A Method and System for Extracting and Optimizing Administrative Text Elements Based on Natural Language Processing

Abstract

This invention discloses a method and system for optimizing administrative text element extraction based on natural language processing, belonging to the field of natural language processing technology. The method performs multi-granularity semantic structure parsing on administrative texts to construct multi-granularity semantic structure representation data capable of simultaneously representing word semantics, sentence functional attributes, and textual organization relationships; identifies semantic fragments of candidate administrative elements and constructs an administrative element structure constraint graph to depict the logical dependencies between administrative elements; introduces the structure constraint graph as explicit constraint conditions to perform structure-aware joint reasoning and consistency constraint optimization on administrative elements, suppressing element combinations that do not conform to administrative semantic logic, and outputting structurally consistent administrative text element extraction results; this invention can effectively improve the logical consistency, completeness, and business usability of administrative text element extraction results, and is applicable to application scenarios such as structured processing of administrative documents, government information management, and intelligent auxiliary decision-making.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, specifically to a method and system for extracting and optimizing administrative text elements based on natural language processing. Background Technology

[0002] An optimized method for extracting administrative text elements based on natural language processing is an intelligent processing approach designed to meet the needs of structured administrative text processing. It achieves accurate and grouped identification of key administrative elements by performing multi-granular semantic structure analysis, modeling the logical relationships between administrative elements, and jointly extracting elements under structural constraints. This method not only focuses on the semantic content of the administrative elements themselves but also ensures that the extraction results conform to the requirements of administrative semantic structure by explicitly characterizing the logical dependencies between basis elements, decision elements, implementing body elements, and effective conditions or execution timeliness elements. This provides a reliable data foundation for the structured storage, intelligent retrieval, compliance review, and automation of government business processes in administrative documents.

[0003] However, existing methods for extracting elements from administrative texts often treat words or sentences as independent processing units, neglecting the inherent logical relationships between elements such as basis, decision, execution, and conditions in administrative texts. This can easily lead to fragmented extraction results, isolated elements, or combinations of elements that do not conform to the semantic logic of administrative texts.

[0004] Existing multi-granularity semantic structure parsing methods have technical problems: they focus on general natural language text processing and are difficult to characterize the unique syntactic function differences and discourse organization rules in administrative texts. This results in the separation of semantic information at the word level, sentence level and discourse level, which cannot provide effective semantic support for subsequent administrative element identification.

[0005] Existing methods for constructing administrative element structure constraint diagrams suffer from technical problems. These problems include reliance on manual rules or simple similarity connections for element relationships, making it difficult to reflect the logical closure constraints in administrative semantics and easily generating element relationship networks that do not conform to the legal logic of administration. Summary of the Invention

[0006] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a method and system for optimizing administrative text element extraction based on natural language processing. The technical solution adopted by this invention is as follows: The method for optimizing administrative text element extraction based on natural language processing provided by this invention includes the following steps:

[0007] Step S1: Multi-granular semantic structure parsing of administrative texts;

[0008] Step S2: Constructing the administrative element structure constraint diagram;

[0009] Step S3: Joint extraction of multi-granularity semantic structure constraint elements;

[0010] Step S4: Feature verification output.

[0011] Further, in step S1, the multi-granularity semantic structure parsing of administrative text is used to perform multi-level semantic parsing of administrative text. Specifically, it involves dividing the administrative text into semantic units at the word level, syntactic function level, and discourse structure level, and performing position alignment and semantic encoding processing on the semantic units at different granularities to obtain multi-granularity semantic structure representation data that can simultaneously represent word semantic features, sentence functional attributes, and discourse organization relationships. This includes the following steps:

[0012] Step S11: Basic structured preprocessing of administrative text, used to identify the structural boundaries of the input administrative text. Specifically, it involves text cleaning, paragraph boundary identification and sentence segmentation of the original administrative text, and assigning position indexes to each sentence and paragraph to obtain basic structured text data containing sentence sequences, paragraph sequences and sentence-paragraph hierarchical relationships.

[0013] Step S12: Word-level semantic unit parsing, used to obtain the basic semantic information of the administrative text. Specifically, the administrative text is segmented based on the sentence sequence, and a corresponding word-level semantic representation vector is generated for each word. At the same time, the position index of the word in the sentence is recorded to obtain the word-level semantic unit and its semantic representation data.

[0014] Step S13: Syntactic function-level semantic unit parsing, used to characterize the semantic function attributes of different statements in administrative texts. Specifically, based on the word-level semantic units and the position information of sentences in the text, functional-aware semantic encoding is performed on each sentence to generate syntactic function-level semantic representations of supporting statements, decisive statements, executive statements, or supplementary statements, thereby obtaining syntactic function-level semantic units and their corresponding semantic representation data.

[0015] Step S14: Text structure-level semantic unit parsing, used to model the organizational structure of administrative texts at the text level. Specifically, based on the syntactic function-level semantic representation and paragraph structure information, multiple sentences are subjected to text-level semantic aggregation to generate text structure-level semantic representations that represent different text structure units, thus obtaining text structure-level semantic units and their semantic representation data.

[0016] Step S15: Align and output structured semantic units to establish correspondences between different semantic granularities. Specifically, based on the positional information of words, sentences, and discourse, construct mapping relationships between word-level semantic units, syntactic function-level semantic units, and discourse structure-level semantic units, and integrate the semantic representations of each granularity to obtain multi-granular semantic structure representation data that can simultaneously represent word semantic features, sentence functional attributes, and discourse organization relationships.

[0017] The multi-granularity semantic structure representation data includes at least the semantic representation vector of each semantic unit, the granularity identifier information of the semantic unit, and the correspondence information between semantic units of different granularities. The multi-granularity semantic structure representation data serves as the basic input data for subsequent administrative element structure modeling and joint element extraction.

[0018] Further, in step S2, the construction of the administrative element structure constraint graph is used to characterize the logical dependency relationship between administrative elements. Specifically, based on the multi-granularity semantic structure representation data, candidate administrative element semantic fragments in the administrative text are identified under semantic constraints at the syntactic function level and the discourse structure level, and the administrative element structure constraint graph is constructed using the candidate administrative element semantic fragments as nodes.

