Digital government affair policy interpretation system and method

Abstract

The invention discloses an interpretation system and method for digital government affair policies, and belongs to the field of digital government affairs, and the method specifically comprises the steps that a policy data obtaining and processing module collects latest government affair policy files from the Internet, and carries out the preprocessing through a natural language, and generates text data; the knowledge graph construction module extracts key elements of the policy according to the policy text content to construct a knowledge graph; the text representation model module converts the policy text data into numerical feature vectors; the relation modeling module defines entities and relations thereof in the knowledge graph as graph data, and constructs a graph model of the relations between the entities; the semantic understanding module performs semantic understanding on the input enterprise information and converts the information into a text sequence through feature search; the reasoning module performs reasoning according to input enterprise information and government affair policies and judges the matching degree of the enterprise information and the government affair policies; the visualization module is used for carrying out visualization display on the generated result; through cooperation of the seven modules, the government affair policy is finally interpreted.

Description

Technical Field

[0001] The present invention belongs to the technical field of digital government affairs, and in particular, relates to a system and method for interpreting digital government affairs policies. Background Art

[0002] Due to the complexity and professionalism of policy content, enterprises often find it difficult to directly understand the specific content and intention of government policies, and are usually unable to clearly judge whether they meet the application requirements and need to consult relevant staff. In addition, the government also needs to arrange manpower to handle a large number of consultations and policy application review work. In the face of this situation, intelligent and accurate policy matching can effectively reduce the workload of manpower and improve application efficiency. However, the existing policy matching system has the following defects:

[0003] 1. The simple labeling system cannot handle the logical problems in the policy matching process and can only provide preliminary screening results. After filling in the information, the enterprise still cannot know whether it can apply for the policy;

[0004] 2. The enterprise needs to fill in too much information, and the information that needs to be filled in cannot be dynamically adjusted according to the enterprise's situation during the information input process;

[0005] 3. Policy sorting relies too much on manual work, which is inefficient and time-consuming for a large number of policy scenarios;

[0006] 4. It is difficult to infer the optimal candidate policy based on user input dynamic information and generate corresponding matching basis and result report;

[0007] Therefore, there is an urgent need for a system and method for interpreting digital government policies to reduce the workload of staff and improve enterprises' understanding of government policies. Summary of the invention

[0008] The purpose of the present invention is to provide a system and method for interpreting digital government policies to solve the problem that enterprises find it difficult to directly understand the specific content and intention of government policies and cannot clearly judge whether they meet the reporting requirements.

[0009] To achieve the above objectives, the present invention provides a system for interpreting digital government policies, including a policy data acquisition and processing module, a knowledge graph construction module, a text representation model module, a relationship modeling module, a semantic understanding module, a reasoning module and a visualization module.

[0010] Preferably, the specific contents of the policy data acquisition and processing module, the knowledge graph construction module, the text representation model module, the relationship modeling module, the semantic understanding module, the reasoning module and the visualization module are as follows:

[0011] Policy data acquisition and processing module: collects the latest government policy documents from the Internet or other sources, pre-processes them through natural language processing technology, and generates processable text data;

[0012] Knowledge graph construction module: Based on the content of the policy text, the key elements of the policy are extracted using extraction, induction, and entity recognition methods to construct a knowledge graph;

[0013] Text representation model module: converts policy text data into numerical feature vectors;

[0014] Relationship modeling module: defines entities and their relationships in the knowledge graph as graph data and builds a graph model of the relationship between entities;

[0015] Semantic understanding module: semantically understands the input enterprise information and converts it into a text sequence through feature search;

[0016] Reasoning module: Reasoning based on the input enterprise information and government policies to determine the matching degree between enterprise information and government policies;

[0017] Visualization module: Visualize the generated results and present them to users intuitively.

[0018] The present invention also provides a method for interpreting digital government policies, comprising the following steps:

[0019] Step 1: Policy acquisition and cleaning: Collect the latest government-related policy documents, remove irrelevant information through natural language processing technology, and organize them into structured data;

[0020] Step 2: Knowledge graph construction: Use policy content to create a knowledge graph, including elements of policy objectives, subjects, measures, and influencing factors, with each element as a node and the relationships between elements as edges;

[0021] Step 3: Semantic parsing and linking: Through the word vector and entity linking technology, the keywords and entities in the policy content are connected with the corresponding nodes in the knowledge graph to form a relationship network of government policies;

