval question information according to the approval process knowledge subgraph and the policy document knowledge subgraph, and obtain the approval answer.

7. A knowledge graph question-answering device for land use approval, characterized in that: It includes a memory and a processor, the memory stores a computer program, and the processor is configured to run the computer program to execute the knowledge graph question-answering method for land use reporting described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program is configured to execute the knowledge graph question-answering method for land use reporting as described in any one of claims 1 to 5 when running.

Images