Government affair text auditing method and system based on knowledge graph reasoning

Abstract

The invention discloses a government affair text auditing method and system based on knowledge graph reasoning. The method and system are used for improving the accuracy and the intelligent level of government affair text auditing. The method comprises the following steps: acquiring an original policy text, and performing semantic analysis processing on the original policy text by utilizing a two-channel semantic disambiguation engine; fusing semantic analysis results of the rule channel and the neural channel to obtain a standardized entity set subjected to semantic disambiguation and structured attribute calibration; mapping the standardized entity set, the case original text data and the declaration material data into a knowledge graph; a graph neural network model is utilized to learn node embedding representation in the knowledge graph so as to realize semantic alignment among the normalized entity set, the case original text data and the declaration material data; on the knowledge graph, performing compliance evaluation by using symbol reasoning based on a predefined logic rule and neural reasoning based on node embedding representation or a graph path; and generating a government affair text auditing result according to a compliance evaluation result.

Description

Technical Field

[0001] This application relates to the interdisciplinary fields of natural language processing and intelligent government affairs, and in particular to a method and system for reviewing government documents based on knowledge graph reasoning. Background Technology

[0002] Government document review is a crucial part of the daily management and public services of government departments. It involves examining various policy documents, legal provisions, and application materials to ensure they comply with relevant laws, regulations, and policy requirements. With the continuous improvement of e-government informatization, the number of documents to be processed has increased dramatically, and traditional review methods that rely primarily on manual processes are no longer sufficient to meet the demand.

[0003] Existing technologies have attempted to leverage information technology to assist in the review of government documents, such as using Natural Language Processing (NLP). These technologies typically employ rule-based matching, keyword recognition, or shallow semantic analysis to identify information within the text or attempt to establish connections between different texts. However, due to the highly specialized and rigorous nature of government documents, especially policies and regulations, and the inevitable semantic complexity and diversity they contain, traditional NLP models struggle to accurately link legal provisions, case studies, and the logical chains of application materials. For example, they cannot dynamically parse the contextual meaning of vague qualifiers such as "shall" or "in principle" in policy clauses. In other words, existing automated government document review methods have limited capabilities in understanding complex contexts, handling ambiguous expressions, and linking different types of government documents, thus failing to meet application requirements. Summary of the Invention

[0004] This application provides a method and system for reviewing government documents based on knowledge graph reasoning, which can improve the accuracy and intelligence level of government document review.

[0005] The first aspect of this application provides a method for reviewing government documents based on knowledge graph reasoning, including: The original policy text is obtained and semantic parsing is performed on the original policy text using a dual-channel semantic disambiguation engine. The dual-channel semantic disambiguation engine includes a rule channel built based on a government affairs domain ontology library and a neural channel built based on a pre-trained language model. By fusing the semantic parsing results of the rule channel and the neural channel, a normalized entity set that has undergone semantic disambiguation and structured attribute labeling is obtained; Obtain application material data and original case text data related to the original policy text, and map the standardized entity set, the original case text data, and the application material data into a knowledge graph; A graph neural network model is used to learn the node embedding representations in the knowledge graph in order to achieve semantic alignment between the standardized entity set, the original case data, and the application material data. On the knowledge graph, symbolic reasoning based on predefined logical rules and neural reasoning based on the node embedding representation or graph path are used to conduct a compliance assessment on the degree of conformity between the application material data and the original policy text. Based on the results of the compliance assessment, a government document review result is generated, which includes the compliance judgment conclusion and the corresponding supporting evidence.

[0006] Optionally, the rule channel is used to perform deterministic semantic parsing on legal terms and normative expressions that conform to a preset pattern in the original policy text using the government affairs domain ontology library; The neural pathway is used to capture the dynamic semantics of ambiguous qualifiers or non-standard expressions in the original policy text through deep context modeling based on an improved ERNIE model.

[0007] Optionally, the semantic parsing results of fusing the rule channel and the neural channel yield a normalized entity set that has undergone semantic disambiguation and structured attribute labeling, including: Calculate the parsing confidence of semantic parsing results that are simultaneously identified by the rule channel and the neural channel and point to the same text content; The decision weights of the rule channel and the neural channel are determined according to a preset disambiguation weight allocation strategy; By combining the analytical confidence and the decision weight, the semantic parsing results of the rule channel and the neural channel are selected, merged, or modified to obtain a normalized entity set that has undergone semantic disambiguation and structured attribute labeling.

[0008] Optionally, the step of acquiring application material data and original case text data related to the original policy text, and mapping the standardized entity set, the original case text data, and the application material data to a knowledge graph, includes: The entities representing policy clauses and legal concepts in the standardized entity set and their relationships are converted into nodes and edges of the policy dimension in a knowledge graph; Information is extracted from the original case data related to the original policy text, extracting target information such as case focus, judgment reasoning, and cited legal provisions. The target information is then converted into nodes of the case dimension in the knowledge graph, and associations are established with the policy dimension nodes. The content of the application materials data is parsed into structured data items, and the data items are converted into nodes and attributes of the material dimension in the knowledge graph.

[0009] Optionally, the step of learning the node embedding representations in the knowledge graph using a graph neural network model to achieve semantic alignment between the normalized entity set, the original case data, and the application material data includes: By utilizing the attention and aggregation mechanisms built into the graph neural network model, the feature information of each entity node from its heterogeneous neighborhood under the policy dimension, the case dimension, and the material dimension is iteratively aggregated to generate a node embedding representation that can capture cross-dimensional contextual dependencies. The heterogeneous neighborhood contains entity nodes of different dimensions and diverse relational edges. The node embedding representation is optimized and learned based on a preset loss function until the core semantic concepts of data in each dimension achieve consistent representation in the embedding space. The loss function is used to maximize the representation similarity of semantically related entity nodes in the embedding space, while minimizing the representation similarity of unrelated entity nodes.