[0022] Step 4: Input enterprise information and convert the input enterprise information into a text sequence through feature search; enterprise information includes industry classification, scale, technical expertise, and business model; and enterprise information is captured through network data using crawler technology;

[0023] Step 5: Each word in the enterprise information text sequence and each node in the government policy relationship network are converted into vectors through the word embedding model, and then the matching degree between the enterprise information and the government policy is compared through cosine similarity calculation;

[0024] Step 6, Judgment of similarity degree; for cases with low similarity or incomplete matching, an anomaly detection algorithm is used for secondary judgment to avoid misjudgment, identify and warn enterprises with declaration information that does not conform to the norm after secondary judgment, and give suggestions; for those with high similarity degree, a final grade report is given;

[0025] Step 7, Visualize the generated results, and finally enterprises can export them by themselves.

[0026] Preferably, the specific process of step 1 is as follows:

[0027] S11. Use web crawlers to obtain policy documents from official websites, policy and regulation databases, news reports, and social media, and preprocess the crawled data. The preprocessing includes removing HTML tags, special characters, and format conversion;

[0028] S12. Clean the data preprocessed in S11, including grammar checking, content filtering, and deduplication;

[0029] S13. Perform natural language processing on the data cleaned in S12 to generate structured data; including word segmentation, stemming, lemmatization, part-of-speech tagging, stop word removal, word frequency statistics, word vector generation, entity recognition, and attribute extraction.

[0030] Preferably, the specific content of S13 is as follows:

[0031] S131. Word segmentation: Segment the text in the policy document into a sequence of words;

[0032] S132. Stemming: Extract the stem of each word and remove the affixes;

[0033] S133. Lemmatization: Restore the stem to the specific word form;

[0034] S134. Part-of-speech tagging: Tag the part of speech of each word, including nouns, verbs, and adjectives;

[0035] S135. Stop word removal: Remove common meaningless words, including "de" (of), "le" (particle indicating completion);

[0036] S136. Word frequency statistics: Count the word frequency of each word and find the word with the highest frequency;

[0037] S137. Word vector generation: Use BERT or Word2Vec word vector models to vectorize the words in the policy document;

[0038] S138. Entity recognition: Perform clustering operations on the word vectors to cluster the words in the policy document into different entities;

[0039] S139, Attribute extraction: Starting from the entity, extract the entity's attributes according to the entity's type and context to generate structured data.

[0040] Preferably, the relationship network of government policies in step 3 is in represents a node set, Represents a set of edges.

[0041] Preferably, the specific process of converting the input enterprise information through feature search in step 4 is as follows:

[0042] S41, initialize an empty generated text sequence D, represented by D = {};

[0043] S42, iterating and traversing the input enterprise text sequence X from the first word to the last word;

[0044] S43. Obtain the generated complete text sequence D, including the text content filled in according to the template requirements.

[0045] Preferably, the specific process of iteratively traversing the input enterprise text sequence X from the first word to the last word in S42 is as follows:

[0046] S421. For each position i, check whether x i Is the start instruction marker, if so, construct a caption P for the current paragraph i , contains the text sequence from the previous start instruction marker to the current position, specifically expressed as in, Indicates the first symbol in the text sequence from the previous start instruction to the current position i, x i-1 Indicates the symbol before the current position i;

[0047] S422, calculate the length L of the current paragraph i , specifically expressed as L i =|P i |;

[0048] S423, using the model generation method, decode the current paragraph to obtain the generated result R i ;

[0049] S424, for each keyword in the template, generate result R i Select the word that best matches it and add it to the caption of the current paragraph. i middle;

[0050] S425, if the keyword has multiple matching words, fill in the words in a loop until a unique match is found;

[0051] S426. Finally, fill in the caption P i Added to the generated text sequence D.

[0052] Preferably, the calculation expression of cosine similarity in step 5 is as follows:

[0053]

[0054] Where A represents the vocabulary vector in the enterprise information text sequence, B represents the node vector of the government policy relationship network, and θ represents the angle between the vocabulary vector in the enterprise information text sequence and the node vector of the government policy relationship network.

[0055] Preferably, the anomaly detection algorithm includes One-class SVM or One-shot SVM technology.