[0019] In the administrative element structure constraint diagram, the logical dependency relationship between administrative elements is explicitly represented by edge relationships. The logical dependency relationship includes at least the supporting relationship between the basis element and the decision element, the directional relationship between the decision element and the implementing body element, and the constraint relationship between the decision element and the effective condition or the execution time limit element.

[0020] The administrative element structure constraint diagram is used to limit the reasonable range of element combinations in the subsequent administrative element extraction process, and is provided as structural constraint input data to subsequent steps.

[0021] The semantic fragments of the candidate administrative elements are used to construct the administrative element structure constraint diagram, and are not used as the final administrative element output results.

[0022] The specific steps for constructing the administrative element structure constraint diagram include:

[0023] Step S21: Candidate administrative element semantic fragment identification, used to discover possible administrative element nodes under multi-granularity semantic structure constraints. Specifically, based on the multi-granularity semantic structure representation data obtained in step S1, under the semantic constraints of syntactic function level and discourse structure level, candidate administrative element semantic fragments are identified from the administrative text, and corresponding semantic representation vectors, position indices, and identification information of the syntactic function level and discourse structure level to be assigned to the candidate administrative element semantic fragments, thereby obtaining a set of candidate administrative element semantic fragments.

[0024] Step S22: Weak type attribution modeling of candidate administrative elements, which is used to provide prior constraints for subsequent logical relationship modeling. Specifically, based on the semantic representation vector of the semantic fragments of the candidate administrative elements and their corresponding syntactic function level and discourse structure level semantic information, weak type attribution modeling is performed on the semantic fragments of each candidate administrative element to generate weak attribution scoring information that represents whether it belongs to the basis element, decision element, execution subject element or effective condition, or execution time limit element.

[0025] Step S23: Generate logical dependency relationship edges for administrative elements, which are used to construct logical relationship connections between candidate administrative elements. Specifically, based on the semantic similarity, structural distance features and weak type attribution scoring information between semantic segments of candidate administrative elements, logical dependency relationship edges are generated between semantic segments of candidate administrative elements, and corresponding relationship types and confidence levels are assigned to the logical dependency relationship edges to obtain an initial set of administrative element relationship edges.

[0026] Step S24: Structural consistency constraint relationship edge screening, used to suppress relationship connections that do not conform to administrative semantic logic. Specifically, based on the logical closure constraints between administrative elements, structural consistency gating screening is performed on the initial set of administrative element relationship edges to remove relationship edges that do not participate in the administrative semantic logic structural closure, resulting in a set of administrative element relationship edges that have been structurally constrained and sparsified.

[0027] Step S25: Generate the administrative element structure constraint graph, which is used to form the structural constraint input data for subsequent joint extraction of elements. Specifically, the administrative element structure constraint graph is constructed by using the semantic fragments of the candidate administrative elements as graph nodes and the relationship edges of the administrative elements after being filtered by structural consistency constraints as graph edges.

[0028] The candidate administrative element semantic fragments are used to construct an administrative element structure constraint diagram, but are not used as the final administrative element output. The administrative element structure constraint diagram is provided as structural constraint input data to subsequent steps to limit the reasonable range of element combinations during the administrative element extraction process.

[0029] Further, in step S3, the joint extraction of multi-granularity semantic structure constraint elements is used to complete the extraction of administrative elements under structural constraints. Specifically, the multi-granularity semantic structure representation data and the administrative element structure constraint diagram are jointly modeled. In the process of administrative element extraction, the administrative element structure constraint diagram is introduced as an explicit structural constraint condition, so that the prediction result of administrative elements must satisfy the administrative semantic logical dependency relationship.

[0030] Under the structural constraint mechanism, candidate administrative elements are predicted synchronously, suppressing the output of isolated or conflicting elements that do not meet the structural constraints, and obtaining administrative text element extraction results with consistent structure. The specific steps include:

[0031] Step S31: Construction of multi-source semantic structure joint representation, used to integrate the semantic and structural information of administrative texts. Specifically, based on the multi-granularity semantic structure representation data obtained in step S1 and the administrative element structure constraint graph constructed in step S2, joint representation encoding is performed on the semantic fragments of each candidate administrative element. The semantic representation, syntactic function-level semantic representation and discourse structure-level semantic representation of the candidate administrative elements are spliced ​​or fused to obtain the joint semantic-structure representation data of the candidate administrative elements.

[0032] Step S32: Joint reasoning of structure-aware elements, used to complete the synchronous prediction of candidate administrative elements under the administrative semantic logic constraint. Specifically, the joint semantic-structure representation data is used as node features, and the administrative element structure constraint graph is used as structural constraint conditions. In the process of structure-aware reasoning, the candidate administrative elements are jointly predicted, so that the prediction results of each candidate administrative element are jointly affected by the semantic and structural information of its logically dependent nodes.

[0033] Step S33: Structural consistency constraint optimization, used to suppress element combinations that do not conform to administrative semantic logic. Specifically, structural consistency constraint loss is introduced in the element prediction process to penalize prediction results that violate administrative element existence constraints, element logical dependency order constraints, or element combination closure constraints, thereby guiding the model to output element combination results that satisfy administrative semantic logical dependency relationships.

[0034] Step S34: Administrative element attribution determination, used to generate usable administrative element extraction results. Specifically, based on the prediction results optimized by structural consistency constraints, the final element type attribution determination is made for the candidate administrative elements. When multiple candidate administrative elements compete for the same structural role, the conflict is resolved through structural contribution or consistency constraints, and the administrative elements that satisfy the administrative semantic logic are retained.

[0035] Step S35: Generate administrative text element extraction results, which are used to form the input data for subsequent element verification. Specifically, the administrative text element extraction results with consistent output structure are generated. The administrative text element extraction results include at least the type information of each administrative element, the corresponding semantic content information, and the structural relationship information between administrative elements, and are used as the input data for subsequent element verification steps.