[0010] Optionally, the compliance assessment of the degree of conformity between the application material data and the original policy text is performed on the knowledge graph using symbolic reasoning based on predefined logical rules and neural reasoning based on the node embedding representation or graph path, including: Extract structured predefined logical rules from the provisions of the original policy text or from the constraints defined in the government affairs ontology library; On the knowledge graph, symbolic reasoning is used to perform preliminary verification on the compliance checkpoints in the application material data that can directly correspond to the predefined logical rules, and output preliminary conclusions of certain compliance, partial compliance, non-compliance, or need for further analysis. For cases where the preliminary conclusion is partially compliant, requires further analysis, or involves semantic complexity in the compliance checkpoints, neural reasoning methods are used to perform deep semantic matching by analyzing the node embedding representation or graph path to generate auxiliary evaluation opinions.

[0011] Optionally, generating the government document review result, which includes a compliance judgment conclusion and corresponding supporting evidence, based on the compliance assessment results includes: The preliminary conclusions output by the symbolic reasoning and the auxiliary evaluation opinions generated by the neural reasoning are integrated to obtain a comprehensive evaluation result, and the final compliance judgment conclusion of the application material data is determined based on the comprehensive evaluation result; From the knowledge graph and the processes of symbolic reasoning and neural reasoning, obtain supporting evidence directly related to the final compliance judgment conclusion; The final compliance judgment and supporting evidence are organized and formatted to generate the government document review result.

[0012] The second aspect of this application provides a government document review system based on knowledge graph reasoning, including: The acquisition unit is used to acquire the original policy text and perform semantic parsing processing on the original policy text using a dual-channel semantic disambiguation engine. The dual-channel semantic disambiguation engine includes a rule channel built based on a government domain ontology library and a neural channel built based on a pre-trained language model. The fusion unit is used to fuse the semantic parsing results of the rule channel and the neural channel to obtain a normalized entity set after semantic disambiguation and structured attribute labeling; The mapping unit is used to acquire application material data and case text data related to the original policy text, and to map the standardized entity set, the case text data and the application material data to the knowledge graph; The alignment unit is used to learn the node embedding representation in the knowledge graph using a graph neural network model, so as to achieve semantic alignment between the normalized entity set, the original case data, and the application material data. The reasoning unit is used to perform a compliance assessment on the knowledge graph by employing symbolic reasoning based on predefined logical rules and neural reasoning based on the node embedding representation or graph path to assess the degree of conformity between the application material data and the original policy text. The generation unit is used to generate a government document review result that includes a compliance judgment conclusion and corresponding supporting evidence based on the results of the compliance assessment.

[0013] A third aspect of this application provides a government document review device based on knowledge graph reasoning, the device comprising: Processor, memory, input / output units, and bus; The processor is connected to the memory, the input / output unit, and the bus; The memory stores a program, which the processor invokes to execute the first aspect and any optional knowledge graph-based reasoning method for reviewing government documents.

[0014] The fourth aspect of this application provides a computer-readable storage medium storing a program that, when executed on a computer, performs the first aspect and any optional one of the first aspects of the knowledge graph-based reasoning method for reviewing government documents.

[0015] As can be seen from the above technical solutions, this application has the following advantages: To address the ambiguous wording of policy provisions, this application employs a dual-channel semantic disambiguation engine. By combining the deterministic parsing of normative expressions through a rule-based channel with the dynamic semantic capture of context-based neural channels, it can more accurately understand the actual meaning of policy provisions, especially ambiguous qualifiers. Furthermore, this application constructs a multi-dimensional knowledge graph based on legal provisions, cases, and application materials, mapping scattered government information into a unified knowledge framework. Effective fusion and semantic alignment of information from different dimensions are achieved through learning node embedding representations. Within this unified framework, symbolic reasoning is used for hard condition verification, and neural reasoning is used for complex association pattern mining. This accurately constructs, verifies, and generates complete logical chains and compliance judgments between legal provisions, cases, and application materials, thereby overcoming the shortcomings of traditional NLP models in handling the complexity of government documents and constructing reliable logical chains, significantly improving the accuracy and intelligence level of government document review. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A schematic flowchart of an embodiment of the government document review method based on knowledge graph reasoning provided in this application; Figure 2 A flowchart illustrating another embodiment of the government document review method based on knowledge graph reasoning provided in this application; Figure 3 A schematic diagram of the structure of an embodiment of the government document review system based on knowledge graph reasoning provided in this application; Figure 4 A schematic diagram of an embodiment of the government document review device based on knowledge graph reasoning provided in this application. Detailed Implementation

[0018] This application provides a method and system for reviewing government documents based on knowledge graph reasoning, which can improve the accuracy and intelligence level of government document review.

[0019] It should be noted that the knowledge graph-based reasoning method for reviewing government documents provided in this application can be applied to terminals as well as servers. For example, a terminal can be a smartphone, computer, tablet, smart TV, smartwatch, portable computer terminal, or a desktop computer, etc. For ease of explanation, this application uses the system running on the terminal as the execution subject for illustration.

[0020] Please see Figure 1 , Figure 1 An embodiment of the government document review method based on knowledge graph reasoning provided in this application includes: 101. Obtain the original policy text and use a dual-channel semantic disambiguation engine to perform semantic parsing on the original policy text. The dual-channel semantic disambiguation engine includes a rule channel built based on the government domain ontology and a neural channel built based on a pre-trained language model. First, the system needs to acquire and parse the original policy text as the basis for subsequent compliance review. The original policy and regulatory text is the direct basis for judging whether the application materials are compliant. However, the original policy text is usually unstructured natural language and often contains legal terminology, normative expressions, and ambiguous or vague qualifying words such as "should" and "in principle." Traditional single natural language processing methods are difficult to understand these texts accurately and comprehensively, especially in dynamically parsing the contextual meaning of ambiguous words. Based on this, this application chooses to use a dual-channel semantic disambiguation engine to perform semantic parsing processing on the original policy text, transforming the natural language text in the original policy text into structured semantics that can be understood by computers. The design concept of this dual-channel semantic disambiguation engine is to combine the advantages of different parsing mechanisms to deal with the characteristics of government documents that are both normative and complex. The dual-channel engine mainly includes two parallel parsing channels: a rule channel and a neural channel, which are described below: The implementation of the rule-based channel relies on a pre-built ontology library for the government domain. This library is a structured knowledge base containing government domain concepts, terms, relationships, and rules. The rule-based channel utilizes the ontological knowledge and pre-defined parsing rules within this library to perform deterministic semantic parsing on legal terms and normative expressions conforming to normative patterns in original policy texts, ensuring accurate understanding of standard terminology. Combined with techniques such as dependency parsing, the rule-based channel can also assist in identifying the modifiers of fuzzy qualifiers and trigger checks on relevant exception conditions when encountering specific fuzzy words. The advantage of the rule-based channel lies in its high accuracy and interpretability when processing standardized and explicit language.