[0056] Therefore, the present invention adopts the above-mentioned system and method for interpreting digital government policies, which has the following beneficial effects:

[0057] (1) The latest policies are transformed into a relationship network of government policies, making it easier for enterprises to view and understand the interpreted information. The entire process is completed automatically without human intervention;

[0058] (2) Enterprises can input their own relevant information and independently obtain the final matching results, giving the conditions or basis for not meeting or meeting the requirements, thereby reducing the manpower input for policy application and review;

[0059] (3) Through feature search methods, the process of converting corporate information into text sequences can be better controlled, making the generated text more accurate and in line with expectations.

[0060] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 A schematic diagram of the structure of a system for interpreting digital government policies according to the present invention;

[0062] Figure 2 This is an overall flow chart of a method for interpreting digital government policies according to the present invention. DETAILED DESCRIPTION

[0063] The following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] See also Figure 1 , a digital government policy interpretation system, including a policy data acquisition and processing module, a knowledge graph construction module, a text representation model module, a relationship modeling module, a semantic understanding module, a reasoning module and a visualization module. The specific contents of each module are as follows:

[0065] Policy data acquisition and processing module: collects the latest government policy documents from the Internet or other sources, pre-processes them through natural language processing technology, and generates processable text data;

[0066] Knowledge graph construction module: Based on the content of the policy text, the key elements of the policy are extracted using extraction, induction, and entity recognition methods to construct a knowledge graph;

[0067] Text representation model module: converts policy text data into numerical feature vectors;

[0068] Relationship modeling module: defines entities and their relationships in the knowledge graph as graph data and builds a graph model of the relationship between entities;

[0069] Semantic understanding module: semantically understands the input enterprise information and converts it into a text sequence through feature search;

[0070] Reasoning module: Reasoning based on the input enterprise information and government policies to determine the matching degree between enterprise information and government policies;

[0071] Visualization module: Visualize the generated results and present them to users intuitively.

[0072] See also Figure 2 , a method for interpreting digital government policies, comprising the following steps:

[0073] Step 1: Policy acquisition and cleaning: Collect the latest government-related policy documents, remove irrelevant information through natural language processing technology, and organize them into structured data; the specific process is as follows:

[0074] S11. Use web crawlers to obtain policy documents from official websites, policy and regulation databases, news reports, and social media, and pre-process the crawled data, including removing HTML tags, special characters, and format conversion;

[0075] S12. Clean the data preprocessed in S11, including grammar checking, content filtering, and duplicate removal;

[0076] S13. Process the data cleaned in S12 using natural language processing to generate structured data, including word segmentation, stemming, lemmatization, part-of-speech tagging, stop word removal, word frequency statistics, word vector generation, entity recognition, and attribute extraction. The specific details are as follows:

[0077] S131. Word segmentation: Segment the text in the policy document into a sequence of words;

[0078] S132. Stemming: Extract the stem of each word and remove the affixes;

[0079] S133. Lemmatization: Restore the stem to the specific word form;

[0080] S134. Part-of-speech tagging: Tag the part of speech of each word, including nouns, verbs, and adjectives;

[0081] S135. Stop word removal: Remove common meaningless words, including "de" and "le";

[0082] S136. Word frequency statistics: Count the word frequency of each word and find the word with the highest frequency;

[0083] S137. Word vector generation: Vectorize the words in the policy document using the BERT or Word2Vec word vector model;

[0084] S138. Entity recognition: Perform clustering operations on the word vectors to cluster the words in the policy document into different entities;

[0085] S139. Attribute extraction: Starting from the entity, extract the attributes of the entity according to the entity type and context to generate structured data.

[0086] Step 2. Knowledge graph construction: Create a knowledge graph using the policy content, including elements such as policy objectives, subjects, measures, and influencing factors. Use these elements as nodes and the associations between the elements as edges;

[0087] Step 3. Semantic parsing and linking: Connect the keywords and entities in the policy content to the corresponding nodes in the knowledge graph through word vectors and entity linking technologies to form a relationship network of government policies. Among them, the formed relationship network of government policies is where represents the node set, represents the edge set.