[0036] Further, in step S4, the element verification output is used to evaluate the logical consistency of the extraction results and output the extracted administrative text elements. Specifically, based on the administrative text element extraction results and combined with the administrative element structure constraint diagram, a structural consistency analysis is performed on the combination of administrative elements to verify whether the logical dependency relationship between each administrative element meets the administrative semantic structure constraint requirements.

[0037] After completing the structural consistency verification, the administrative elements that pass the verification are output, and a structural consistency assessment result reflecting the rationality and completeness of the combination of administrative elements is generated. The structural consistency assessment result is used to characterize the credibility of the administrative text element extraction results.

[0038] The administrative text element extraction optimization system based on natural language processing provided by this invention includes a structure parsing module, an element constraint module, a joint extraction module, and a stability evaluation module;

[0039] The structure parsing module is used for multi-granularity semantic structure parsing of administrative texts. Through multi-granularity semantic structure parsing of administrative texts, multi-granularity semantic structure representation data is obtained, and the multi-granularity semantic structure representation data is sent to the element constraint module and the joint extraction module.

[0040] The element constraint module is used to construct an administrative element structure constraint diagram. Through the construction of the administrative element structure constraint diagram, the element structure constraint relationship is obtained, and the element structure constraint relationship is sent to the joint extraction module.

[0041] The joint extraction module is used for joint extraction of multi-granularity semantic structure constraint elements. Through joint extraction of multi-granularity semantic structure constraint elements, the administrative text element extraction result is obtained, and the administrative text element extraction result is sent to the stability assessment module.

[0042] The stability assessment module is used for element verification output, and the structural consistency assessment result is obtained through element verification output.

[0043] The beneficial effects achieved by the present invention using the above solution are as follows:

[0044] (1) In view of the technical problems in existing administrative text element extraction methods, which usually treat words or sentences as independent processing units and ignore the inherent logical dependency relationship between elements such as basis, decision, execution and conditions in administrative texts, the element extraction results are prone to fragmentation, isolated existence of elements or combination of elements that do not conform to administrative semantic logic. This solution creatively adopts a comprehensive administrative text element extraction method that combines multi-granularity semantic structure analysis, construction of administrative element structure constraint diagram and joint extraction of elements. It realizes the introduction of explicit structural constraints in the extraction process, so that the predicted results of administrative elements must meet the administrative semantic logical dependency relationship. Thus, in, for example, in administrative penalty decision documents, elements such as "penalty basis - penalty decision - execution subject - effective period" can be extracted synchronously and in groups, avoiding the problem of extracting only a single penalty clause or isolated basis clause, and improving the overall consistency and business usability of administrative text element extraction results.

[0045] (2) In view of the technical problem that existing multi-granularity semantic structure parsing methods have multiple focuses on general natural language text processing, and are difficult to characterize the unique syntactic function differences and discourse organization rules in administrative texts, resulting in the separation of semantic information at the word level, sentence level and discourse level, and failing to provide effective semantic support for subsequent administrative element identification, this solution creatively adopts a multi-granularity semantic structure collaborative parsing method for administrative texts. By parsing administrative texts at the word level, syntactic function level, and discourse structure level, and establishing alignment relationships between semantic units of different granularities, it realizes accurate modeling of functional statements such as basis statements, decision statements, and execution statements in administrative texts. For example, the same penalty decision can still be uniformly included in the same semantic structure framework in different paragraphs and different expressions, providing a stable and reusable semantic foundation for subsequent structured modeling and element extraction.

[0046] (3) In view of the technical problem that existing methods for constructing administrative element structure constraint diagrams often rely on manual rules or simple similarity connections for element relationships, making it difficult to reflect the logical closure constraints in administrative semantics and easily generating element relationship networks that do not conform to the legal logic of administration, this solution creatively adopts an element logical dependency constraint diagram construction method oriented towards administrative texts. Through candidate administrative element semantic fragment identification, weak type attribution modeling, and structural consistency constraint gating screening, logical closure constraints are introduced while constructing administrative element relationship edges to ensure that the generated structure constraint diagram can truly reflect the administrative semantic logical structure of "based on supporting decisions, decisions pointing to the implementing body, and decisions subject to effective conditions or time limits". This effectively avoids structural abnormal relationships such as "unfounded licensing decisions" or "administrative decisions without implementing bodies" in administrative licensing texts.

[0047] (4) In view of the technical problem that existing joint element extraction methods usually adopt the strategy of independent prediction of each element, which makes it difficult to suppress structural conflicts in the prediction stage, resulting in element type conflicts, role duplication or incomplete element combination in the final output results, this solution creatively adopts the joint extraction method of administrative elements based on structural consistency constraints. In the element prediction process, a structural consistency constraint optimization mechanism is introduced to punish the prediction results that violate the administrative semantic logical dependency relationship. In the final attribution determination stage, the conflict is resolved by combining the structural contribution. This realizes the synchronous and collaborative prediction of administrative elements, so that the combination of administrative elements that meets the logical closure requirements can be stably output in complex administrative texts, which significantly improves the accuracy, stability and credibility of the element extraction results of administrative texts. Attached Figure Description

[0048] Figure 1 A flowchart illustrating the administrative text element extraction optimization method based on natural language processing provided by this invention;

[0049] Figure 2 A schematic diagram of the administrative text element extraction and optimization system based on natural language processing provided by the present invention;

[0050] Figure 3 This is a flowchart illustrating the process of parsing the multi-granular semantic structure of administrative texts in step S1.

[0051] Figure 4 A flowchart illustrating the construction of the administrative element structure constraint diagram for step S2;

[0052] Figure 5 This is a flowchart illustrating the process of jointly extracting multi-granularity semantic structure constraint elements in step S3.

[0053] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0054] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0055] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0056] Example 1, see Figure 1 The present invention provides an optimization method for extracting administrative text elements based on natural language processing. The method includes the following steps:

[0057] Step S1: Multi-granular semantic structure parsing of administrative texts;

[0058] Step S2: Constructing the administrative element structure constraint diagram;

[0059] Step S3: Joint extraction of multi-granularity semantic structure constraint elements;

[0060] Step S4: Feature verification output.

