Knowledge graph-based nuclear industry law and regulation database construction method

Abstract

The invention belongs to the field of information technology and nuclear energy industry, and particularly relates to a nuclear industry law and regulation database construction method based on a knowledge graph, and the method comprises the steps: 1, carrying out the automatic collection of nuclear industry laws and regulations; 2, constructing a nuclear industry law and regulation knowledge graph; and step 3, constructing an intelligent application model based on the nuclear industry law and regulation knowledge graph. According to the method, the accuracy, the traceability and the usability of the nuclear industry law and regulation data are effectively improved, and the problems of an existing nuclear industry law and regulation database in the aspects of data management, query efficiency and updating maintenance are solved; and through intelligent query and analysis functions, a user is helped to more effectively understand and apply laws and regulations of the nuclear industry, and safe and compliant development of the nuclear industry is supported.

Description

Technical Field

[0001] The present invention belongs to the fields of information technology and nuclear energy industry, and particularly relates to a method for constructing a nuclear industry laws and regulations database based on a knowledge graph. Background Art

[0002] With the rapid development of the nuclear energy industry, the importance of nuclear safety and laws and regulations has become increasingly prominent. The number of laws and regulations involved in the nuclear industry is huge and updated frequently, which brings challenges to nuclear industry practitioners in obtaining, managing, and applying these laws and regulations. Traditional laws and regulations databases have problems such as data islands, untimely updates, and low query efficiency, and it is difficult to meet the current nuclear industry's requirements for laws and regulations information processing.

[0003] Existing nuclear industry laws and regulations databases mostly adopt traditional relational databases for construction, and this structure has limitations in processing complex queries, data association analysis, etc. At the same time, with the rise of big data and artificial intelligence technologies, especially the development of knowledge graph technology, it provides new ideas and methods for the construction of nuclear industry laws and regulations databases. A knowledge graph can graphically represent entities, concepts, attributes, and their relationships in laws and regulations, thereby improving the readability, queryability, and intelligent level of data.

[0004] Knowledge graph technology can extract useful information from unstructured regulation texts through natural language processing (NLP) technologies such as entity recognition and relationship extraction, and construct a structured knowledge graph. Such a graph can not only provide accurate retrieval of laws and regulations, but also conduct association analysis between regulations, providing support for nuclear industry decision-making. In addition, the dynamic update mechanism of the knowledge graph can ensure the timeliness of the database content, which is particularly important for the nuclear industry where laws and regulations change rapidly.

[0005] Therefore, it is urgent to use knowledge graph technology to construct an efficient and intelligent nuclear industry laws and regulations database to solve the deficiencies of existing databases in data processing and query, and improve the management efficiency and application level of nuclear industry laws and regulations. Summary of the Invention

[0006] The object of the present invention is to provide a method for constructing a nuclear industry laws and regulations database based on a knowledge graph. By constructing a database based on the knowledge graph, this method provides a laws and regulations management platform that can be updated in real time, queried quickly, and analyzed intelligently. This platform can integrate a large amount of laws and regulations data, extract key information, and establish the association relationships between entities, thereby improving the accuracy, traceability, and usability of nuclear industry laws and regulations data, and effectively solving the problems existing in the existing nuclear industry laws and regulations database in terms of data management, query efficiency, and update and maintenance. In addition, this method also helps users understand and apply the laws and regulations of the nuclear industry more effectively through intelligent query and analysis functions, and supports the safe and compliant development of the nuclear energy industry.

[0007] The technical solution for realizing the object of the present invention:

[0008] A method for constructing a nuclear industry laws and regulations database based on a knowledge graph, the method comprising:

[0009] Step 1, automatic collection of nuclear industry laws and regulations;

[0010] Step 2, constructing a knowledge graph of nuclear industry laws and regulations;

[0011] Step 3, constructing an intelligent application model based on the knowledge graph of nuclear industry laws and regulations.

[0012] The said Step 1 includes: investigating and studying the data types and field types of laws and regulations published on government websites, industry association publications, and legal databases, and collecting nuclear industry laws and regulations text data; inputting, modifying, querying, viewing, and exporting the laws and regulations text data; storing the extracted and maintained structured laws and regulations information into the laws and regulations database for subsequent querying, viewing, knowledge graph construction, and intelligent application of laws and regulations information.

[0013] The said Step 2 includes:

[0014] Data preprocessing: performing data cleaning on nuclear industry laws and regulations: using regular expressions to remove punctuation marks and stop words in the text, and removing abnormal fragments in the text; constructing a domain thesaurus, adding the domain thesaurus to the word segmentation tool, and performing word segmentation operations on the cleaned text data using a natural language word segmentation tool;

[0015] Natural language processing: applying the word2vec model to obtain a word embedding matrix; representing the segmented text in the form of a one-hot encoded vector and multiplying it by the word embedding matrix to obtain the corresponding word vector; applying Bi-LSTM+CRF for sequence labeling to obtain the labeling of named entities and entity attributes; applying Bi-LSTM+ATT for relationship extraction to obtain the relationships between entities for the embedding of the knowledge graph.

[0016] Knowledge Graph Embedding: Integrate data using the TransE or DistMult model to achieve data storage with a unified semantic data structure.

[0017] The data includes unstructured data, semi-structured data, and structured data.

[0018] The domain thesaurus includes a nuclear industry laws and regulations professional thesaurus, a general thesaurus, and a synonym thesaurus.