[0088] Step 4: Input enterprise information and convert the input enterprise information into a text sequence through feature search; enterprise information includes industry classification, scale, technical expertise, and business model; and enterprise information is captured from network data through crawler technology; in the feature search method, we first define a series of templates, each of which describes the structure and requirements of the expected generated text. Then, in the process of generating text, we use these templates as guidance to guide the model to generate text that meets the requirements of the template. Specifically, we first identify key information in the input text, such as start tags or specific words, as filling points for the template. Then, according to the requirements of the template, the model will give priority to filling in these key information when generating text to ensure that the generated text meets the structure of the template. The feature search method can not only improve the consistency and structure of the generated text, but also ensure that the generated text is related to a specific field or task, thereby improving the relevance and applicability of the generated results. By designing the feature search method, we can better control the process of converting enterprise information into a text sequence, making the generated text more accurate and in line with expectations; among them, the specific process of converting the input enterprise information through feature search is as follows:

[0089] S41, initialize an empty generated text sequence D, represented by D = {};

[0090] S42, iterate the input enterprise text sequence X from the first word to the last word; the specific process is as follows:

[0091] S421. For each position i, check whether x i Is the start instruction marker, if so, construct a caption P for the current paragraph i , contains the text sequence from the previous start instruction marker to the current position, specifically expressed as in, Indicates the first symbol in the text sequence from the previous start instruction to the current position i, x i-1 Indicates the symbol before the current position i;

[0092] S422, calculate the length L of the current paragraph i , specifically expressed as L i =|Pi|;

[0093] S423, using the model generation method, decode the current paragraph to obtain the generated result R i ;

[0094] S424, for each keyword in the template, generate result R i Select the word that best matches it and add it to the caption of the current paragraph. i middle;

[0095] S425, if the keyword has multiple matching words, fill in the words in a loop until a unique match is found;

[0096] S426. Finally, fill in the caption P i Added to the generated text sequence D.

[0097] S43. Obtain the generated complete text sequence D, including the text content filled in according to the template requirements.

[0098] Step 5: Convert each word in the enterprise information text sequence and each node in the government policy relationship network into a vector through the word embedding model, and then compare the matching degree between the enterprise information and the government policy through cosine similarity calculation; the calculation expression of cosine similarity is as follows:

[0099]

[0100] Where A represents the vocabulary vector in the enterprise information text sequence, B represents the node vector of the government policy relationship network, and θ represents the angle between the vocabulary vector in the enterprise information text sequence and the node vector of the government policy relationship network.

[0101] Step 6: Determine the degree of similarity. For cases where the similarity is low or cannot be fully matched, anomaly detection algorithms are used for secondary judgment to avoid misjudgment. Enterprises that do not meet the conventional declaration information after secondary judgment are identified and warned, and suggestions are given. For cases with high similarity, a final grade report is given. The anomaly detection algorithm includes One-class SVM or One-shot SVM technology.

[0102] Step 7: Visualize the generated results, and finally the enterprise can export them by itself.

[0103] Therefore, the present invention adopts the above-mentioned system and method for interpreting digital government policies, and forms a relationship network of government policies with the latest policies, so that enterprises can view and understand the interpreted information. Subsequently, the enterprises can input their own relevant information into the system to independently obtain the final matching results, and give the conditions or basis that are not met or met, thereby reducing the manpower investment in policy declaration and review; finally, through the feature search method, the process of converting enterprise information into text sequences can be better controlled, so that the generated text is more accurate and in line with expectations, and automatic analysis of enterprise information can be realized, thereby providing enterprises with targeted decision-making support.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A system for interpreting digital government policies, characterized by: It includes policy data acquisition and processing module, knowledge graph construction module, text representation model module, relationship modeling module, semantic understanding module, reasoning module and visualization module.

2. A digital government policy interpretation system according to claim 1, characterized in that: The specific contents of the policy data acquisition and processing module, knowledge graph construction module, text representation model module, relationship modeling module, semantic understanding module, reasoning module and visualization module are as follows: Policy data acquisition and processing module: collects the latest government policy documents from the Internet or other sources, pre-processes them through natural language processing technology, and generates processable text data; Knowledge graph construction module: Based on the content of the policy text, the key elements of the policy are extracted using extraction, induction, and entity recognition methods to construct a knowledge graph; Text representation model module: converts policy text data into numerical feature vectors; Relationship modeling module: defines entities and their relationships in the knowledge graph as graph data and builds a graph model of the relationship between entities; Semantic understanding module: semantically understands the input enterprise information and converts it into a text sequence through feature search; Reasoning module: Reasoning based on the input enterprise information and government policies to determine the matching degree between enterprise information and government policies; Visualization module: Visualize the generated results and present them to users intuitively.