[0061] By performing the above operations, this solution addresses the technical problems inherent in existing administrative text element extraction methods. These methods typically treat words or sentences as independent processing units, neglecting the inherent logical dependencies between elements such as basis, decision, execution, and conditions within the administrative text. This often results in fragmented extraction results, isolated elements, or element combinations that do not conform to administrative semantic logic. This solution creatively adopts a comprehensive administrative text element extraction method that combines multi-granularity semantic structure analysis, administrative element structure constraint diagram construction, and joint element extraction. It introduces explicit structural constraints during the extraction process, ensuring that the predicted results of administrative elements satisfy administrative semantic logical dependencies. This allows for the simultaneous and grouped extraction of elements such as "basis for punishment—punishment decision—executing body—effective period" in documents such as administrative penalty decisions. This avoids the problem of extracting only a single penalty clause or an isolated basis clause, improving the overall consistency and business usability of administrative text element extraction results.

[0062] Example 2, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S1, the multi-granularity semantic structure parsing of administrative text is used to perform multi-level semantic parsing of administrative text. Specifically, it divides the administrative text into semantic units at the word level, syntactic function level, and discourse structure level, and performs position alignment and semantic encoding processing on the semantic units at different granularities to obtain multi-granularity semantic structure representation data that can simultaneously represent word semantic features, sentence functional attributes, and discourse organization relationships. The steps include:

[0063] Step S11: Basic structured preprocessing of administrative text, used to identify the structural boundaries of the input administrative text. Specifically, it involves text cleaning, paragraph boundary identification and sentence segmentation of the original administrative text, and assigning position indexes to each sentence and paragraph to obtain basic structured text data containing sentence sequences, paragraph sequences and sentence-paragraph hierarchical relationships.

[0064] In this embodiment, the text cleaning includes removing abnormal spaces, illegal characters, and repeated punctuation marks, and standardizing common structural markers in administrative texts; paragraph boundary recognition is based on common structural words in administrative documents, such as "after investigation," "based on," "the decision is as follows," and "a penalty is hereby imposed," etc.; sentence segmentation is performed using a combination of punctuation rules and semantic rules, where initial segmentation is first performed based on punctuation marks such as "。; !? etc.", and then sentence boundaries are corrected based on the long sentence structures present in the administrative text.

[0065] After sentence segmentation, a sentence position index is generated for each sentence, and its paragraph number and sequential position within the paragraph are recorded to form a sentence sequence set. At the same time, a paragraph number is generated for each paragraph, and the number of sentences within the paragraph is counted to establish a mapping of the membership relationship between sentences and paragraphs. Through the above processing, basic structured text data is formed, which includes at least sentence sequences, paragraph sequences, and sentence-paragraph membership information, and serves as input data for subsequent word-level semantic parsing and syntactic function recognition.

[0066] Step S12: Word-level semantic unit parsing, used to obtain the basic semantic information of the administrative text. Specifically, the administrative text is segmented based on the sentence sequence, and a corresponding word-level semantic representation vector is generated for each word. At the same time, the position index of the word in the sentence is recorded to obtain the word-level semantic unit and its semantic representation data.

[0067] In this embodiment, the word segmentation process is implemented using a word segmentation algorithm based on a pre-trained language model. For example, WordPiece segmentation or a BERT-CRF-based word segmentation model can be used to segment the sentence to obtain a word sequence. Subsequently, the pre-trained language model is used to semantically encode each word to generate a word-level semantic representation vector. In a specific implementation, the BERT model can be used to perform contextual semantic encoding on the words to obtain the semantic representation vector of each word. At the same time, the position index information of the words in the sentence is recorded to represent the positional relationship of the words in the sentence sequence.

[0068] Through the above processing, a set of word-level semantic units is obtained, wherein each semantic unit includes at least word content, corresponding semantic representation vector and word position index information. This set of word-level semantic units is used to provide basic semantic features for subsequent syntactic function recognition and discourse structure modeling.

[0069] Step S13: Syntactic function-level semantic unit parsing, used to characterize the semantic function attributes of different statements in administrative texts. Specifically, based on the word-level semantic units and the position information of sentences in the text, functional-aware semantic encoding is performed on each sentence to generate syntactic function-level semantic representations of supporting statements, decisive statements, executive statements, or supplementary statements, thereby obtaining syntactic function-level semantic units and their corresponding semantic representation data.

[0070] In this embodiment, sentence-level semantic aggregation is first performed on the semantic vectors of words in the sentence to obtain the overall semantic representation of the sentence. In a specific implementation, the semantic vectors of words in the sentence can be fused by average pooling, max pooling, or attention-weighted aggregation to obtain the sentence semantic representation vector. Subsequently, a syntactic function recognition model is constructed to classify the semantic function of the sentence semantic representation vector. In one implementation, a fully connected classification network or a Transformer-based classification model can be used to identify the semantic function of the sentence, thereby generating a probability distribution of the sentence belonging to different semantic function categories.

[0071] Preferably, the sentence function categories include at least four types: basis statements, decision statements, execution statements, and supplementary explanatory statements. By classifying the semantic representation of sentences, the function category of each sentence is determined, and syntactic function-level semantic representation data is generated. Furthermore, in a preferred embodiment, to improve the stability of administrative text semantic function recognition, administrative semantic prompt vectors can be introduced to enhance the sentence semantic representation. That is, a preset administrative semantic prompt vector is integrated into the sentence semantic representation vector, enabling the model to more accurately identify structured semantic functions such as basis statements, decision statements, and execution statements in administrative texts, thereby improving the accuracy of syntactic function recognition.