[0019] Step 3 includes: constructing an accurate semantic retrieval model for nuclear industry laws and regulations, constructing an intelligent push model for nuclear industry laws and regulations, constructing a knowledge Q&A model for nuclear industry laws and regulations, and constructing a comparative analysis model for domestic and foreign nuclear industry laws and regulations.

[0020] Specifically, the construction of the accurate semantic retrieval model for nuclear industry laws and regulations is as follows:

[0021] Step A1, Query Processing: Process the query input by the user, perform word segmentation and stop word removal operations; represent the query input using the same semantic modeling model as used for constructing the laws and regulations thesaurus and semantic library.

[0022] Step A2, Graph Traversal: Based on the semantic representation of the query, search in the knowledge graph through a graph traversal algorithm.

[0023] Step A3, Similarity Calculation: Use a similarity calculation method to measure the semantic similarity between the query and the laws and regulations text.

[0024] Step A4, Sorting and Retrieval: According to the results of the similarity calculation, based on a similarity-based sorting algorithm, sort the laws and regulations text so that the most relevant text is ranked at the front.

[0025] Step A5, Interface Interaction: Design a user-friendly interface that enables users to input queries and view retrieval results.

[0026] Step A6, Evaluation and Optimization: Evaluate the retrieval framework, use some standard test sets or manual evaluation to measure the accuracy and relevance of the retrieval results, and optimize according to the evaluation results.

[0027] Specifically, the construction of the intelligent push model for nuclear industry laws and regulations is as follows:

[0028] Step B1, Identify Named Entities in the Input: According to relevant user characteristics and business scenarios, extract key entities based on the input fields.

[0029] Step B2, Entity Disambiguation and Entity Linking: Disambiguate ambiguous entities, and after completing entity disambiguation, link the extracted entities to the corresponding entities in the knowledge base.

[0030] Step B3. Apply TransE to calculate the embedded vectors of the input triples: Construct the input triples based on the filled information; for the input triples, apply the TransE model to transform them into embedded vectors.

[0031] Step B4. Match with the triples in the knowledge graph to obtain the recommended content: Compare and match the obtained triples with the corresponding triples in the knowledge graph, and select the recommended result with the highest matching degree for output.

[0032] Step B5. Adjust the obtained recommended results in combination with specific matching rules: In the previous step, the machine learning algorithm automatically learns the recommendation degree of each historical data according to the knowledge in the knowledge graph, that is, the vector cosine matching value. The larger this value is, the higher the recommendation degree, that is, the matching degree, and this value is marked as score1; in this step, the algorithm will calculate a recommendation degree score for each historical record again according to the set rules and record it as score2; finally, the final recommendation degree score will be obtained according to the following calculation formula:

[0033] score final = α * score1 + (1 - α)score2

[0034] Among them, α is a number greater than 0 and less than 1, used to control the weights of the two parts of the scores. This value will be obtained through multiple experiments and assisted by artificial experience during the algorithm implementation; score2 is obtained according to different rules in different application modules.

[0035] The specific construction of the nuclear industry laws and regulations knowledge Q&A model is as follows:

[0036] Step C1. Data collection and knowledge graph construction: Collect relevant laws and regulations data and construct a knowledge graph.

[0037] Step C2. Entity recognition and relationship extraction: Use natural language processing technology to perform entity recognition and relationship extraction on the input questions.

[0038] Step C3. Question parsing and matching: Parse the questions raised by users, extract the key information therein, and match them with the entities and relationships in the knowledge graph.

[0039] Step C4. Answer generation and display: Generate corresponding answers according to the matching results and display them to the front end.

[0040] Step C5. User interaction and feedback: Provide a user interaction interface to allow users to further ask questions or provide feedback.

[0041] The specific construction of the domestic and foreign nuclear industry laws and regulations comparative analysis model is as follows:

[0042] Step D1, entity recognition and relationship extraction: Use natural language processing techniques to perform entity recognition and relationship extraction on the collected laws and regulations texts.

[0043] Step D2, constructing a knowledge graph: Collect relevant data on domestic and international nuclear industry laws and regulations, and construct a knowledge graph of domestic and international nuclear industry laws and regulations.

[0044] Step D3, relevance analysis algorithm: Design a suitable relevance analysis algorithm for calculating the relevance between laws and regulations, and measure the similarity, correlation or degree of association between laws and regulations by using graph algorithms and similarity calculation methods.

[0045] Step D4, visual chart display: Based on the knowledge graph and the relevance analysis algorithm, conduct relevance analysis of laws and regulations and visually display the results.

[0046] Step D5, result interpretation and application: Interpret the results of the relevance analysis to help users understand the relationships between laws and regulations and apply them as needed.