[0021] The neural pathway is implemented based on pre-trained language model technology, specifically using the ERNIE model, an improved ERNIE model optimized and trained for the political and legal domain. The ERNIE model is a pre-trained language model that, compared to traditional pre-trained models (such as BERT), incorporates knowledge units such as words, entities, and phrases in addition to learning the language patterns of the text itself during the pre-training phase. This enhances the model's ability to understand language, particularly semantics and conceptual relationships. The improved ERNIE model can be trained on a large number of political and legal texts to improve its language comprehension capabilities. The neural pathway utilizes deep contextual modeling capabilities to analyze the complex relationships between words in sentences and paragraphs, more accurately capturing the dynamic semantics in political texts that depend on complex contexts. It is particularly effective for vague qualifying words or non-standard expressions such as "should" and "in principle," inferring more precise meanings based on their contextual information. The advantage of the neural pathway lies in its flexibility and context-dependent language processing, compensating for the shortcomings of rule-based pathways in understanding complex contexts.

[0022] 102. By fusing the semantic parsing results of the rule-based channel and the neural channel, a normalized entity set that has undergone semantic disambiguation and structured attribute labeling is obtained; Since the two channels perform semantic parsing based on different principles, their outputs may differ, conflict, or be complementary. Therefore, it is necessary to fuse the semantic parsing results of the rule-based channel and the neural channel, comprehensively utilizing the certainty of the rule-based channel in handling normative expressions and the flexibility of the neural channel in capturing contextual semantics, ultimately generating a single, accurate semantic parsing result. Specifically, there are several possible ways to achieve this fusion. One approach is based on preset priority rules; for example, for specific types of legal terms, the parsing result of the rule-based channel can be set as the preferred choice. Another approach is to use a voting mechanism, increasing the confidence level of each channel if both channels provide the same parsing for the same text segment. Alternatively, a weighted combination can be performed based on the confidence scores of each channel's output, scoring different results and selecting the parsing with the highest score. Or, an independent machine learning model can be trained to learn how to optimally combine the outputs of the two channels. These methods can all be implemented algorithmically, and specific implementations are not limited here. The logic of the fusion is to take the preliminary results of the two channels as input and derive a normalized set of entities through logical judgment, calculation, or model prediction.

[0023] This normalized entity set contains all the key semantic information extracted from the original policy text. This information has undergone semantic disambiguation to resolve potential ambiguities and has been labeled with structured attributes. The normalized entity set exists in a unified, machine-friendly format and can be directly used as input data for subsequent steps in building a multi-dimensional knowledge graph. Suppose the original policy text fragment is: “According to Article 8 of the ‘Management Measures,’ the applicant should be an enterprise registered in our city, and its registered capital shall not be less than RMB 500,000.” This text contains: policy reference, entity (applicant, enterprise, our city), attribute (registered capital), value (RMB 500,000), qualifiers (should be, shall not be less than), and location ambiguity (our city). The rule channel may identify POLICY("Article 8 of the ‘Management Measures’"), ENTITY_TYPE("enterprise"), LOCATION("our city"), CONCEPT("registered capital"), AMOUNT("RMB 500,000"), and, based on the rules, identify that “should be” and “shall not be less than” represent “requirements” or “restrictions.” The neural pathway, through contextual modeling, may identify that ENTITY("applicant unit") refers to ENTITY_TYPE("enterprise"), LOCATION("our city") is a location entity, and understand that "registered capital shall not be less than RMB 500,000" is a numerical requirement for "enterprise," and that "shall be" represents a qualification requirement. Finally, the fused, normalized entity set decomposes the information in the original text into independent entities. The specific content of each entity is extracted and stored as its attributes, the constraints in the text are parsed and stored as structured attribute annotations for the registered capital entity, and the semantic relationships between entities are identified and represented as relations.

[0024] 103. Obtain application material data and original case text data related to the original policy text, and map the standardized entity set, original case text data and application material data into a knowledge graph; The core of this step is to aggregate and integrate all the information sources required for government document review—preliminarily processed policy information (the standardized entity set output from step 102), case information to aid in policy understanding (original case data), and application information to be reviewed (application material data)—into a unified data model, i.e., constructing a multi-dimensional knowledge graph. Original case data provides the practical application and interpretation of policies, while application material data contains specific factual information about the applicant or the reviewed entity. Because government document review involves diverse data sources, such as policy provisions, legal cases, and application forms, these sources vary in format and semantic expression, and are often isolated from each other. Mapping this data to a knowledge graph transforms this scattered and heterogeneous information into a structured, interconnected unified model, resulting in a comprehensive knowledge graph containing policy norms, case facts, and application details. This provides a foundation for efficient and accurate correlation queries and logical reasoning between different information types.

[0025] 104. Utilize graph neural network models to learn node embedding representations in knowledge graphs to achieve semantic alignment between standardized entity sets, case text data, and application material data; Although step 103 integrates policy, case, and material information into a single graph, simply establishing structural connections is insufficient for a computer to understand the deep semantic relationships between entities representing the same concept or highly related entities across different dimensions. For example, the regulation node regarding "registered capital" in the policy dimension and the "registered capital" attribute node of a specific enterprise entity in the material dimension may be structurally distant, but they are semantically closely related. The purpose of semantic alignment is to learn how to bring entities representing the same or related concepts across different dimensions closer together in the numerical representation space, thereby enabling the computer to identify the implicit semantic relationships between them.

[0026] The core technology for achieving semantic alignment is the Graph Neural Network (GNN) model. GNNs are neural network models capable of learning directly on graph-structured data, and in this application, they are used to learn the embedding representation of each node in a knowledge graph. A node embedding representation is a numerical vector representation of each node in a multi-dimensional space. This vector captures the node's own information and its structure and semantic context within the entire graph. The learning process enables the GNN model to analyze the connections between nodes and the information they carry, mapping nodes with similar semantics or representing related concepts to closer positions in the embedding space. This process achieves semantic alignment between data from different dimensions, allowing entities or information from different sources (policies, cases, materials) that are semantically related to each other to have similar numerical representations in the embedding space, thus improving the usability of the knowledge graph.