[0072] As a further optimization of this embodiment, in order to enable the model to accurately capture the semantic features of specific legal logic in administrative documents, in a specific example of this embodiment, the administrative semantic prompt vector is generated by constructing an administrative domain semantic feature library. The administrative domain semantic feature library can be constructed through statistical analysis of historical administrative document corpora or expert annotation. Specifically, for statements of different functional categories, their legal effect keywords in administrative procedures are extracted as prior knowledge features. Table 1 below shows the specific mapping relationship and enhancement logic between syntactic functional categories and administrative semantic features:

[0073] Table 1. Instantiation Mapping Table of Administrative Syntactic Functional-Level Semantic Features;

[0074]

[0075] In practical processing, the administrative semantic prompt vector and the sentence semantic representation vector can be fused by concatenation or addition to obtain an enhanced syntactic function representation, thereby increasing the distance between different functional statements in the vector space;

[0076] Step S14: Text structure-level semantic unit parsing, used to model the organizational structure of administrative texts at the text level. Specifically, based on the syntactic function-level semantic representation and paragraph structure information, multiple sentences are subjected to text-level semantic aggregation to generate text structure-level semantic representations that represent different text structure units, thus obtaining text structure-level semantic units and their semantic representation data.

[0077] In this embodiment, the semantic representation of a sentence is first aggregated at the paragraph level according to the paragraph relationship to which the sentence belongs, that is, the semantic representations of multiple sentences in the same paragraph are merged to generate a paragraph semantic representation vector. In a specific implementation, the semantic vectors of sentences within a paragraph can be merged by average aggregation or attention aggregation to obtain the paragraph semantic representation.

[0078] Subsequently, the semantic relationships between different paragraphs are modeled using a discourse structure coding model. In one implementation, a Transformer encoder or a graph structure semantic coding model can be used to encode the paragraph sequence in order to capture the semantic dependencies between different paragraphs and the discourse organization structure.

[0079] Through the above processing, text structure-level semantic unit data is generated, wherein each text structure unit includes at least paragraph semantic representation and paragraph structural position information in the document. This text structure-level semantic unit is used to characterize the organizational relationship of administrative texts at the text level and to provide text-level semantic features for subsequent administrative element structural constraint modeling.

[0080] Step S15: Align and output structured semantic units to establish correspondences between different semantic granularities. Specifically, based on the positional information of words, sentences, and discourse, construct mapping relationships between word-level semantic units, syntactic function-level semantic units, and discourse structure-level semantic units, and integrate the semantic representations of each granularity to obtain multi-granular semantic structure representation data that can simultaneously represent word semantic features, sentence functional attributes, and discourse organization relationships.

[0081] In this embodiment, the hierarchical mapping relationship between word-level semantic units and sentences is first established based on the word position index, that is, the sentence to which each word semantic unit belongs is determined; then, the mapping relationship between syntactic function-level semantic units and discourse structure-level semantic units is established based on the hierarchical relationship between sentences and paragraphs, thereby forming a multi-layer semantic mapping structure between word level, sentence level and discourse level.

[0082] After completing the semantic unit mapping, the semantic representations of different granularities are integrated in a unified manner to generate multi-granularity semantic structure representation data. This data includes at least the semantic representation vector corresponding to each semantic unit, the granularity identifier information of the semantic unit, and the mapping relationship information between semantic units of different granularities, thereby forming a multi-granularity semantic structure representation that simultaneously contains word semantic information, sentence function information, and discourse structure information.

[0083] The core of the multi-granularity semantic structure representation data lies in establishing cross-granularity index anchors to ensure that word semantics, sentence function and text logic are fully aligned on the coordinate system, and serve as the basic input data for subsequent administrative element structure constraint diagram construction and joint extraction of administrative elements.

[0084] Specifically, the multi-granularity semantic structure represents the data by constructing a multi-dimensional tensor mapping table to record the hierarchical affiliation of each semantic unit. Table 2 below uses a simplified "Administrative Penalty Decision" as an example to illustrate the specific data structure of the aligned output:

[0085] Table 2. Example table of data alignment instantiation for multi-granularity semantic structure representation;

[0086]

[0087] The multi-granularity semantic structure representation data includes at least the semantic representation vector of each semantic unit, the granularity identifier information of the semantic unit, and the correspondence information between semantic units of different granularities. The multi-granularity semantic structure representation data serves as the basic input data for subsequent administrative element structure modeling and joint element extraction.

[0088] By performing the above operations, this solution addresses the technical problem that existing multi-granularity semantic structure parsing methods often focus on general natural language text processing, making it difficult to characterize the unique syntactic functional differences and discourse organization patterns in administrative texts. This results in the separation of semantic information at the word, sentence, and discourse levels, failing to provide effective semantic support for subsequent administrative element identification. This solution creatively adopts a multi-granularity semantic structure collaborative parsing method for administrative texts. By performing word-level, syntactic function-level, and discourse structure-level semantic unit parsing on administrative texts, and establishing alignment relationships between semantic units of different granularities, it achieves accurate modeling of functional statements such as basis statements, decision statements, and execution statements in administrative texts. For example, the same penalty decision can still be uniformly incorporated into the same semantic structure framework in different paragraphs and with different expressions, providing a stable and reusable semantic foundation for subsequent structured modeling and element extraction.

[0089] Example 3, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S2, the construction of the administrative element structure constraint graph is used to depict the logical dependency relationship between administrative elements. Specifically, based on the multi-granularity semantic structure representation data, candidate administrative element semantic fragments in administrative text are identified under semantic constraints at the syntactic function level and the discourse structure level, and the administrative element structure constraint graph is constructed using the candidate administrative element semantic fragments as nodes.

[0090] In the administrative element structure constraint diagram, the logical dependency relationship between administrative elements is explicitly represented by edge relationships. The logical dependency relationship includes at least the supporting relationship between the basis element and the decision element, the directional relationship between the decision element and the implementing body element, and the constraint relationship between the decision element and the effective condition or the execution time limit element.

[0091] The administrative element structure constraint diagram is used to limit the reasonable range of element combinations in the subsequent administrative element extraction process, and is provided as structural constraint input data to subsequent steps.

[0092] The semantic fragments of the candidate administrative elements are used to construct the administrative element structure constraint diagram, and are not used as the final administrative element output results.