[0047] The beneficial technical effects of the present invention are as follows:

[0048] A method for constructing a nuclear industry laws and regulations database based on a knowledge graph provided by the present invention. The method for constructing a nuclear industry laws and regulations database based on a knowledge graph realizes rapid retrieval, in-depth analysis and dynamic update of laws and regulations information by integrating and structuring a large amount of unstructured text data, significantly improves the accessibility, accuracy and practicability of nuclear industry laws and regulations data, thereby enhancing the response speed of nuclear industry practitioners to regulatory changes and the efficiency of compliance management. At the same time, it also provides a powerful tool for policymakers, researchers and legal experts to support their decision-making, research analysis and compliance review work. Description of the Drawings

[0049] Figure 1 It is the system architecture diagram of the nuclear industry laws and regulations knowledge graph in the embodiment of the present invention;

[0050] Figure 2 It is the algorithm flow framework diagram of the nuclear industry knowledge graph in the embodiment of the present invention;

[0051] Figure 3 It is the business model diagram of the nuclear industry laws and regulations knowledge graph in the embodiment of the present invention;

[0052] Figure 4 It is the accurate semantic retrieval diagram of the nuclear industry laws and regulations in the embodiment of the present invention;

[0053] Figure 5 It is the intelligent push diagram of the nuclear industry laws and regulations in the embodiment of the present invention;

[0054] Figure 6 This is the Q&A chart of nuclear industry laws and regulations in the embodiments of the present invention. Detailed implementation manners

[0055] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0056] A method for constructing a nuclear industry laws and regulations database based on a knowledge graph provided by the present invention specifically includes the following steps:

[0057] Step 1: Automatic collection of nuclear industry laws and regulations

[0058] Investigate and study the data types, field types, etc. of laws and regulations published by institutions such as government websites, industry association publications, and legal databases. Through the automatic collection module, collect the text data of nuclear industry laws and regulations. The text data includes national laws, administrative regulations, departmental rules, regulatory documents, guiding documents, policy interpretations, notices and announcements, policy documents, etc. related to the nuclear industry. The collected text data covers all aspects of the nuclear industry, such as nuclear safety, nuclear energy utilization, nuclear waste treatment, etc., to ensure the comprehensiveness and accuracy of the knowledge graph.

[0059] Through the nuclear safety laws and regulations management module, functions such as input, modification, query, viewing, and export of the text data of laws and regulations are realized. Store the extracted and maintained structured laws and regulations information in the laws and regulations database for subsequent query and viewing of laws and regulations information, knowledge graph construction, and intelligent applications.

[0060] Step 2: Classification of nuclear industry laws and regulations

[0061] Classify and sort out the collected nuclear industry laws and regulations. As shown in Table 1, they are mainly divided into two categories: central regulations and local regulations.

[0062] Among the central regulations, they are further subdivided into sub-categories such as national laws, administrative regulations, departmental rules, regulatory documents, policy interpretations, notices and announcements, etc. National laws are approved by the National People's Congress and its Standing Committee and have the highest legal effect, such as the Constitution. Administrative regulations are regulatory documents formulated by the State Council according to the Constitution and laws, usually with the document number starting with "Guo Ling". Departmental rules come from various ministries and commissions of the State Council and directly affiliated institutions with administrative management functions and are published in the national regulations library. Regulatory documents include documents issued by the State Council and various departments, while policy interpretations elaborate on the specific content and spirit of the policies. Notices and announcements involve important matters issued by government departments.

[0063] Local regulations include administrative regulations and departmental rules. Among them, administrative regulations are formulated by local state power organs, and departmental rules come from local government websites and bulletins.

[0064] Such classification and collation helps to clearly display the hierarchy and sources of nuclear industry laws and regulations, ensuring the effective implementation and management of laws and regulations.

[0065] Table 1 Data Sources of Nuclear Industry Laws and Regulations

[0066]

[0067] Step 3: Construct the Knowledge Graph of Nuclear Industry Laws and Regulations

[0068] Based on the collected laws and regulations data, using knowledge graph technology, construct the business model of the knowledge graph of nuclear industry laws and regulations, build the thesaurus and semantic library of nuclear industry laws and regulations, sort out the triples of legal entity, relationship and attribute, and form the knowledge base and knowledge system of laws and regulations. The specific implementation methods are as follows:

[0069] (1) Architecture Overview

[0070] The knowledge graph model of nuclear industry laws and regulations transforms the original laws and regulations data that is isolated, has a small amount of information per unit, and a low value density into a higher-value data form through knowledge graph association. The "entity-relationship" data form of the knowledge graph can bring about qualitative changes in products such as data fusion, information retrieval, interactive analysis, and multi-dimensional display in terms of functions, performance, and utility, thereby realizing ultra-depth analysis and application of nuclear industry laws and regulations data.

[0071] Figure 1 Shows the complete system architecture of the knowledge graph of nuclear industry laws and regulations. This system integrates various operations of the knowledge graph from knowledge system design, knowledge graph construction to knowledge reasoning application, forming a systematic and platform-based one-stop platform with data connection and function integration.

[0072] The specific process is as follows:

[0073] 1) Data Processing: Data processing mainly refers to extracting, cleaning, transforming, and standardizing the information in various types of data sources. The main data forms in the data sources include unstructured data, semi-structured data, and structured data.

[0074] Structured data refers to the data from the first-phase database of laws and regulations and the databases of other relevant platforms, and this part of the data can be directly mapped and transformed into the triple data structure of the knowledge graph.

[0075] Semi-structured data refers to the JSON and XML format files downloaded from the APIs provided by relevant laws and regulations websites.

[0076] Unstructured data refers to legal text files such as PDF and Word.

[0077] 2) Knowledge Modeling: Knowledge modeling refers to the way of organizing and representing nuclear industry laws and regulations data in the form of a knowledge graph. The key steps include entity extraction, attribute extraction, and relationship extraction.

[0078] Entity Extraction: Identify entities from nuclear industry laws and regulations data, such as regulation names, institution names, terms, industries, equipment, locations, etc.