[0027] 105. On the knowledge graph, use symbolic reasoning based on predefined logical rules and neural reasoning based on node embedding representation or graph path to conduct a compliance assessment on the degree of conformity between the data in the application materials and the original policy text. After completing the information preparation steps 101-104, actual logical judgment is required to determine whether the application materials meet policy requirements. This involves comparing the specific facts in the application materials with the normative conditions in the policy text, and considering relevant case interpretations. Due to the complexity of government document review, simple matching is insufficient; a combination of reasoning methods is necessary. This application chooses to employ a hybrid reasoning approach on the constructed knowledge graph, including symbolic reasoning based on predefined logical rules and neural reasoning based on node embedding representations or graph paths. The knowledge graph, as a unified data carrier, provides the foundation for complex associations and reasoning.

[0028] Symbolic reasoning is a reasoning method based on formal logic and predefined rules. Specifically, it uses predefined logical rules to make deterministic judgments about facts in a knowledge graph. For example, if the rule stipulates that "registered capital must be greater than or equal to 500,000 yuan," symbolic reasoning will directly check whether the node in the knowledge graph representing the registered capital value in the application materials meets this condition. The advantages of symbolic reasoning lie in its transparent process, deterministic results, and strict logical validity.

[0029] Neural reasoning is a reasoning method based on machine learning models. It can identify and utilize complex patterns, implicit associations, or semantically similar connections that have not been distilled into explicit symbolic rules. It is suitable for handling semantically ambiguous situations, situations requiring contextual inference, or situations involving complex relational paths.

[0030] By comprehensively applying these two reasoning mechanisms on a knowledge graph, this application is able to conduct a comprehensive and in-depth compliance assessment of the degree of conformity between the data in the application materials and the original policy text. In this application, symbolic reasoning is responsible for handling explicit hard conditions to ensure basic compliance; neural reasoning is responsible for handling complex and implicit semantics or relationships to provide more refined or auxiliary judgments.

[0031] 106. Generate government document review results that include compliance judgment conclusions and corresponding supporting evidence based on the compliance assessment results.

[0032] After obtaining the compliance assessment results, the system first integrates all assessment results and draws a clear compliance judgment conclusion based on this information. To ensure the final conclusion is persuasive and transparent, the system also needs to trace and obtain supporting evidence directly related to the final compliance judgment conclusion from the entire review process. This evidence comes from multiple sources, including but not limited to: original application material data and policy clause nodes represented in the knowledge graph; specific predefined logical rules triggered or not triggered during symbolic reasoning; key graph paths, semantic relationships, or patterns identified during neural reasoning; and relevant case nodes and their association with policies or materials. The process of obtaining evidence needs to be able to trace the reasoning chain in reverse to locate the key facts and rules leading to the final conclusion. Finally, the system needs to organize and format the determined final compliance judgment conclusion and the obtained supporting evidence to generate government document review results that are easy for users to understand. Specifically, the conclusion and evidence can be filled into a preset report template to generate structured data files (such as XML or JSON), or visualized on the user interface. Supporting evidence usually needs to be presented in a clear way, such as listing the violated clauses, relevant application data items, case descriptions to assist in the judgment, or key relational graph fragments identified during the reasoning process.

[0033] In this embodiment, to address the ambiguous expressions in policy clauses, this application employs a dual-channel semantic disambiguation engine. By combining the deterministic parsing of normative expressions through a rule-based channel with the dynamic semantic capture of context-based neural channels, it can more accurately understand the actual meaning of policy clauses, especially vague qualifiers. Furthermore, this application constructs a multi-dimensional knowledge graph based on legal provisions, cases, and materials, mapping scattered government information into a unified knowledge framework. Effective fusion and semantic alignment of information from different dimensions are achieved through learning node embedding representations. On this unified framework, symbolic reasoning is used for hard condition verification, and neural reasoning is used for complex association pattern mining. This accurately constructs, verifies, and generates complete logical chains and compliance judgments between legal provisions, cases, and application materials, thereby overcoming the shortcomings of traditional NLP models in handling the complexity of government documents and constructing reliable logical chains, significantly improving the accuracy and intelligence level of government document review.

[0034] The following provides a detailed explanation of the government document review method based on knowledge graph reasoning provided in this application. Please refer to [link / reference]. Figure 2 , Figure 2 Another embodiment of the knowledge graph-based reasoning method for government document review provided in this application includes: 201. Obtain the original policy text and use a dual-channel semantic disambiguation engine to perform semantic parsing on the original policy text. The dual-channel semantic disambiguation engine includes a rule channel built based on the government domain ontology and a neural channel built based on a pre-trained language model. In this embodiment, step 201 is similar to step 101 in the previous embodiment, and will not be described again here.

[0035] 202. Calculate the parsing confidence of semantic parsing results that are simultaneously identified by the rule channel and the neural channel and point to the same text content; Steps 202-204 involve fusing the semantic parsing results from the rule-based channel and the neural channel. The two channels of the dual-channel engine may both identify entities, relations, or attributes for the same part of the original text (e.g., the same word, phrase, or sentence fragment). Therefore, the system needs to calculate a parsing confidence score for each of the parsing results identified by the rule-based channel and the neural channel, each pointing to the same segment of the original text. The parsing confidence score is a numerical value used to quantify the channel's confidence level in a specific parsing result (e.g., the identified entity type, relation type, or attribute value). The parsing confidence score of the rule-based channel can be calculated based on the accuracy of the matched rules, the range of text covered, or preset rule weights. The parsing confidence score of the neural channel can be directly determined using the probability scores of the prediction results from the underlying pre-trained language model.