3. A method for interpreting digital government policies as described in any one of claims 1-2, characterized in that: The following steps are involved: Step 1: Policy acquisition and cleaning; Collect the latest government-related policy documents, remove irrelevant information through natural language processing technology, and organize them into structured data; Step 2: Knowledge graph construction; Use policy content to create a knowledge graph, including elements of policy goals, subjects, measures, and influencing factors, with each element as a node and the relationships between elements as edges; Step 3: Semantic parsing and linking: Through the word vector and entity linking technology, the keywords and entities in the policy content are connected with the corresponding nodes in the knowledge graph to form a relationship network of government policies; Step 4: Input enterprise information and convert the input enterprise information into a text sequence through feature search; enterprise information includes industry classification, scale, technical expertise, and business model; and enterprise information is captured through network data using crawler technology; Step 5: Each word in the enterprise information text sequence and each node in the government policy relationship network are converted into vectors through the word embedding model, and then the matching degree between the enterprise information and the government policy is compared through cosine similarity calculation; Step 6: Determine the degree of similarity. For cases where the similarity is low or cannot be completely matched, anomaly detection algorithms are used for secondary judgment. Enterprises whose declared information does not conform to the conventional ones after secondary judgment are identified and warned, and suggestions are given. For those with high similarity, a final grade report is given; Step 7: Visualize the generated results, and finally the enterprise can export them by itself.

4. A method for interpreting digital government policies according to claim 3, characterized in that: The specific process of step 1 is as follows: S11. Use web crawlers to obtain policy documents from official websites, policy and regulation databases, news reports, and social media, and pre-process the crawled data, including removing HTML tags, special characters, and format conversion; S12. Clean the data preprocessed in S11, including grammar checking, content filtering, and deduplication; S13. Process the data cleaned in S12 in natural language to generate structured data; Including word segmentation, stemming, lemmatization, part-of-speech tagging, stop word removal, word frequency statistics, word vector generation, entity recognition, and attribute extraction.

5. A method for interpreting digital government policies according to claim 4, characterized in that: The specific content of S13 is as follows: S131. Word segmentation: Segment the text in the policy document into a sequence of words; S132. Stemming: Extract the stem of each word and remove the affixes; S133. Lemmatization: Restore the stem to the specific word form; S134. Part-of-speech tagging: Tag the part of speech of each word, including nouns, verbs, and adjectives; S135. Stop word removal: Remove common meaningless words, including "de" (的) and "le" (了); S136. Word frequency statistics: Count the word frequency of each word and find the word with the highest frequency; S137. Word vector generation: Vectorize the words in the policy document using the BERT or Word2Vec word vector model; S138. Entity recognition: Perform clustering operations on the word vectors to cluster the words in the policy document into different entities; S139. Attribute extraction: Starting from the entity, extract the attributes of the entity according to the entity type and context to generate structured data.

6. A method for interpreting digital government policies according to claim 5, characterized in that: The relationship network of government policies in step 3 is in represents a node set, Represents a set of edges.

7. A method for interpreting digital government policies according to claim 6, characterized in that: The specific process of converting the input enterprise information through feature search in step 4 is as follows: S41. Initialize an empty generated text sequence D, denoted as D = {}; S42. Iteratively traverse the input enterprise text sequence X from the first word to the last word; S43. Obtain the generated complete text sequence D, including the text content filled according to the template requirements.

8. A method for interpreting digital government policies according to claim 7, characterized in that: The specific process of iteratively traversing the input enterprise text sequence X from the first word to the last word in S42 is as follows: S421. For each position i, check whether x i Is the start instruction marker, if so, construct a caption P for the current paragraph i , contains the text sequence from the previous start instruction marker to the current position, specifically expressed as in, Indicates the first symbol in the text sequence from the previous start instruction to the current position i, x i-1 Indicates the symbol before the current position i; S422, calculate the length L of the current paragraph i , specifically expressed as L i =|P i |; S423, using the model generation method, decode the current paragraph to obtain the generated result R i ; S424, for each keyword in the template, generate result R i Select the word that best matches it and add it to the caption of the current paragraph. i middle; S425. If there are multiple matching words for the keyword, loop and fill until a unique match is found; S426. Finally, fill in the caption P i Added to the generated text sequence D.

9. A digital government policy interpretation system according to claim 8, characterized in that: The calculation expression of the cosine similarity in step 5 is as follows: In the formula, A represents the vocabulary vector in the enterprise information text sequence, B represents the node vector of the government affairs policy relationship network, and θ represents the angle between the vocabulary vector in the enterprise information text sequence and the node vector of the government affairs policy relationship network.

10. A digital government policy interpretation system according to claim 9, characterized in that: The anomaly detection algorithm includes One-class SVM or One-shot SVM technology.

Images