[0079] Attribute Extraction: Identify entity attributes from nuclear industry laws and regulations data, such as the release date of regulations, version numbers, regulation issuing agencies, scope of application, etc.

[0080] Relationship Extraction: Extract citation relationships, substitution relationships, application relationships between regulations and the field in nuclear industry regulations, such as "regulations apply to certain types of facilities".

[0081] 3) Text Representation: The text representation of nuclear industry laws and regulations based on the knowledge graph can map the text data of laws and regulations to the entities and relationships in the knowledge graph, thus better integrating structured knowledge and text information. Related algorithms include Bi-LSTM-CRF, Bi-LSTM-ATT, Word2Vec, etc.

[0082] 4) Knowledge Reasoning: The entire data platform needs to aggregate the knowledge storage and computing modules required for knowledge reasoning to provide high-performance and stable services for reading and writing the knowledge graph. Knowledge storage and computing are usually supported by a graph database, a graph algorithm library, a graph computing engine, and a graph reasoning engine. The knowledge storage and computing modules need to integrate data storage and computing components for domesticated environments at the same time. For example, the knowledge graph is very suitable for storing data such as citation relationships, substitution relationships, and attribute relationships of nuclear industry laws and regulations, enabling it to give clear entity results in reasoning applications such as question-and-answer retrieval and association analysis.

[0083] 5) Query Service: The query service provides a visual interface for the knowledge reasoning module and supports knowledge graph queries, including single or batch data services. In terms of deep applications, it is necessary to provide interface capabilities such as statistical queries, conditional queries, fuzzy queries, keyword matching, semantic retrieval, and intelligent question answering.

[0084] 6) Application Scenarios: Build intelligent applications based on the knowledge graph, including semantic retrieval, intelligent recommendation, search question answering, and association analysis. Intelligent applications need to perform data analysis, model training optimization, etc. based on the knowledge graph data, and encapsulate the model deployment and services into recognizable intelligent application capabilities.

[0085] Semantic Retrieval: Based on the semantic associations between entities and relationships in laws and regulations data, combined with natural language processing and graph technology, more accurate retrieved text results can be provided.

[0086] Intelligent Recommendation: Provide a personalized recommendation system based on the semantic associations in the content of query-based laws and regulations data.

[0087] Search and Q&A: Identify keywords, entities, relationships, and constraint conditions in the question, retrieve in the knowledge graph to find entities, relationships, or attributes that match the question. Then combine template generation, natural language generation technology, or text information extracted from the knowledge graph to construct the answer.

[0088] Association Analysis: Use knowledge graph data to identify and analyze the association relationships between laws and regulations entities, and mine the relationship patterns between domestic and foreign laws and regulations.

[0089] (2) Algorithm Process of the Knowledge Graph of Nuclear Industry Laws and Regulations

[0090] The algorithm process framework of the complete knowledge graph of nuclear industry laws and regulations is as Figure 3 shown. For the construction of the knowledge graph of nuclear industry laws and regulations, first analyze the business processes, experience information, and data sources of specific business modules to establish a knowledge model that can represent rich engineering semantics and context; on this basis, develop intelligent algorithms to perform intelligent analysis, extraction, association, and knowledge embedding on relevant data. The specific algorithm part mainly consists of three parts: data preprocessing, natural language processing, and knowledge graph embedding. The following is a detailed description of the algorithms for the three parts.

[0091] 1) Data Preprocessing: Before officially processing the data, first perform data cleaning. Use regular expressions to remove punctuation marks and stop words in the text, and remove abnormal fragments in the text. Then, construct professional thesauruses, including thesauruses of nuclear industry laws and regulations, general thesauruses, and synonym thesauruses. After initially constructing the professional thesauruses, perform word segmentation on the text data. Use natural language word segmentation tools to segment the cleaned text. Add the professional thesauruses and general thesauruses to the word segmentation tool to obtain better word segmentation results.

[0092] In the above work, the two parts of constructing the thesaurus and word segmentation are in a mutually assisting and promoting relationship. In the initial stage, the work of constructing the thesaurus is completed manually. Using the thesaurus can improve the accuracy of the word segmentation work. After that, use statistics such as information entropy to discover new words in the segmented text, and obtain new words not recorded in the thesaurus to further improve the thesaurus.

[0093] 2) Natural Language Processing: Numerical representation of text. Apply the word2vec model to obtain the word embedding matrix. Represent the segmented text in the form of one-hot encoded vectors and multiply it by the word embedding matrix to obtain the corresponding word vectors. The word vectors contain rich semantic information, improving the accuracy of subsequent work. Sequence labeling, including the labeling of named entities and entity attributes. Apply Bi-LSTM+CRF to complete this task and obtain the named entities and entity attributes. Relationship extraction, including relationships between entities such as "reference" and "causing". Apply the Bi-LSTM+ATT model to obtain the relationships between entities for the embedding of the knowledge graph.

[0094] 3) Knowledge Graph Embedding: Apply the TransE or DistMult model to integrate structured and unstructured data and achieve data storage in a unified semantic data structure such as triple RDF form.

[0095] (3) Business Model of the Knowledge Graph of Nuclear Industry Laws and Regulations

[0096] When building the knowledge graph of nuclear industry laws and regulations, each laws and regulations document is usually used as a node, and specific clauses or requirements can be used as attributes or sub-nodes related to these nodes. The relationships between entities are used as edges. The arrow direction indicates that the superior entity points to the inferior entity. The business model of the knowledge graph of nuclear industry laws and regulations Figure 3 is shown as follows. According to the business model of the knowledge graph of nuclear industry laws and regulations, sort out the entities, relationships, and attributes as shown in Table 2-4.