[0036] 203. Determine the decision weights of the rule channel and the neural channel according to the preset disambiguation weight allocation strategy; Beyond the confidence level of specific parsing results, the overall fusion strategy may require assigning higher priority to rule-based or neural channels. For example, rule-based channels might be favored when processing standardized legal terminology, while neural channels might be favored when processing ambiguous or colloquial expressions. Specifically, this can be achieved using a pre-designed disambiguation weight allocation strategy. This strategy is an algorithm or set of rules that determines the decision weights of rule-based and neural channels in this fusion decision based on the current parsing context (e.g., the type of text paragraph being processed, the type of entity identified). The decision weight is also a numerical value, representing the relative importance of the results from each channel when considering the parsing confidence level. This strategy can be configured as a fixed weight allocation strategy during system initialization or dynamically adjusted as needed during operation. For example, the strategy might stipulate that rule-based channels have higher decision weights for terms explicitly defined in the ontology, while neural channels have higher decision weights for complex sentence structures.

[0037] 204. By comprehensively analyzing confidence and decision weights, the semantic parsing results of the rule channel and neural channel are selected, merged or modified to obtain a normalized entity set after semantic disambiguation and structured attribute labeling. The system combines the parsing confidence score from step 202 and the decision weight from step 203 to make decisions on the semantic parsing results of the rule-based and neural channels. This process can be algorithmically evaluated to determine the overall score of the parsing results for each channel on the same text content (e.g., parsing confidence score multiplied by the decision weight, plus other factors). Based on this overall score, the system will take one or more of the following actions: Selection: Prioritize the parsing results of the channel with the highest overall score, discarding parsing results of other channels for the same content; Merging: If the parsing results of two channels are complementary (e.g., the rule channel identifies an entity, and the neural channel identifies a certain attribute or modifier of that entity), then merge this information into the same final structured representation; Correction: If the parsing result of one channel (e.g., a channel with high decision weight and high parsing confidence) conflicts with the parsing result of another channel, then use the former to correct or cover the latter.

[0038] After selecting, merging, or correcting the semantic parsing results from the two channels, a standardized set of entities that has undergone semantic disambiguation and structured attribute labeling can be obtained. This means that each entity, relation, and attribute in the set has a definite meaning (eliminating ambiguity) and is given a standard structure and attributes (e.g., entity type, attribute name, relation type), with a unified format, and can be directly used for subsequent knowledge graph construction.

[0039] The following example, using the policy text fragment "The applicant should have no major tax violations in the past three years," illustrates how the parsing results from the rule-based channel and the neural channel are fused through steps 202 to 204. Regarding the phrase "past three years" in the text, step 202 calculates the parsing confidence of the results from both the rule-based and neural channels (e.g., the rule-based channel, using exact matching, parses it as "counting back 36 months," with a confidence of 0.9; the neural channel, understanding the context, parses it as "the past three years," with a confidence of 0.7). Subsequently, step 203 determines the decision weights for the two channels based on a preset disambiguation weight allocation strategy (the strategy sets the rule weight at 0.6 and the neural weight at 0.4). Finally, in step 204, the system comprehensively considers these parsing confidences and decision weights to make a decision (e.g., calculating the weighted score: rule-based: 0.9 * 0.6 = 0.54; neural: 0.7 * 0.4 = 0.28). Because the weighted score of the rule channel (0.54) is higher, the system ultimately selects the parsing result of the rule channel, "counting back 36 months", and incorporates it into the final normalized entity set to complete the final semantic determination of the expression "the past three years".

[0040] 205. Convert the entities and their relationships representing policy clauses and legal concepts in the standardized entity set into nodes and edges of the policy dimension in the knowledge graph; Steps 205-207 map the standardized entity set, original case data, and application material data to a knowledge graph. First, the standardized entity set is traversed and processed. Each entity representing a policy clause or legal concept in the set is converted into a node representing the policy dimension in the knowledge graph. For example, "a certain regulation," Article 15, and registered capital may all become independent nodes. The relationships between the entities identified in the set (such as citation, definition, application, etc.) are then converted into edges connecting these corresponding nodes. For example, between the node representing "Article 15" and the node representing "High-tech Enterprise Certification," an edge indicating "applicable" might be established based on the original text relationship.

[0041] 206. Extract information from the original case data related to the original policy text, extract the target information such as the focus of the case, the reasoning of the judgment, and the cited legal provisions, and convert the target information into nodes of the case dimension in the knowledge graph, and establish the association with the policy dimension nodes; For the original case data, key elements closely related to policy application, such as the case focus, the reasoning behind the judgment, and the legal provisions explicitly cited in the case, are extracted through information extraction. This extracted structured target information is then converted into nodes in the case dimension of a knowledge graph. For example, a specific case record can become a node, its main point of contention can become another related node, and the key points of the reasoning behind the judgment can also serve as related nodes. It should be noted that this step analyzes the extracted cited legal provisions and establishes connections (edges) between the corresponding case dimension nodes and the policy dimension nodes constructed in step 205. For example, if a case cites a clause of a certain law, a "citation" or "interpretation" relationship edge will be created between the node representing the case and the policy node representing that clause.

[0042] 207. Parse the content of the application materials data into structured data items, and convert the data items into nodes and attributes of the material dimension in the knowledge graph; Finally, the application materials data are parsed and processed, especially the structured or semi-structured content such as tables, fields, and lists, which are converted into structured data items. For example, the values ​​of the "Registered Capital" field, the dates of the "Incorporation Date" field, and the numbers of the "Qualification Certificates" column are extracted from the enterprise application form. These structured data items are converted into nodes and attributes of the material dimension in a knowledge graph. Typically, the main body in the application materials (such as the applicant enterprise or individual) is represented as a node, while the specific data items extracted from the materials are represented as attributes of that node or related attribute nodes (for example, "Registered Capital" is an attribute of the enterprise node with a value of "5 million yuan"). At the same time, associations may also be established between material dimension nodes and policy dimension nodes based on the content of the materials (for example, connecting the node representing the enterprise's qualifications to the clause nodes in the policy regarding enterprise qualification requirements).

[0043] Through steps 205-207, the system processes and maps policy information, case information, and application material information to different dimensions of the knowledge graph. By establishing cross-dimensional connections, it ultimately constructs a multi-dimensional knowledge graph that integrates all information sources required for the review of government documents.

