A method for tracking governance structure changes based on policy citation correlation

Abstract

The present invention discloses a method for tracking governance structure changes based on policy citation association, comprising the following steps: S1 constructing an industrial policy pool and performing data cleaning operations; S2 extracting policy meta-elements; S3 extracting policy context elements; S4 identifying policy citation relationships; S5 constructing a weighted, directed policy element citation network; S6 ranking the importance of policy elements; S7 performing key topology mining and governance structure change tracking on a single type of policy element citation network and a time series policy combination citation network; the present invention traces the structural changes of the industrial policy knowledge system by identifying the citation and cited relationships of policy documents, thereby compensating for the problems of expert subjectivity, coarse detection granularity and insufficient accuracy of existing subject evolution tracking methods, and can improve the traceability and interpretability of policy quantitative analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of data representation and computer system, and in particular to a method for tracking governance structure changes based on policy citation association. Background Art

[0002] The rapid growth of large-scale and massive policy document resources has made it difficult for a large amount of fragmented government data and policy knowledge to support decision-making quickly and accurately. In particular, the current level of intelligent analysis of the structural characteristics and knowledge system of large-scale policy texts is weak. Most policy contents are fragmented and scattered, lacking semantic associations, and there is no effective reference between policies, and the knowledge contained cannot be effectively reused. Policy document resources contain multiple types of knowledge elements such as policy goals, tools, measures, and the associated states of cooperation, coordination, and reference between elements. Intelligent analysis and identification of policy content elements and their potential associations is an important way to eliminate the risk of policy system fragmentation, improve the ability of policy knowledge elements to be reused collaboratively, and promote the scientific and holistic nature of policy making. Based on natural language processing technology, multi-dimensional analysis and intelligent calculation of large-scale policy texts are carried out to describe, explain, predict and evaluate the structure, evolution process and development trend of specific industrial policy systems, which has important research and practical value for promoting government evidence-based decision-making, inferring the construction of digital government and improving the level of government governance.

[0003] Current research mainly focuses on small-scale policy samples, and the extraction and calculation of policy elements, element intensity, and element relationships still rely on manual indexing, which is difficult to support efficient, large-scale, and fine-grained policy processing projects. Moreover, existing research is usually limited to a single element and its associated network, focusing on the focus and evolution of policy elements in the temporal dimension from the perspectives of policy theme derivation and diffusion, policy goal continuation and deviation, policy tool configuration and innovation, etc., but pays less attention to the common political phenomenon of policy document citations, and ignores the integrity and systematicity of multi-type policy elements, relationship topology, and combination changes in the policy system.

[0004] In summary, existing methods for quantitative policy analysis and tracking the evolution of governance structures are more based on topic evolution analysis and policy element co-occurrence network structure mining. The granularity of identifying and expressing policy elements is relatively coarse, and it is difficult to detect the accumulation, continuity and inheritance of policy knowledge, and it is difficult to accurately trace the source of knowledge and administrative basis in the evolution of the policy system. Summary of the invention

[0005] Purpose of the invention: The purpose of the present invention is to provide a method for tracking governance structure changes based on policy citation associations, so as to solve the problems of expert subjectivity, coarse detection granularity and insufficient accuracy in existing subject evolution tracking methods; and to solve the limitations of the association of single policy elements.

[0006] Technical solution: The method for tracking governance structure changes based on policy citation association described in the present invention includes the following steps:

[0007] S1 collects policy documents issued by governments at all levels according to the target industry, builds an industrial policy pool, and performs data cleaning operations;

[0008] S2 Extracting policy meta-elements: Identifying basic meta-elements and secondary meta-elements based on the unified macro-text layout and functional structure pattern followed by policy documents;

[0009] S3 Extraction of policy context elements: Use deep learning algorithms and domain element dictionaries to perform fine-grained analysis and automatic indexing of lexical and formal features in policy document content, and identify three types of policy context elements: policy objectives, policy measures, and policy tools;

[0010] S4 Identify policy citation relationships: Extract the citation and cited relationships between policy documents based on the book title marks and citation prompts in the content of policy documents;

[0011] S5 builds a weighted and directed policy element citation network: according to the mention intensity of different policy elements, optimizes the edge weight calculation method, merges and removes duplicates of directed links of the same policy elements, and generates a time-series weighted and directed policy element citation network;

[0012] S6 ranks the importance of policy elements: introduces the unique attributes and citation features of policy documents, designs a PageRank algorithm that optimizes node jump probability, and identifies core policy elements and dynamic changes in different periods;

[0013] S7 conducts key topology mining and governance structure change tracking on the citation network of single-type policy elements and the citation network of temporal policy combinations.