[0097] Table 2 Entity Table of Nuclear Industry Laws and Regulations

[0098]

[0099] Table 1 Attribute Table of Nuclear Industry Laws and Regulations

[0100]

[0101]

[0102] Table 2 Attribute Table of Nuclear Industry Laws and Regulations

[0103]

[0104] Step 4: Intelligent Application Based on the Knowledge Graph Model of Nuclear Industry Laws and Regulations

[0105] (1) Construct a Precise Semantic Retrieval Model for Nuclear Industry Laws and Regulations

[0106] Based on the constructed knowledge graph model and knowledge system of nuclear industry laws and regulations, using technologies such as natural language processing and search engines, the accurate semantic retrieval function of domestic and foreign nuclear industry laws and regulations, nuclear industry information and frontier consultations is realized.

[0107] The accurate semantic retrieval of nuclear industry laws and regulations realizes the accurate semantic retrieval of domestic and foreign nuclear industry laws and regulations. This module includes the semantic retrieval and data display functions of nuclear industry big data. Users first enter the information to be queried in the input box. Through the natural language processing engine, Chinese word segmentation, part-of-speech tagging, stop word processing, feature engineering (vectorization), sentence similarity, and document similarity processing of laws and regulations information and input corpus are realized, and finally the semantic recognition function is realized. Among them, by optimizing the recognition of professional terms in nuclear industry regulations and the processing of synonyms and co-referential pronouns, the comprehensiveness of retrieval is improved. Then use the Solr search engine to accurately retrieve domestic and foreign nuclear industry laws and regulations. Consider the matching degree of each word segment through the matching rule and then comprehensively calculate the matching degree of the entire input statement, and sort according to the matching degree from high to low to avoid repeated screening and display. At the same time, combine functions such as dynamic summary and keyword highlighting to present the retrieved information to users. Set up a secondary screening box, and users can perform secondary screening on the matched information to more accurately locate the information to be queried.

[0108] After semantic modeling of the laws and regulations data, the semantic information of words and sentences is captured. On this basis, a semantic retrieval module is constructed as Figure 5 shown.

[0109] 1) Query processing: Process the query input by the user, including operations such as word segmentation and stop word removal. Use the same semantic modeling model as the construction of the laws and regulations thesaurus and semantic library to represent the query input.

[0110] 2) Graph traversal: Based on the semantic representation of the query, search in the knowledge graph through the graph traversal algorithm. It can start from the entity node or relationship node of the query and traverse along the associated edges to find entities or relationships related to the query.

[0111] 3) Similarity calculation: Use appropriate similarity calculation methods (such as cosine similarity, Jaccard similarity) to measure the semantic similarity between the query and the laws and regulations text. This can help determine the relevance of the text to the query.

[0112] 4) Sorting and retrieval: According to the results of similarity calculation, sort the laws and regulations text so that the most relevant text is ranked first. Sorting algorithms based on similarity can be used, such as inverted index or vector-based sorting methods.

[0113] 5) Interface Interaction: Design a user-friendly interface that enables users to input queries and view retrieval results. The interface can provide functions such as search suggestions, filtering options, and summaries of relevant legal provisions to enhance the user experience.

[0114] 6) Evaluation and Optimization: Evaluate the retrieval framework using some standard test sets or manual evaluation to measure the accuracy and relevance of retrieval results. Optimize according to the evaluation results, which may require adjusting model parameters, improving query processing, or improving sorting algorithms, etc.

[0115] (2) Construct an intelligent push model for nuclear industry laws and regulations

[0116] Based on the constructed nuclear industry laws and regulations knowledge graph model and knowledge system, using technologies such as natural language processing, intelligent recommendation, and machine learning, based on the business needs of nuclear industry personnel and the current laws and regulations information browsed by users, realize the intelligent recommendation function of domestic and foreign nuclear industry laws and regulations.

[0117] Embedded Knowledge Graph (KGE) can represent entities and relationships in a continuous vector space while retaining certain network information. The goal of applying the KGE method is to simplify the processing of the KG while maintaining its structure. In a recommendation system, KGE can be used to enrich the information of entities and attributes, and then the embedded representation can be used to calculate the similarity between the two. The most popular KGE models include TransE, TransH, TransD, and TransR.

[0118] The intelligent recommendation system framework based on embedded knowledge graph technology is as Figure 5 shown.

[0119] The specific algorithm development is as follows:

[0120] 1) Identify named entities in the input: According to relevant user characteristics and business scenarios, extract key entities based on the input fields.

[0121] 2) Entity disambiguation and entity linking: First, disambiguate the ambiguous entities. For example, "People's Republic of China" and "China" refer to the same entity, which is uniformly called "People's Republic of China" in the knowledge base. When "China" is input, the algorithm automatically identifies this entity as "People's Republic of China". After completing entity disambiguation, link the extracted entities with the corresponding entities in the knowledge base.

[0122] 3) Apply TransE to calculate the embedded vector of the input triple: Construct the input triple according to the filled information. For the input triple, apply the TransE model to transform it into an embedded vector.