[0044] 208. Utilizing the built-in attention and aggregation mechanisms of graph neural network models, feature information of each entity node from its heterogeneous neighborhood under the policy dimension, case dimension, and material dimension is iteratively aggregated to generate node embedding representations that can capture cross-dimensional contextual dependencies. The heterogeneous neighborhood contains entity nodes of different dimensions and diverse relational edges. After the knowledge graph is constructed, each node has its initial feature information. However, to achieve semantic alignment, the node representation must not only reflect its own features but also incorporate its contextual information within the entire graph, especially the information of nodes and edges associated with it in different dimensions. Specifically, this can be achieved by utilizing the attention and aggregation mechanisms built into the Graph Neural Network (GNN) model to learn node embedding representations. Specifically, for each entity node in the knowledge graph, the GNN model examines its heterogeneous neighborhood. A heterogeneous neighborhood refers to the set of surrounding nodes directly connected to the node, containing entity nodes from different dimensions (policies, cases, materials) and diverse relational edges of different types connecting them (e.g., citation relationships, applicability relationships, attribute relationships, case association relationships, etc.). The aggregation mechanism is responsible for collecting feature information from each neighboring node and its connecting edges within its heterogeneous neighborhood. The attention mechanism, during aggregation, assigns different weights to different neighboring nodes or different types of relational edges, allowing the model to focus on neighborhood information that is more important for understanding the semantics of the current node. This aggregation process is typically iteratively performed in multiple layers, allowing nodes to aggregate information from more distant neighbors in the graph and capture a broader context of the graph structure. In this way, the model can generate node embedding representations for each node, and these vector representations can capture cross-dimensional contextual dependencies and incorporate information about nodes associated with other dimensions.

[0045] 209. Optimize the node embedding representation based on the preset loss function until the core semantic concepts among the data of each dimension achieve consistent representation in the embedding space. The loss function is used to maximize the representation similarity of semantically related entity nodes in the embedding space, while minimizing the representation similarity of unrelated entity nodes. Step 209 is the training and optimization process of node embedding representations. Specifically, it involves using a pre-defined loss function to guide the initial node embedding representations generated in step 208 to converge towards the goal of achieving semantic alignment. Optimization learning is an iterative process that continuously improves the quality of node embeddings by adjusting the internal parameters of the GNN model to minimize the value of the loss function. The loss function is designed to guide the model to learn to maximize the representational similarity of semantically related entity nodes in the embedding space while minimizing the representational similarity of unrelated entity nodes. Semantically related entity nodes refer to nodes that represent the same or highly related core semantic concepts across different dimensions (e.g., the node defining "registered capital" in a policy and the node corresponding to the numerical value of "registered capital" in the application materials). Core semantic concepts are important concepts shared across text sources in the political and legal fields. The goal of learning is to achieve consistent representations of these semantically related nodes in the embedding space, meaning their vector representations become similar, reflecting their semantic connections. Optimization learning continues until the loss function converges, reaching a pre-defined stopping condition. At this point, it can be considered that the core semantic concepts across the data in each dimension have achieved effective semantic alignment in the embedding space.

[0046] 210. Extract structured predefined logical rules from the provisions of the original policy text or from the constraints defined in the government affairs ontology; Symbolic reasoning presupposes an executable set of rules, i.e., predefined logical rules. Policy texts contain numerous norms and requirements, while e-government ontology libraries may formally define the constraints between entities. This information serves as the basis for human understanding of policies and compliance judgments, and needs to be transformed into a logical form that machines can understand and apply. In this embodiment, there are two ways to obtain predefined logical rules: first, directly identifying and extracting sentences or fragments with clear structure and judgment logic from the provisions of the original policy text; second, obtaining the constraints defined in the e-government ontology library, which typically contains rules or restrictions expressed in formal language. This process is usually automated, and these predefined logical rules can be maintained and updated when policies are updated or the ontology library is revised.

[0047] 211. On the knowledge graph, symbolic reasoning is used to perform preliminary verification on compliance checkpoints in the application material data that can directly correspond to predefined logical rules, and output preliminary conclusions of certain compliance, partial compliance, non-compliance, or need for further analysis. In actual compliance assessments, symbolic reasoning is first employed on a constructed and semantically aligned knowledge graph. Symbolic reasoning is a computational method based on explicit rules and logic. The system uses pre-extracted, predefined logical rules to verify specific information in the knowledge graph. These verified information points are called compliance checkpoints, which are representations of the application material data in the knowledge graph and directly match and correspond to specific predefined logical rules (for example, if a rule is about "registered capital >= 500,000," the node in the knowledge graph representing the "registered capital" value in the application material is the corresponding checkpoint). During symbolic reasoning, a preliminary verification is performed on each such checkpoint to determine whether it meets the corresponding rule. The verification result is output as a preliminary conclusion, whose state is discrete, including certain compliance (clearly meeting the rule), non-compliance (clearly violating the rule), partial compliance (e.g., meeting some conditions of the rule), or requiring further analysis (the rule cannot be directly judged and more information or more complex reasoning is needed). The advantage of symbolic reasoning lies in the logical rigor and high certainty of its judgment results, enabling it to handle the most direct and certain compliance conditions in government documents.

[0048] 212. For cases where the preliminary conclusion is partially compliant, requires further analysis, or involves semantic complexity in compliance checkpoints, neural reasoning methods are used to perform deep semantic matching by analyzing node embedding representations or graph paths to generate auxiliary evaluation opinions. Step 212 is the second stage of the compliance assessment. Its necessity lies in dealing with compliance issues where symbolic reasoning cannot provide a definitive conclusion or which itself involves complex semantic understanding. These include ambiguous situations that cannot be fully covered by rules, situations that require inference by combining information from multiple sources, or situations that involve understanding the deeper meaning of the text.

[0049] This step is triggered when the initial conclusion output in step 211 is partially compliant and requires further analysis, or when the compliance checkpoint itself involves semantic complexity (e.g., determining whether a description meets the "innovation" standard). In these cases, the system utilizes neural inference for deeper analysis. Neural inference relies on previously learned knowledge graph representations, making judgments by analyzing node embeddings or exploring graph paths within the knowledge graph. Node embeddings capture the semantic location of nodes in the embedding space and their relevance to other nodes, while graph paths reflect the structure formed by a series of relationships connecting entities. By calculating and pattern matching these representations and structures, neural inference can perform deep semantic matching (e.g., determining whether a description in the application materials is semantically highly relevant to the "core technology" concept defined in the policy), capture association patterns that are difficult to discover using symbolic rules, or make probabilistic inferences about compliance. The output of neural inference is an auxiliary assessment opinion, providing supplementary judgments, probability assessments, or relevant insights for complex or uncertain checkpoints, but is generally not a final definitive conclusion.