[0093] The specific steps for constructing the administrative element structure constraint diagram include:

[0094] Step S21: Candidate administrative element semantic fragment identification, used to discover possible administrative element nodes under multi-granularity semantic structure constraints. Specifically, based on the multi-granularity semantic structure representation data obtained in step S1, under the semantic constraints of syntactic function level and discourse structure level, candidate administrative element semantic fragments are identified from the administrative text, and corresponding semantic representation vectors, position indices, and identification information of the syntactic function level and discourse structure level to be assigned to the candidate administrative element semantic fragments, thereby obtaining a set of candidate administrative element semantic fragments.

[0095] In this embodiment, the semantic fragments of candidate administrative elements are obtained through the collaborative identification of syntactic function-level semantic information and word-level semantic information. Specifically, firstly, based on the syntactic function-level semantic units obtained in step S1, the sentences are semantically filtered. In one implementation, sentences identified as basis statements, decision statements, or executive statements are preferentially selected as candidate semantic regions. Subsequently, fragment identification is performed within the candidate semantic regions based on word-level semantic vectors, and continuous word fragments that may constitute administrative elements are identified through sliding windows or sequence labeling models.

[0096] In one specific implementation, a sequence labeling model can be used to label sentences with BIO tags to identify the boundaries of administrative semantic segments and obtain a set of candidate administrative element semantic segments. For each candidate administrative element semantic segment, its semantic representation vector, sentence position index, syntactic function category, and paragraph number are recorded to form a set of candidate administrative element semantic segments, which serves as the input for subsequent weak type attribution modeling.

[0097] Step S22: Weak type attribution modeling of candidate administrative elements, which is used to provide prior constraints for subsequent logical relationship modeling. Specifically, based on the semantic representation vector of the semantic fragments of the candidate administrative elements and their corresponding syntactic function level and discourse structure level semantic information, weak type attribution modeling is performed on the semantic fragments of each candidate administrative element to generate weak attribution scoring information that represents whether it belongs to the basis element, decision element, execution subject element or effective condition, or execution time limit element.

[0098] In this embodiment, a preliminary type determination of the semantic fragments of candidate administrative elements is made by constructing an administrative semantic weak type attribution model. Specifically, firstly, a semantic representation vector is calculated for each semantic fragment of candidate administrative elements. The calculation method is to aggregate the semantic vectors of words within the fragment to obtain the fragment semantic vector. Then, the semantic fragments of candidate administrative elements are weakly typed by combining syntactic functional level semantic information and text position features.

[0099] In one implementation, the weak type attribution modeling aims to utilize the multi-dimensional semantic features extracted in the preceding steps to calculate the probability that a candidate fragment belongs to a specific administrative element type. The calculation formula is as follows:

[0100] ;

[0101] In the formula, It is the probability distribution vector of the candidate fragments belonging to each administrative element type. It is the semantic fragment of the i-th candidate administrative element. It is a weight matrix. It is the semantic vector of the semantic fragment of the i-th candidate administrative element. It is a vector concatenation operator. It contains candidate fragments Syntactic functional level semantic units semantic representation, It contains candidate fragments Chapter structure-level semantic units Location features are represented. It is a bias parameter;

[0102] Furthermore, it can be remembered The component on type t is , used to represent candidate fragments The probability of belonging to feature type t, where the feature type set is... ;

[0103] Step S23: Generate logical dependency relationship edges for administrative elements, which are used to construct logical relationship connections between candidate administrative elements. Specifically, based on the semantic similarity, structural distance features and weak type attribution scoring information between semantic segments of candidate administrative elements, logical dependency relationship edges are generated between semantic segments of candidate administrative elements, and corresponding relationship types and confidence levels are assigned to the logical dependency relationship edges to obtain an initial set of administrative element relationship edges.

[0104] In this embodiment, logical relationship edges between candidate elements are generated by constructing an administrative element dependency relationship scoring function. Specifically, for any two candidate administrative element semantic fragments, their semantic similarity, structural distance, and weak type matching degree are first calculated, and the three types of features are fused to calculate the relationship score;

[0105] The formula for calculating semantic similarity is:

[0106] ;

[0107] In the formula, It is semantic similarity. It is the semantic vector of the semantic neighbor segment of the j-th candidate administrative element;

[0108] The formula for calculating the structural distance is:

[0109] ;

[0110] In the formula, It is the structural distance. It is the position index of the semantic fragment of the i-th candidate administrative element. It is the position index of the semantic neighbor segment of the j-th candidate administrative element. In one embodiment, the position index can be represented by a combination index of "paragraph number - sentence number - segment start word number" or an equivalent global offset index.

[0111] The formula for calculating the weakly typed matching degree is:

[0112] ;

[0113] In the formula, It is a weak type matching degree. It represents the probability that a fragment belongs to a specific administrative element type t1. The probability that a neighboring segment belongs to a specific administrative element type t2;

[0114] Among them, t1 and t2 are determined according to the type of logical dependency relationship to be generated: the supporting relationship corresponds to (t1, t2) = (basis, decision), the pointing relationship corresponds to (t1, t2) = (decision, execution subject), and the constraint relationship corresponds to (t1, t2) = (decision, effective condition) or (decision, execution time limit).

[0115] The process of fusing the three types of features to calculate the relationship score is specifically achieved by constructing a relationship score function, the calculation formula of which is as follows:

[0116] ;

[0117] In the formula, It's a relationship rating. This is the similarity weight, with a preferred value of 0.45. This is the structural distance weight, with a preferred value of 0.25. This is the type matching weight, with a preferred value of 0.3;

[0118] Step S24: Structural consistency constraint relationship edge screening, used to suppress relationship connections that do not conform to administrative semantic logic. Specifically, based on the logical closure constraints between administrative elements, structural consistency gating screening is performed on the initial set of administrative element relationship edges to remove relationship edges that do not participate in the administrative semantic logic structural closure, resulting in a set of administrative element relationship edges that have been structurally constrained and sparsified.