[0123] 4) Match with the triples in the knowledge graph to obtain the recommended content: Compare and match the triples obtained in the previous step with the corresponding triples in the knowledge graph, and select the result with the highest matching degree for output. In this step, a specific number of historical data records ranked at the top can be recommended according to user needs. The historical data records recommended here are ranked from high to low according to the recommended degree values calculated by the algorithm. The recommended degree is calculated from the cosine similarity of the triple vectors. The larger the value calculated from the cosine similarity of a certain historical data, the higher the matching degree, the higher its corresponding recommended degree, and the more forward it ranks in the recommended results.

[0124] 5) Combine specific matching rules to adjust the recommended results obtained in the previous step: In the previous step, the machine learning algorithm automatically learned the recommended degrees of each historical data according to the knowledge in the knowledge graph, that is, the vector cosine matching value. The larger this value, the higher the recommended degree, that is, the matching degree. Mark this value as score1. In this step, the algorithm will calculate a recommended degree score for each historical record again according to the set rules and record it as score2. Finally, the final recommended degree score will be obtained according to the following calculation formula:

[0125] score final = α * score1 + (1 - α)score2

[0126] Among them, α is a number greater than 0 and less than 1, used to control the weights of the two parts of the scores. This value will be adjusted through multiple experiments and artificial experience assistance during the algorithm implementation. score2 is obtained according to different rules in different application modules.

[0127] (3) Construct a knowledge Q&A model for nuclear industry laws and regulations

[0128] Based on the constructed knowledge graph model and knowledge system of nuclear industry laws and regulations, use natural language processing, knowledge Q&A and other technologies to realize the knowledge Q&A function of domestic and foreign nuclear industry laws and regulations.

[0129] The function framework of knowledge Q&A for laws and regulations based on the knowledge graph is as Figure 6 shown.

[0130] 1) Data collection and construction of knowledge graph: Collect relevant laws and regulations data and construct a knowledge graph.

[0131] 2) Entity recognition and relationship extraction: Use natural language processing technology to perform entity recognition and relationship extraction on the input questions. This can help the system understand the laws and regulations entities involved in the questions and their relationships.

[0132] 3) Problem analysis and matching: Analyze the questions raised by users, extract the key information, and match it with the entities and relationships in the knowledge graph. This can help the system find relevant laws and regulations.

[0133] 4) Answer generation and display: Generate corresponding answers based on the matching results and display them to the front end. The answers can be summaries of laws and regulations, content of specific articles, etc.

[0134] 5) User interaction and feedback: Provide a user interaction interface that allows users to further ask questions or provide feedback. This can help the system continuously improve and optimize the accuracy and completeness of the answers.

[0135] 1) Question design

[0136] In this patent, the knowledge graph is mainly constructed based on the titles of laws and regulations in the nuclear industry. Since there is relatively little detailed content of laws and regulations in the knowledge graph, questions are mainly designed based on the attributes and relationships of laws and regulations. For example:

[0137] What are the laws and regulations in the nuclear industry?

[0138] Please provide the names of the main laws and regulations related to nuclear safety.

[0139] What is the scope of application (field) of the XXX regulation?

[0140] What are the laws and regulations related to nuclear radiation?

[0141] What are the relevant regulations on nuclear waste management?

[0142] What are the regulatory requirements for the transportation of nuclear materials?

[0143] What are the regulations issued during the period from XXX to XXX?

[0144] What nuclear safety conventions has China signed with France?

[0145] What are the regulations related to the XXX regulation?

[0146] These questions cover different topics of laws and regulations in the nuclear industry, including time, field, relationships between laws and regulations, etc. According to the content of the knowledge graph, the scope of questions can be expanded to meet the needs of users.

[0147] 2) Q&A algorithm design

[0148] In this patent, question design mainly focuses on the attributes and relationships of laws and regulations. Therefore, a Q&A system based on keyword matching is designed. The most important part in the keyword matching algorithm is keyword extraction. This patent uses the TF-IDF algorithm to evaluate the importance of keywords.

[0149] The TF-IDF (Term Frequency-Inverse Document Frequency) algorithm is a statistical method used to evaluate the importance of a word for a document set or a single document in a corpus. The importance of a word increases proportionally with the number of times it appears in a document, but decreases inversely with the frequency of its occurrence in the corpus. This algorithm has been widely applied in fields such as data mining, text processing, and information retrieval, such as finding the keywords in an article.

[0150] The main idea of TF-IDF is that if a word or phrase has a high term frequency (TF) in an article and rarely appears in other articles, it is considered to have good class discrimination ability and is suitable for classification. TF-IDF is actually TF * IDF. Here, TF (Term Frequency) represents the frequency of a term in a document; IDF (Inverse Document Frequency), the main idea of which is that if the number of documents containing a word is less, the discrimination of this word is greater, that is, the IDF is larger. For how to obtain the keywords of an article, we can calculate the TF-IDF of all nouns in this article. The larger the TF-IDF, the higher the discrimination of this noun for this article. Selecting several words with larger TF-IDF values can be regarded as the keywords of this article.

[0151] Thus, it can be seen that the more documents a word is contained in, the smaller the IDF value, that is, this word is very common and not the most important keyword that can distinguish the characteristics of an article.

[0152] In this way, the TF-IDF value of a certain term, that is, a term that appears frequently in a file and appears less frequently in the entire corpus of files, can generate a high-weight TF-IDF. Therefore, TF-IDF tends to filter out common words and retain important words. That is: the importance of a word increases proportionally with the number of times it appears in a document, but decreases inversely with the frequency of its occurrence in the corpus.