[0014] Furthermore, in step S1, the data cleaning operation includes deleting duplicate policy files, eliminating forwarding files that only require the execution of existing files and do not publish any substantive content, and removing policy files that are not closely related to the target field.

[0015] Furthermore, in step S2, the basic elements include: policy title, issuing agency, document type, drafting and effective date; the secondary elements include: issuing area, effectiveness level, and administrative level.

[0016] Furthermore, step S3 is specifically as follows:

[0017] S31 uses the BERT model to extract keywords from the top-N important policy documents released in the target industry field;

[0018] S32 preliminarily determined the seed terminology set of three types of policy elements, namely policy objectives, policy measures and policy tools, through manual review and existing policy element dictionaries;

[0019] S33 imports the seed term sets of three types of policy elements into the deep learning model to expand the similar word set of each seed word;

[0020] S34 identifies these element term sets for each policy document based on the maximum forward matching algorithm. For policy documents that cannot be matched with the predetermined element term set, relevant terms are extracted through manual review, and the three types of policy element seed terms are expanded. The feature dictionary of the three types of policy elements is constructed through multiple iterations;

[0021] S35 represents each policy document as an N-dimensional vector consisting of three types of elements: policy objectives, policy measures, and policy tools.

[0022] Further, step S4 is specifically as follows:

[0023] S41 identifies all citation and cited relationships embedded in the content of each policy document based on the quotation marks in the content of the policy document and the prompt words "basis", "implementation", "based on", "regulation", "combination" and "compliance" before and after the citation;

[0024] S42 cumulatively weights multiple mentions of the same policy document in the content of a policy document and calculates the frequency of citations.

[0025] Furthermore, step S5 is specifically as follows:

[0026] S51 calculates the mention intensity and intensity proportion of the three types of policy elements in each policy document;

[0027] S52 considers the relative importance of policy elements and optimizes the edge weight calculation method between policy elements in two cited policy documents according to the mention intensity of policy elements;

[0028] S53 accumulates and adds the edge weights between policy elements in different policy documents to construct a time-series citation network of a single type of policy element and a citation network of multiple types of policy combinations.

[0029] Furthermore, step S6 is specifically as follows:

[0030] S61 calculates the effectiveness level to which each policy element belongs based on the document distribution characteristics of the policy element in the industrial policy pool;

[0031] S62 uses logarithmic normalization to standardize the citation intensity and effectiveness level of policy elements and reduce the span of values;

[0032] S63 uses standardized policy elements to incorporate the attributes of citation intensity and effectiveness level into the traditional PageRank algorithm to initialize the PageRank value;

[0033] S64 optimizes the PageRank algorithm’s jump probability calculation strategy based on the weights of citation links between policy elements, and stops iteration when the importance of all policy elements remains unchanged;

[0034] S65 is compared with the results of relevant indicators based on policy attributes and the classic node ranking method, and the consistency and reliability of the optimization strategy in ranking the importance of policy elements are evaluated based on the Spearman correlation coefficient;

[0035] S66 obtains the importance ranking results of three types of core policy elements in different periods and analyzes their spatiotemporal dynamic change characteristics.

[0036] Further, step S7 is specifically as follows:

[0037] S71 is based on the importance ranking of core policy elements in different periods, selects Top-N policy elements to filter the policy element citation association network to construct a single type of core policy target citation network, core policy measure citation network, and core policy tool citation network;

[0038] S72 explores the key topology and structural evolution of the core policy elements in different time and space dimensions based on social network analysis methods, and tracks the changes in focus and evolution paths in governance goals, governance means and measures;

[0039] S73 constructs a time-series policy combination citation network based on the intensity of Top-N policy elements and their citation associations in different periods;

[0040] S74 focuses on the policy combination citation network over time, measures the complexity, consistency, coherence and comprehensiveness of policy combinations in different time and space dimensions, explores the key topological and structural evolution laws of the policy combination network, and tracks the changes in the concept of collaborative governance and governing philosophy in different time and space.

[0041] An electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, it implements any one of the methods for tracking governance structure changes based on policy citation association.