[0050] 213. Integrate the preliminary conclusions output by symbolic reasoning and the auxiliary evaluation opinions generated by neural reasoning to obtain a comprehensive evaluation result, and determine the final compliance judgment conclusion of the application material data based on the comprehensive evaluation result; Symbolic reasoning provides a preliminary judgment, either deterministic or categorical, while neural reasoning offers supplementary opinions, potentially containing probabilistic or relational pattern information, for complex situations. The system needs to integrate the outputs of symbolic and neural reasoning—the preliminary conclusions and supplementary assessment opinions—to obtain a comprehensive evaluation result. For example, for the same checkpoint, if symbolic reasoning indicates "partial compliance," while neural reasoning gives an 80% probability of compliance, the comprehensive evaluation result is based on the combined score or classification of both. Subsequently, the system determines the final compliance judgment conclusion for the submitted material data based on the comprehensive evaluation result. This is a decision mapping process, mapping the comprehensive evaluation result to the final compliance status presented to the user. The final compliance judgment conclusion is typically one of predefined discrete states, such as "compliant," "non-compliant," or "requires manual review." The logic for the judgment can be rule-based (e.g., if the comprehensive evaluation result for any key checkpoint is "non-compliant," then the final conclusion is "non-compliant") or based on a trained classification model; specific implementation is not limited here. This judgment is the final result given by the system's automated review.

[0051] 214. Obtain supporting evidence directly related to the final compliance judgment conclusion from knowledge graphs, symbolic reasoning, and neural reasoning processes; After determining the final compliance assessment, the system needs to perform backtracking and retrieval to obtain supporting evidence directly related to that conclusion. This evidence comes from two core components: 1. Knowledge Graph: The knowledge graph stores all the original information that has been integrated and semantically aligned, including the specific content of policy clauses, data facts in application materials, and relevant case records. As supporting evidence, the system will retrieve and extract key nodes and edges in the knowledge graph that directly participate in the final judgment, such as specific data item nodes being verified in the application materials, related policy clause nodes, or case nodes that assist in the judgment.

[0052] 2. Symbolic and Neural Reasoning Processes: During compliance assessments, the inference engine records its workflow. For symbolic reasoning, the system records which specific predefined logical rules are applied, at which compliance checkpoints in the knowledge graph these rules are validated, and what the validation results (preliminary conclusions) are. For neural reasoning, the system records which node embeddings or graph paths in the knowledge graph are analyzed, which deep semantic matching or pattern mining operations are performed, and what auxiliary assessment opinions are generated. As supporting evidence, the system extracts from these records the application of key rules that directly lead to the final conclusions, important graph path discoveries, or auxiliary assessment opinions with decisive influence.

[0053] By extracting raw facts from the knowledge graph and tracing the logical chain of reasoning, the system can comprehensively obtain supporting evidence that directly supports the final conclusion. This evidence can be used to generate key components of the final audit report, ensuring the traceability and credibility of the audit results.

[0054] 215. Organize and format the final compliance judgment conclusion and supporting evidence to generate the government document review results.

[0055] Taking the final compliance assessment conclusion and the collected supporting evidence as input, and performing organization and formatting operations, the system ultimately generates the government document review result. This result not only includes the final judgment of the automated review, but also all the evidence supporting that judgment, greatly improving the transparency, credibility, and interpretability of the review result.

[0056] The following provides a detailed description of the government document review system based on knowledge graph reasoning provided in this application. Please refer to [link / reference]. Figure 3 , Figure 3 Another embodiment of the government document review system based on knowledge graph reasoning provided in this application, the system includes: The acquisition unit 301 is used to acquire the original policy text and perform semantic parsing processing on the original policy text using a dual-channel semantic disambiguation engine. The dual-channel semantic disambiguation engine includes a rule channel built based on the government domain ontology library and a neural channel built based on a pre-trained language model. The fusion unit 302 is used to fuse the semantic parsing results of the rule channel and the neural channel to obtain a normalized entity set after semantic disambiguation and structured attribute labeling; Mapping unit 303 is used to acquire application material data and case text data related to the original policy text, and to map the standardized entity set, case text data and application material data to the knowledge graph; Alignment unit 304 is used to learn the node embedding representation in the knowledge graph using a graph neural network model in order to achieve semantic alignment between the standardized entity set, case text data and application material data; Reasoning unit 305 is used to conduct a compliance assessment on the degree of conformity between the application material data and the original policy text by using symbolic reasoning based on predefined logical rules and neural reasoning based on node embedding representation or graph path on the knowledge graph. The generation unit 306 is used to generate the government document review results, which include the compliance judgment conclusions and corresponding supporting evidence, based on the compliance assessment results.

[0057] In this embodiment, the functions of each unit are the same as described above. Figure 1 or Figure 2 The steps in the method embodiments shown correspond to those in the examples, and will not be repeated here.

[0058] This application also provides a government document review device based on knowledge graph reasoning; please refer to [link / reference]. Figure 4 , Figure 4 An embodiment of the government document review device based on knowledge graph reasoning provided in this application includes: Processor 401, memory 402, input / output unit 403, bus 404; The processor 401 is connected to the memory 402, the input / output unit 403, and the bus 404; The memory 402 stores a program, and the processor 401 calls the program to execute any of the above-mentioned knowledge graph-based reasoning-based government document review methods.

[0059] This application also relates to a computer-readable storage medium on which a program is stored, which, when run on a computer, causes the computer to execute any of the above-mentioned knowledge graph-based reasoning-based government document review methods.