[0153] TF-IDF combines two important concepts: term frequency (TF) and inverse document frequency (IDF).

[0154] The term frequency (TF) represents the frequency of a word in a document and is usually calculated using the following formula:

[0155]

[0156] The Inverse Document Frequency (IDF) represents the importance of a term in a document collection and can be calculated using the following formula:

[0157]

[0158] Combining TF and IDF, calculate the TF-IDF weight:

[0159] TF-IDF(t, d) = TF(t, d) * IDF(t)

[0160] (4) Construct a comparative analysis model of domestic and foreign nuclear industry laws and regulations

[0161] Based on the constructed nuclear industry laws and regulations knowledge graph model and knowledge system, using natural language processing, semantic similarity and other technologies, establish the association of domestic and foreign similar laws and regulations, realize the function of associative display of domestic and foreign nuclear industry laws and regulations, and lay a foundation for subsequent benchmarking research and analysis.

[0162] The associative analysis framework of domestic and foreign nuclear industry laws and regulations based on the knowledge graph can include the following steps.

[0163] 1) Entity recognition and relationship extraction: Use natural language processing technology to perform entity recognition and relationship extraction on the collected laws and regulations texts. This can help the system understand the entities involved in the laws and regulations (such as law names, clauses, key concepts, etc.) and the relationships between them (such as citation relationships, similarity relationships, etc.).

[0164] 2) Construct a knowledge graph: Collect relevant data on domestic and foreign nuclear industry laws and regulations, and construct a knowledge graph of domestic and foreign nuclear industry laws and regulations, where the nuclear industry laws and regulations are represented as nodes and the relationships between them are represented as edges.

[0165] 3) Associative analysis algorithm: Design a suitable associative analysis algorithm for calculating the association between laws and regulations. Graph algorithms, similarity calculations and other methods can be considered to measure the similarity, correlation or association degree between laws and regulations.

[0166] 4) Visualization chart display: Based on the knowledge graph and associative analysis algorithm, conduct associative analysis of laws and regulations and visually display the results. A graphical interface or visualization tool can be used to display the results of the associative analysis in the form of charts, network diagrams, etc.

[0167] 5) Result interpretation and application: Interpret the results of the associative analysis to help users understand the relationships between laws and regulations and apply them as needed. For example, based on the results of the associative analysis, the commonalities, differences and dependencies between different laws and regulations can be found, providing reference and decision-making support for the nuclear industry.

[0168] The correlation analysis based on the knowledge graph mainly focuses on the connections and comparisons of domestic and foreign laws and regulations. Therefore, the algorithm mainly focuses on entity alignment. Entity alignment in the knowledge graph refers to the process of matching and linking similar entities in different knowledge graphs. In the knowledge graphs of domestic and foreign laws and regulations, entities refer to specific events, devices, institutions, etc., and the goal of entity alignment is to identify the corresponding relationships that describe the same entity in different knowledge graphs.

[0169] The process of entity alignment can be divided into two main steps: entity recognition and entity link.

[0170] Entity recognition: In different knowledge graphs, entities may be represented in different ways and even use different naming methods. The goal of entity recognition is to find similar entities in different knowledge graphs and determine that they may be the same entity.

[0171] Entity link: Once similar entities are recognized, the goal of entity link is to match and link them to establish associations across knowledge graphs. This can be achieved by comparing the attributes, relationships, and context information of the entities.

[0172] Comparative analysis: Based on the results of entity alignment, conduct a comparative analysis of domestic and foreign knowledge graphs to discover the differences and laws of domestic and foreign laws and regulations knowledge.

[0173] The present invention has been described in detail above with reference to the accompanying drawings and embodiments. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those of ordinary skill in the art, various changes can be made without departing from the spirit of the present invention. The content not described in detail in the present invention can all adopt the prior art.

Claims

1. A method for constructing a nuclear industry laws and regulations database based on a knowledge graph, characterized in that The method includes: Step 1, automatically collecting nuclear industry laws and regulations; Step 2, constructing a knowledge graph of nuclear industry laws and regulations; Step 3, constructing an intelligent application model based on the knowledge graph of nuclear industry laws and regulations.

2. The method for constructing a nuclear industry laws and regulations database based on a knowledge graph according to claim 1, wherein The said Step 1 includes: researching and studying the data types and field types of laws and regulations published on government websites, industry association publications, and legal databases, and collecting nuclear industry laws and regulations text data; inputting, modifying, querying, viewing, and exporting the laws and regulations text data; storing the extracted and maintained structured laws and regulations information into a laws and regulations database for subsequent querying, viewing, knowledge graph construction, and intelligent application of laws and regulations information.

3. A method for constructing a nuclear industry laws and regulations database based on a knowledge graph according to claim 1, characterized in that The said Step 2 includes: Data preprocessing: cleaning the data of nuclear industry laws and regulations: using regular expressions to remove punctuation marks and stop words in the text, and removing abnormal segments in the text; constructing a domain word library, adding the domain word library to the word segmentation tool, and using a natural language word segmentation tool to perform word segmentation on the cleaned text data; Natural language processing: applying the word2vec model to obtain a word embedding matrix; representing the segmented text in the form of a one-hot encoded vector and multiplying it by the word embedding matrix to obtain the corresponding word vector; applying Bi-LSTM+CRF for sequence annotation to obtain the annotation of named entities and entity attributes; applying Bi-LSTM+ATT for relationship extraction to obtain the relationships between entities for embedding in the knowledge graph; Knowledge graph embedding: applying the TransE or DistMult model to integrate the data and realize data storage with a unified semantic data structure.