[0042] A storage medium described in the present invention stores a computer program, and when the computer program is executed by a processor, it implements any one of the methods for tracking governance structure changes based on policy citation association.

[0043] Beneficial effects: Compared with the existing technology, the present invention has the following significant advantages: it can identify the core policy elements and governance structure in the policy system, and track the changes and evolution of policy governance concepts; it introduces the perspective of policy citation association analysis to objectively and truly trace the accumulation, continuity and inheritance of legal knowledge, improve the traceability and interpretability of policy quantitative analysis, and the policy citation relationship can reflect the flow, inheritance, emergence and dependency patterns of policy knowledge elements, and provide a novel relational perspective for revealing the policy element selection path and identifying core policy elements; it overcomes the limitations of the association of a single policy element, and focuses on multiple types of policy elements, relationship topology and Taking into account the integrity and systematicness of the combination change, a single type of policy element citation network and multiple types of policy combination citation networks are constructed, and the change path and governance structure of the core policy objectives, measures and tool elements and their combinations of specific industries are revealed based on policy content mining and network analysis methods; considering the Matthew effect in the policy element selection process, the citation relationship weights are optimized based on the importance of policy elements, and the policy effectiveness attributes are introduced to improve the node jump probability of the PageRank algorithm, which is applied to the optimized policy element citation network to accurately identify the core policy elements in different periods, and reveal the change path, spatiotemporal configuration structure and government governance rules of the core elements of the target industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0045] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0046] like Figure 1 As shown, an embodiment of the present invention provides a governance structure change tracking method based on policy citation association, comprising the following steps:

[0047] S1. Select the target industry, collect large-scale policy documents issued by the government to build an industrial policy pool, and perform data cleaning operations; including the following steps:

[0048] S11. Select a typical emerging industry research field as the target field. The requirement for the target field is to publish large-scale policy documents in the industry field. Based on the selected target field, the relevant policy documents are sourced from the Peking University Law Database (https: / / www.pkulaw. com / ). The search formula is “title=((keyword 1) OR (keyword 2) OR… OR (keyword 𝑛 )), PY = (Start Time - Now);

[0049] S12. Perform a combination match based on the policy document number and legal document citation code in the industrial policy pool to remove duplicate policy documents, and only retain one policy document;

[0050] S13. Use the Cosine similarity calculation method to detect the text content similarity between policy documents, and set the upper limit of the similarity threshold γ. When 90% ≤γ<100%, it is considered that the two policy documents only have government forwarding behavior. These forwarding documents that only require the implementation of existing documents and do not publish any substantive content are eliminated;

[0051] S14. Through manual review, remove policy documents that are not closely related to the substantive content of the target industry to ensure the effectiveness and accuracy of policy document collection;

[0052] S2. Extraction of policy meta-elements: Based on the unified macro-text layout and functional structure pattern followed by policy documents, identify the basic meta-elements of policy title, issuing agency, document type, and date of writing and effectiveness, as well as the secondary meta-elements of issuing region, effectiveness level, and administrative level; including the following steps:

[0053] S21. Automatically parse the text genre structure of policy documents based on natural language processing technology to identify the basic elements of policy title, policy text, issuing agency, timeliness, document type, and drafting and effective date;

[0054] S22. Match the policy titles and issuing agencies according to the city list to accurately identify the policy issuing areas;

[0055] S23. Determine the effectiveness level of policy documents according to the legal effectiveness hierarchy system;

[0056] S24. Clean, complete and normalize each policy attribute field through a hybrid method of precise word matching and manual indexing;

[0057] S3. Extraction of policy context elements: Use deep learning algorithms and domain element dictionaries to perform fine-grained analysis and automatic indexing of lexical features and formal features in the content of policy documents, and identify three types of elements: policy objectives, policy measures, and policy tools; including the following steps:

[0058] S31. Use the BERT model to extract keywords from the top-N important policy documents released in the target industry, and perform keyword cleaning, disambiguation and alignment preprocessing;

[0059] S32. To determine the initial dictionary parameters, the seed term sets of three types of elements, namely policy objectives, policy measures and policy tools, are preliminarily determined through manual review and the existing policy element dictionary. Specifically, policy objectives are divided into three types: economic concern, environmental concern, and social concern; policy tools are divided into three types: supply-oriented, environmental, and demand-oriented; policy measures are divided into eight types: administrative measures, regulatory measures, supervisory and guarantee measures, personnel measures, guiding measures, fiscal measures, financial measures, and electricity price measures;