[0060] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0061] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0062] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0063] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0064] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for reviewing government documents based on knowledge graph reasoning, characterized in that, The methods for reviewing government documents include: The original policy text is obtained and semantic parsing is performed on the original policy text using a dual-channel semantic disambiguation engine. The dual-channel semantic disambiguation engine includes a rule channel built based on a government affairs domain ontology library and a neural channel built based on a pre-trained language model. By fusing the semantic parsing results of the rule channel and the neural channel, a normalized entity set that has undergone semantic disambiguation and structured attribute labeling is obtained; Obtain application material data and original case text data related to the original policy text, and map the standardized entity set, the original case text data, and the application material data into a knowledge graph; A graph neural network model is used to learn the node embedding representations in the knowledge graph in order to achieve semantic alignment between the standardized entity set, the original case data, and the application material data. On the knowledge graph, symbolic reasoning based on predefined logical rules and neural reasoning based on the node embedding representation or graph path are used to conduct a compliance assessment on the degree of conformity between the application material data and the original policy text. Based on the results of the compliance assessment, a government document review result is generated, which includes the compliance judgment conclusion and the corresponding supporting evidence.

2. The method for reviewing government documents according to claim 1, characterized in that, The rule channel is used to perform deterministic semantic parsing on legal terms and normative expressions that conform to a preset pattern in the original policy text using the government affairs ontology library; The neural pathway is used to capture the dynamic semantics of ambiguous qualifiers or non-standard expressions in the original policy text through deep context modeling based on an improved ERNIE model.

3. The method for reviewing government documents according to claim 1, characterized in that, The semantic parsing results of fusing the rule channels and the neural channels yield a normalized entity set that has undergone semantic disambiguation and structured attribute labeling, including: Calculate the parsing confidence of semantic parsing results that are simultaneously identified by the rule channel and the neural channel and point to the same text content; The decision weights of the rule channel and the neural channel are determined according to a preset disambiguation weight allocation strategy; By combining the analytical confidence and the decision weight, the semantic parsing results of the rule channel and the neural channel are selected, merged, or modified to obtain a normalized entity set that has undergone semantic disambiguation and structured attribute labeling.

4. The method according to claim 1, characterized in that, The process of acquiring application material data and original case text data related to the original policy text, and mapping the standardized entity set, the original case text data, and the application material data to a knowledge graph, includes: The entities representing policy clauses and legal concepts in the standardized entity set and their relationships are converted into nodes and edges of the policy dimension in a knowledge graph; Information is extracted from the original case data related to the original policy text, extracting target information such as case focus, judgment reasoning, and cited legal provisions. The target information is then converted into nodes of the case dimension in the knowledge graph, and associations are established with the policy dimension nodes. The content of the application materials data is parsed into structured data items, and the data items are converted into nodes and attributes of the material dimension in the knowledge graph.

5. The method according to claim 4, characterized in that, The step of learning the node embedding representations in the knowledge graph using a graph neural network model to achieve semantic alignment between the standardized entity set, the original case data, and the application material data includes: By utilizing the attention and aggregation mechanisms built into the graph neural network model, the feature information of each entity node from its heterogeneous neighborhood under the policy dimension, the case dimension, and the material dimension is iteratively aggregated to generate a node embedding representation that can capture cross-dimensional contextual dependencies. The heterogeneous neighborhood contains entity nodes of different dimensions and diverse relational edges. The node embedding representation is optimized and learned based on a preset loss function until the core semantic concepts of data in each dimension achieve consistent representation in the embedding space. The loss function is used to maximize the representation similarity of semantically related entity nodes in the embedding space, while minimizing the representation similarity of unrelated entity nodes.

6. The method according to any one of claims 1 to 5, characterized in that, The process involves using symbolic reasoning based on predefined logical rules and neural reasoning based on node embeddings or graph paths on the knowledge graph to conduct a compliance assessment of the degree of conformity between the application material data and the original policy text, including: Extract structured predefined logical rules from the provisions of the original policy text or from the constraints defined in the government affairs ontology library; On the knowledge graph, symbolic reasoning is used to perform preliminary verification on the compliance checkpoints in the application material data that can directly correspond to the predefined logical rules, and output preliminary conclusions of certain compliance, partial compliance, non-compliance, or need for further analysis. For cases where the preliminary conclusion is partially compliant, requires further analysis, or involves semantic complexity in the compliance checkpoints, neural reasoning methods are used to perform deep semantic matching by analyzing the node embedding representation or graph path to generate auxiliary evaluation opinions.

7. The method according to claim 6, characterized in that, The process of generating a government document review result that includes a compliance judgment conclusion and corresponding supporting evidence based on the compliance assessment results includes: The preliminary conclusions output by the symbolic reasoning and the auxiliary evaluation opinions generated by the neural reasoning are integrated to obtain a comprehensive evaluation result, and the final compliance judgment conclusion of the application material data is determined based on the comprehensive evaluation result; From the knowledge graph and the processes of symbolic reasoning and neural reasoning, obtain supporting evidence directly related to the final compliance judgment conclusion; The final compliance judgment and supporting evidence are organized and formatted to generate the government document review result.

8. A government document review system based on knowledge graph reasoning, characterized in that, The government document review system includes: The acquisition unit is used to acquire the original policy text and perform semantic parsing processing on the original policy text using a dual-channel semantic disambiguation engine. The dual-channel semantic disambiguation engine includes a rule channel built based on a government domain ontology library and a neural channel built based on a pre-trained language model. The fusion unit is used to fuse the semantic parsing results of the rule channel and the neural channel to obtain a normalized entity set after semantic disambiguation and structured attribute labeling; The mapping unit is used to acquire application material data and case text data related to the original policy text, and to map the standardized entity set, the case text data and the application material data to the knowledge graph; The alignment unit is used to learn the node embedding representation in the knowledge graph using a graph neural network model, so as to achieve semantic alignment between the normalized entity set, the original case data, and the application material data. The reasoning unit is used to perform a compliance assessment on the knowledge graph by employing symbolic reasoning based on predefined logical rules and neural reasoning based on the node embedding representation or graph path to assess the degree of conformity between the application material data and the original policy text. The generation unit is used to generate a government document review result that includes a compliance judgment conclusion and corresponding supporting evidence based on the results of the compliance assessment.

9. A government document review device based on knowledge graph reasoning, characterized in that, The government document review device includes: Processor, memory, input / output units, and bus; The processor is connected to the memory, the input / output unit, and the bus; The memory stores a program, which the processor invokes to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains a program that, when executed on a computer, performs the method as described in any one of claims 1 to 7.

Images