4. A method for constructing a nuclear industry laws and regulations database based on a knowledge graph according to claim 3, characterized in that, The said data includes unstructured data, semi-structured data, and structured data.

5. A method for constructing a nuclear industry laws and regulations database based on a knowledge graph according to claim 3, characterized in that, The said domain word library includes a nuclear industry laws and regulations professional word library, a general word library, and a synonym word library.

6. The method for constructing a nuclear industry laws and regulations database based on a knowledge graph according to claim 1, wherein The said Step 3 includes: constructing a precise semantic retrieval model of nuclear industry laws and regulations, constructing an intelligent push model of nuclear industry laws and regulations, constructing a knowledge Q&A model of nuclear industry laws and regulations, and constructing a comparative analysis model of domestic and foreign nuclear industry laws and regulations.

7. A method for constructing a nuclear industry laws and regulations database based on a knowledge graph according to claim 6, characterized in that, The specific construction of the precise semantic retrieval model of nuclear industry laws and regulations is as follows: Step A1, query processing: processing the query input by the user, performing word segmentation and stop word removal operations; using the same semantic modeling model as that for constructing the laws and regulations word library and semantic library to represent the query input; Step A2, graph traversal: based on the semantic representation of the query, searching in the knowledge graph through a graph traversal algorithm; Step A3, similarity calculation: using a similarity calculation method to measure the semantic similarity between the query and the laws and regulations text; Step A4, sorting and retrieval: according to the results of similarity calculation, based on a similarity-based sorting algorithm, sorting the laws and regulations text so that the most relevant text is ranked first; Step A5, interface interaction: designing a user-friendly interface that enables users to input queries and view retrieval results; Step A6, evaluation and optimization: evaluating the retrieval framework, using some standard test sets or manual evaluation to measure the accuracy and relevance of retrieval results, and optimizing according to the evaluation results.

8. A method for constructing a nuclear industry laws and regulations database based on a knowledge graph according to claim 6, characterized in that, The specific construction of the intelligent push model of nuclear industry laws and regulations is as follows: Step B1. Identify named entities in the input: Extract key entities according to relevant user characteristics, business scenarios, and input fields. Step B2. Entity disambiguation and entity linking: Disambiguate ambiguous entities. After entity disambiguation, link the extracted entities to the corresponding entities in the knowledge base. Step B3. Apply TransE to calculate the embedded vector of the input triple: Construct the input triple based on the filled information. For the input triple, apply the TransE model to transform it into an embedded vector. Step B4. Match with triples in the knowledge graph to obtain recommended content: Compare and match the obtained triples with the corresponding triples in the knowledge graph, and select the recommended result with the highest matching degree for output. Step B5. Adjust the obtained recommended results in combination with specific matching rules: In the previous step, the machine learning algorithm automatically learns the recommendation degree of each historical data, that is, the vector cosine matching value, from the knowledge in the knowledge graph. The larger this value, the higher the recommendation degree, that is, the matching degree. Mark this value as score1. In this step, the algorithm will calculate a recommendation degree score for each historical record again according to the set rules and record it as score2. Finally, the final recommendation degree score will be obtained according to the following calculation formula: score final = α * score1 + (1 - α) score2 where α is a number greater than 0 and less than 1, used to control the weights of the two parts of the scores. This value will be obtained through multiple experiments and manual experience assistance during algorithm implementation. score2 is obtained according to different rules in different application modules.

9. A method for constructing a nuclear industry laws and regulations database based on a knowledge graph according to claim 6, wherein, The construction of the nuclear industry laws and regulations knowledge Q&A model is specifically as follows: Step C1. Data collection and knowledge graph construction: Collect relevant laws and regulations data and construct a knowledge graph. Step C2. Entity recognition and relationship extraction: Use natural language processing technology to perform entity recognition and relationship extraction on the input questions. Step C3. Question parsing and matching: Parse the questions raised by users, extract key information, and match it with the entities and relationships in the knowledge graph. Step C4. Answer generation and display: Generate corresponding answers according to the matching results and display them to the front end. Step C5. User interaction and feedback: Provide a user interaction interface to allow users to further ask questions or provide feedback.

10. A method for constructing a nuclear industry laws and regulations database based on a knowledge graph according to claim 6, characterized in that, The construction of the domestic and foreign nuclear industry laws and regulations comparative analysis model is specifically as follows: Step D1. Entity recognition and relationship extraction: Use natural language processing technology to perform entity recognition and relationship extraction on the collected laws and regulations texts. Step D2. Knowledge graph construction: Collect relevant data on domestic and foreign nuclear industry laws and regulations and construct a knowledge graph of domestic and foreign nuclear industry laws and regulations. Step D3. Relevance analysis algorithm: Design a suitable relevance analysis algorithm for calculating the relevance between laws and regulations, and use graph algorithms and similarity calculation methods to measure the similarity, correlation, or association degree between laws and regulations. Step D4. Visualization chart display: Based on the knowledge graph and the relevance analysis algorithm, conduct relevance analysis of laws and regulations and visually display the results. Step D5, Result Interpretation and Application: Interpret the results of the relevance analysis to help users understand the relationships between laws and regulations and apply them as needed.

Images