[0119] In this embodiment, the initial relation edges are filtered by constructing administrative semantic logic closure rules;

[0120] During the relationship screening process, the decision element nodes in the candidate graph are first identified, and then it is determined whether there is at least one supporting relationship with the basis element node and one pointing relationship with the execution subject node. If the above relationships are not satisfied, the corresponding relationship edge is determined to be a structurally inconsistent relationship and is deleted.

[0121] In a preferred embodiment, relation edges can be filtered using a structural consistency gating function, calculated as follows:

[0122] ;

[0123] And perform gating and filtering on the edges:

[0124] ;

[0125] In the formula, It is a structural consistency gating function. It is an indicator function. The relation edge represents the relation edge weight or relation confidence degree between candidate administrative element nodes. In one implementation, the relation score calculated in step S23 can be used. , These are the relation edges after gating and filtering. This is the in-degree check condition, where k is the index of the node being checked for in-degree. It is the set of candidate node indexes that are determined as the basis for the criteria. This is the out-degree check condition, where m is the out-degree check index. It is the set of candidate node indices that are determined to be elements of the execution entity;

[0126] Step S25: Generate the administrative element structure constraint graph, which is used to form the structural constraint input data for subsequent joint extraction of elements. Specifically, the administrative element structure constraint graph is constructed by using the semantic fragments of the candidate administrative elements as graph nodes and the relationship edges of the administrative elements after being filtered by structural consistency constraints as graph edges.

[0127] In this embodiment, each semantic fragment in the candidate administrative element semantic fragment set is used as a graph node, and the relation edges after structural consistency screening are used as graph edges, thereby constructing an administrative element structural constraint graph; in this structural constraint graph, each node contains at least the semantic fragment semantic representation, position index and weak type attribution probability information, and each relation edge contains at least the relation type and relation confidence information;

[0128] By constructing the above-mentioned administrative element structure constraint diagram, the logical dependency relationship between administrative elements can be explicitly expressed, and structural constraint information can be provided for the subsequent joint extraction steps of administrative elements. This restricts the combination of elements that do not conform to the administrative semantic logic during the element extraction process, thereby improving the overall consistency of the administrative text element extraction results.

[0129] The candidate administrative element semantic fragments are used to construct an administrative element structure constraint diagram, but are not used as the final administrative element output. The administrative element structure constraint diagram is provided as structural constraint input data to subsequent steps to limit the reasonable range of element combinations during the administrative element extraction process.

[0130] By performing the above operations, this solution addresses the technical problem in existing methods for constructing administrative element structure constraint diagrams, which often rely on manual rules or simple similarity connections for element relationships, making it difficult to reflect the logical closure constraints in administrative semantics and easily generating element relationship networks that do not conform to the legal logic of administration. This solution creatively adopts an element logical dependency constraint diagram construction method oriented towards administrative texts. Through candidate administrative element semantic fragment identification, weak type attribution modeling, and structural consistency constraint gating screening, logical closure constraints are introduced while constructing administrative element relationship edges. This ensures that the generated structure constraint diagram can truly reflect the administrative semantic logical structure of "the decision is based on the decision, the decision points to the implementing body, and the decision is subject to the conditions for effectiveness or the time limit." This effectively avoids structurally abnormal relationships such as "unfounded licensing decisions" or "administrative decisions without an implementing body" in, for example, administrative licensing texts.

[0131] Example 4, see Figure 1 , Figure 2 and Figure 5 This embodiment is based on the above embodiment. In step S3, the joint extraction of multi-granularity semantic structure constraint elements is used to complete the extraction of administrative elements under structural constraints. Specifically, the multi-granularity semantic structure representation data and the administrative element structure constraint diagram are jointly modeled. In the process of administrative element extraction, the administrative element structure constraint diagram is introduced as an explicit structural constraint condition so that the prediction result of administrative elements must satisfy the administrative semantic logical dependency relationship.

[0132] Under the structural constraint mechanism, candidate administrative elements are predicted synchronously, suppressing the output of isolated or conflicting elements that do not meet the structural constraints, and obtaining administrative text element extraction results with consistent structure. The specific steps include:

[0133] Step S31: Construction of multi-source semantic structure joint representation, used to integrate the semantic and structural information of administrative texts. Specifically, based on the multi-granularity semantic structure representation data obtained in step S1 and the administrative element structure constraint graph constructed in step S2, joint representation encoding is performed on the semantic fragments of each candidate administrative element. The semantic representation, syntactic function-level semantic representation and discourse structure-level semantic representation of the candidate administrative elements are spliced ​​or fused to obtain the joint semantic-structure representation data of the candidate administrative elements.

[0134] Step S32: Joint reasoning of structure-aware elements, used to complete the synchronous prediction of candidate administrative elements under the administrative semantic logic constraint. Specifically, the joint semantic-structure representation data is used as node features, and the administrative element structure constraint graph is used as structural constraint conditions. In the process of structure-aware reasoning, the candidate administrative elements are jointly predicted, so that the prediction results of each candidate administrative element are jointly affected by the semantic and structural information of its logically dependent nodes.

[0135] In this embodiment, a structure-aware graph reasoning model is constructed to jointly predict candidate administrative elements. Specifically, the semantic fragments of candidate administrative elements are regarded as graph nodes, and the relation edges in the administrative element structural constraint graph are regarded as dependency connections between nodes. The semantic representation of nodes is updated through a graph structure message passing mechanism. The structure-aware graph reasoning model can use a graph convolutional network (GCN) or a graph attention network (GAT) to perform structure-aware reasoning on the administrative element structural constraint graph.

[0136] Step S33: Structural consistency constraint optimization, used to suppress element combinations that do not conform to administrative semantic logic. Specifically, structural consistency constraint loss is introduced in the element prediction process to penalize prediction results that violate administrative element existence constraints, element logical dependency order constraints, or element combination closure constraints, thereby guiding the model to output element combination results that satisfy administrative semantic logical dependency relationships.