[0060] S33, importing seed term sets related to various types of policy elements into the deep learning model to expand the similar word set of each seed word;

[0061] S34, identifying these element term sets of each policy document according to the Maximum Matching segmentation (MM) algorithm, extracting relevant terms through manual review for policy documents that cannot be matched with the predetermined element term set, and expanding the three types of element seed term sets;

[0062] S35, iteratively construct feature dictionaries of three types of policy elements to annotate the existence and frequency of policy elements in policy documents;

[0063] S36. Each policy document is represented by N 1 policy goal, N 2 policy measures and N Three policy tools N dimensional vector, N = N 1+ N 2+ N 3;

[0064] S4, policy citation relationship identification: extract the citation and cited relationship between policy documents according to the book title mark and citation prompt words in the content of policy documents; including the following steps:

[0065] S41. Cited policies are usually marked with book title marks when they are mentioned. The string matching algorithm (Knuth-Morris-Pratt, KMP) is used to efficiently identify the book title marks embedded in the content of policy documents.

[0066] S42, identify the prompt words "basis", "implementation", "according to", "regulation", "combination", and "compliance" before and after the quotation marks, determine the citation context and function type, and classify the citation relationship;

[0067] S43, counting the total number of citations, total number of citations and total number of citations of each policy document in the industrial policy pool, and ranking them;

[0068] S5. Construction of weighted and directed policy element citation network: Optimize the edge weight calculation method according to the mention intensity of different policy elements, merge and remove the directed links of the same policy elements, and generate a time-series weighted and directed policy element citation network; including the following steps:

[0069] S51. Calculation of policy objectives a , Policy measures b and policy tools c Citation intensity in each policy document m and intensity ratio w = m / M ;in, M Indicates its frequency of mention in the entire policy pool;

[0070] S52. Consider the relative importance of policy elements and optimize the two cited policy documents according to the intensity of policy element mentions. p , q Edge weight calculation method between internal policy elements weight i,j ( p , q ) = m i / M p + m j / M q ; m i It is a policy element i In Policy p The strength of m j It is a policy element j In Policy q The strength of M p All policy elements are in the policy p The total intensity of M q All policy elements are in the policy q The total intensity of

[0071] S53. Policy elements in different policy documents in the policy pool i, j The edge weights between them are cumulatively added to obtain the policy elements i, j The final edge weight weight i,j = weight i,j ( p , q ) + weight i,j (p , k ) +···+ weight i,j (q, k );

[0072] S54. With the same type of policy elements a i 、b i or c i As a node, a i and a j The weights between them are used as edges to construct a policy target citation network G target ( t )= graph ( a i , l target );by b i and b j The weights between them are used as edges to construct the policy measure citation network G measure ( t )= graph ( b i , l measure );by c i and c j The weights between them are used as edges to construct the policy tool citation network G tool ( t )= graph ( c i , l tool );

[0073] S55. Different types of policy elements a i 、b i and c i As a node, a i 、b i and c i The weights between them are used to construct a policy hybrid citation network G mix (t ) = graph ( mix i , l mix );

[0074] S6. Ranking of policy elements by importance: Introducing the unique attributes and citation features of policy documents, designing a PageRank algorithm that optimizes node jump probability, and identifying core policy elements and dynamic changes in different periods; including the following steps:

[0075] S61. Calculate the effectiveness level of each policy element based on the document distribution characteristics of the policy elements in the industrial policy pool;

[0076] S62. Use the logarithmic normalization method to normalize the citation intensity and effectiveness level of policy elements from 0 to 1 to reduce the span of the values;

[0077] S63. Use the standardized policy element citation intensity and effectiveness level attributes to integrate into the traditional PageRank algorithm and initialize the PageRank value. The calculation formula is as follows:

[0078] ;

[0079] in, d is the damping factor; CN ( e i ) indicates policy elements e i Frequency of occurrence in policy documents; N ( e i ) indicates policy elements e i Frequency of occurrence in all policies;

[0080] S64. Optimize the jump probability calculation strategy of the PageRank algorithm according to the weight of the citation links between policy elements. The calculation formula is as follows:

[0081] ;

[0082] Among them, the jump probability of node i is expressed as ; Reference link weights for features.

[0083] S65, the PageRank algorithm that introduces policy attribute optimization is operated, and the iteration is stopped when the importance ranking of all factors remains unchanged;

[0084] S66. Compare with the relevant indicators based on policy attributes and the results of the classic node ranking method, and evaluate the consistency and reliability of the optimization strategy in the ranking of factor importance based on the Spearman correlation coefficient. The calculation formula is as follows:

[0085] ;

[0086] in, ρ represents the correlation coefficient, is the difference between the two ranking results, n is the sample size;

[0087] S67. Using five-year periods as the basis for division, obtain the importance ranking of the three core policy elements at different stages and analyze their dynamic change characteristics;

[0088] S7. Key network topology and structural change: Key topology mining and structural change tracking are carried out for the citation network of a single type of policy element and the citation network of multiple types of policy combinations. The following steps are included:

[0089] S71. Based on the importance ranking of core policy elements in different periods, select the Top-N policy elements in each period to filter the policy element citation association network, and construct a time-series, single-type core policy target citation network, core policy measure citation network, and core policy tool citation network;

[0090] S72. Based on social network analysis methods, explore the key topology and structural evolution of the core policy elements in different time and space dimensions, and track the changes in focus and evolution paths in governance goals, governance means and measures;

[0091] S73. Construct a time-series policy portfolio citation network based on the citation intensity and citation relationship of the Top-N policy elements in different periods;

[0092] S74. Measure the complexity, consistency, coherence and comprehensiveness of policy combinations in different time and space dimensions based on the policy combination citation network, explore the key topological and structural evolution laws of the policy combination network, and track the changes in the concept of collaborative governance and governing philosophy in different time and space.

Claims

1. A governance structure change tracking method based on policy citation association, characterized in that: The following steps are involved: S1 collects policy documents issued by governments at all levels according to the target industry, builds an industrial policy pool, and performs data cleaning operations; S2 Extracting policy meta-elements: Identifying basic meta-elements and secondary meta-elements based on the unified macro-text layout and functional structure pattern followed by policy documents; S3 Extraction of policy context elements: Use deep learning algorithms and domain element dictionaries to perform fine-grained analysis and automatic indexing of lexical features and formal features in policy document content, and identify three types of policy context elements: policy objectives, policy measures, and policy tools; S4 Identify policy citation relationships: Extract the citation and cited relationships between policy documents based on the book title marks and citation prompts in the content of policy documents; S5 builds a weighted and directed policy element citation network: according to the mention intensity of different policy elements, optimizes the edge weight calculation method, merges and removes duplicates of directed links of the same policy elements, and generates a time-series weighted and directed policy element citation network; including the following steps: S51 Computational policy objectives a , Policy measures b and policy tools c Citation intensity in each policy document m and intensity ratio w = m / M ;in, M Indicates its frequency of mention in the entire policy pool; S52 Considering the relative importance of policy elements, optimize the two cited policy documents according to the intensity of policy element mentions p , q Edge weight calculation method between internal policy elements weight i,j ( p , q )= m i / M p + m j / M q ; m i It is a policy element i In Policy p The strength of m j It is a policy element j In Policy q The strength of M p All policy elements are in the policy p The total intensity of M q All policy elements are in the policy q The total intensity of S53 combines policy elements from different policy documents in the policy pool i, j The edge weights between them are cumulatively added to obtain the policy elements i, j The final edge weight weight i,j = weight i,j ( p , q ) + weight i,j ( p , k ) +···+ weight i,j (q, k ); S54 Policy elements of the same type a i 、b i or c i As a node, a i and a j The weights between them are used as edges to construct the policy target citation network G target ( t )= graph ( a i , l target );by b i and b j The weights between them are used as edges to construct the policy measure citation network G measure ( t )= graph ( b i , l measure );by c i and c j The weights between them are used as edges to construct the policy tool citation network G tool ( t )= graph ( c i , l tool ); S55 Different types of policy elements a i 、b i and c i As a node, a i 、b i and c i The weights between them are used to construct a policy hybrid citation network G mix ( t ) = graph ( mix i , l mix ); S6 ranks the importance of policy elements: introduces the unique attributes and citation features of policy documents, designs a PageRank algorithm that optimizes node jump probability, and identifies core policy elements and dynamic changes in different periods; including the following steps: S61 calculates the effectiveness level to which each policy element belongs based on the document distribution characteristics of the policy element in the industrial policy pool; S62 uses the logarithmic normalization method to standardize the citation intensity and effectiveness level of policy elements from 0 to 1 to reduce the span of the values; S63 uses the standardized policy element citation intensity and effectiveness level attributes to integrate into the traditional PageRank algorithm and initialize the PageRank value. The calculation formula is as follows: ； in, d is the damping factor; CN ( e i ) indicates policy elements e i Frequency of appearance in central and local policy documents; N ( e i ) indicates policy elements e i Frequency of occurrence in all policies; S64 optimizes the jump probability calculation strategy of the PageRank algorithm according to the weight of the citation links between policy elements. The calculation formula is as follows: ； Among them, the jump probability of node i is expressed as ; Refer to link weights for features; The S65 operation introduces the PageRank algorithm for policy attribute optimization, and stops iteration when the importance ranking of all factors remains unchanged; S66 is compared with the relevant indicators based on policy attributes and the results of the classic node ranking method. The consistency and reliability of the optimization strategy in the ranking of element importance are evaluated based on the Spearman correlation coefficient. The calculation formula is as follows: ； in, ρ represents the correlation coefficient, is the difference between the two ranking results, n is the sample size; S67 uses five-year periods as the basis for division, obtains the importance ranking of three types of core policy elements at different stages, and analyzes their dynamic change characteristics; S7 Network Key Topology and Structural Changes: Key topology mining and structural change tracking for single-type policy element citation networks and multi-type policy combination citation networks; including the following steps: S71 is based on the importance ranking of core policy elements in different periods, selects the Top-N policy elements in each period to filter the policy element citation association network, and constructs a time-series, single-type core policy target citation network, core policy measure citation network, and core policy tool citation network; S72 explores the key topology and structural evolution of the core policy elements in different time and space dimensions based on social network analysis methods, and tracks the focus changes and evolution paths in governance goals, governance means and measures; S73 constructs a time-series policy combination citation network based on the citation intensity and citation relationship of Top-N policy elements in different periods; S74 measures the complexity, consistency, coherence and comprehensiveness of policy combinations in different time and space dimensions based on the policy combination citation network, explores the key topological and structural evolution laws of the policy combination network, and tracks the changes in the concepts of collaborative governance and governing philosophy in different time and space.

2. According to claim 1, a governance structure change tracking method based on policy citation association is characterized in that: In step S1, the data cleaning operation includes deleting duplicate policy files, eliminating forwarding files that only require the execution of existing files and do not publish any substantive content, and removing policy files that are not closely related to the target field.

3. According to claim 1, a governance structure change tracking method based on policy citation association is characterized in that: In step S2, the basic elements include: policy title, issuing agency, document type, drafting and effective date; the secondary elements include: issuing area, effectiveness level, and administrative level.

4. According to claim 1, a governance structure change tracking method based on policy citation association is characterized in that: Step S3 includes the following steps: S31 uses the BERT model to extract keywords from the top-N important policy documents released by the central government in the target industry field; S32 preliminarily determined the seed terminology set of three types of policy elements, namely policy objectives, policy measures and policy tools, through manual review and existing policy element dictionaries; S33 imports the seed term sets of three types of policy elements into the deep learning model to expand the similar word set of each seed word; S34 identifies these element term sets for each policy document based on the maximum forward matching algorithm. For policy documents that cannot be matched with the predetermined element term set, relevant terms are extracted through manual review, and the three types of policy element seed terms are expanded. The feature dictionary of the three types of policy elements is constructed through multiple iterations; S35 represents each policy document as an N-dimensional vector consisting of three types of elements: policy objectives, policy measures, and policy tools.

5. According to claim 1, a governance structure change tracking method based on policy citation association is characterized in that: Step S4 includes the following steps: S41 identifies all citation and cited relationships embedded in the content of each policy document based on the quotation marks in the content of the policy document and the prompt words "basis", "implementation", "based on", "regulation", "combination", and "compliance" before and after the citation; S42 cumulatively weights multiple mentions of the same policy document in the content of a policy document and calculates the frequency of citations.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into a processor, it implements a method for tracking governance structure changes based on policy citation association according to any one of claims 1 to 5.

7. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it implements a method for tracking governance structure changes based on policy citation association according to any one of claims 1 to 5.

Images