[0137] In this embodiment, the model prediction results are optimized by constructing a structural consistency loss function. The structural consistency loss consists of two parts: feature classification loss and structural dependency constraint loss. The calculation method is as follows:

[0138] ;

[0139] In the formula, It is the structural consistency loss function. It is the loss of element classification. It is the structural dependency constraint loss weight. It is the structural dependency constraint loss, calculated using the following formula:

[0140] ;

[0141] In the formula, The first set is the set of candidate node indices that determine the feature set, while the second set is the set of candidate node indices that determine the feature based on. It is the probability that it will be predicted as a determining factor. It is the probability of being predicted based on the factor. It is the relation edge weight from the node corresponding to the semantic neighbor segment j of the candidate administrative element to the node corresponding to the semantic segment i of the candidate administrative element;

[0142] Step S34: Administrative element attribution determination, used to generate usable administrative element extraction results. Specifically, based on the prediction results optimized by structural consistency constraints, the final element type attribution determination is made for the candidate administrative elements. When multiple candidate administrative elements compete for the same structural role, the conflict is resolved through structural contribution or consistency constraints, and the administrative elements that satisfy the administrative semantic logic are retained.

[0143] In this embodiment, when multiple candidate semantic fragments are predicted to be of the same administrative element type, the candidate nodes are ranked by structural contribution score to determine the final retained administrative element nodes. The formula for calculating the structural contribution score is as follows:

[0144] ;

[0145] In the formula, C i It is a structural contribution score. It is the prediction probability weight. It is the probability of individual type prediction. It is the relational scoring weight. It is the set of neighboring segments of the semantic fragments of the candidate administrative elements. It is a semantic integrity weight. It is a semantic integrity score, which can be calculated based on semantic fragment length, semantic continuity, and contextual consistency;

[0146] Step S35: Generate administrative text element extraction results, which are used to form the input data for subsequent element verification. Specifically, the administrative text element extraction results with consistent output structure are generated. The administrative text element extraction results include at least the type information of each administrative element, the corresponding semantic content information, and the structural relationship information between administrative elements, and are used as the input data for subsequent element verification steps.

[0147] By performing the above operations, this solution addresses the technical problem in existing joint element extraction methods, which typically employ an independent prediction strategy for each element, making it difficult to suppress structural conflicts during the prediction stage. This results in element type conflicts, role duplication, or incomplete element combinations in the final output. The solution creatively adopts a joint administrative element extraction method based on structural consistency constraints. This method introduces a structural consistency constraint optimization mechanism during element prediction, penalizing prediction results that violate administrative semantic logical dependencies. Furthermore, it resolves conflicts by incorporating structural contribution in the final attribution determination stage. This achieves synchronous and collaborative prediction of administrative elements, enabling the stable output of administrative element combinations that meet logical closure requirements in complex administrative texts. This significantly improves the accuracy, stability, and reliability of administrative text element extraction results.

[0148] Example 5, see Figure 1 , Figure 2 This embodiment is based on the above embodiment. In step S4, the element verification output is used to evaluate the logical consistency of the extraction results and output the extracted administrative text elements. Specifically, based on the administrative text element extraction results and combined with the administrative element structure constraint diagram, the structure consistency analysis of the administrative element combination is performed to verify whether the logical dependency relationship between each administrative element meets the administrative semantic structure constraint requirements.

[0149] After completing the structural consistency verification, the administrative elements that pass the verification are output, and a structural consistency assessment result reflecting the rationality and completeness of the combination of administrative elements is generated. The structural consistency assessment result is used to characterize the credibility of the administrative text element extraction result.

[0150] In this embodiment, the administrative text element extraction result output in step S3 is first used as the object to be verified. The administrative text element extraction result includes at least the type information of each administrative element, the corresponding original semantic content information, and the structural association information between administrative elements. At the same time, the administrative element structure constraint diagram constructed in step S2 is used as a logical reference to perform consistency analysis on the element combination relationship in the administrative text element extraction result.

[0151] Specifically, by verifying the logical dependencies between the extracted administrative elements one by one, it is determined whether the administrative semantic structure constraints are met. This includes whether there are implementing subject elements corresponding to the decision elements, whether there are basis elements matching the decision elements, and whether a reasonable constraint relationship is formed between the decision elements and the effective conditions or execution time limits. When a combination of administrative elements is detected to violate the above administrative semantic logical dependencies, the corresponding elements are marked as structurally inconsistent elements and removed or downgraded from the final output.

[0152] Furthermore, after completing the structural consistency verification, administrative elements that meet the administrative semantic structure constraints are uniformly output, and based on the number of elements that pass the verification, the coverage of element types, and the degree of closure of element structures, a structural consistency assessment result reflecting the rationality and completeness of the combination of administrative elements is generated. The structural consistency assessment result is output in the form of a structural consistency score or a credibility index, which is used to characterize the overall credibility of the current administrative text element extraction results and provide a reliable basis for subsequent structured storage, retrieval analysis, or government business process applications of administrative texts.

[0153] Example 6, see Figure 1 and Figure 2 Based on the above embodiments, the administrative text element extraction and optimization system based on natural language processing provided by the present invention includes a structure parsing module, an element constraint module, a joint extraction module, and a stability evaluation module.

[0154] The structure parsing module is used for multi-granularity semantic structure parsing of administrative texts. Through multi-granularity semantic structure parsing of administrative texts, multi-granularity semantic structure representation data is obtained, and the multi-granularity semantic structure representation data is sent to the element constraint module and the joint extraction module.

[0155] The element constraint module is used to construct an administrative element structure constraint diagram. Through the construction of the administrative element structure constraint diagram, the element structure constraint relationship is obtained, and the element structure constraint relationship is sent to the joint extraction module.

[0156] The joint extraction module is used for joint extraction of multi-granularity semantic structure constraint elements. Through joint extraction of multi-granularity semantic structure constraint elements, the administrative text element extraction result is obtained, and the administrative text element extraction result is sent to the stability assessment module.

[0157] The stability assessment module is used for element verification output, and the structural consistency assessment result is obtained through element verification output.

[0158] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0159] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0160] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. An optimization method for extracting administrative text elements based on natural language processing, characterized by: The method includes the following steps: Step S1: Multi-granular semantic structure analysis of administrative texts. The administrative texts are divided into semantic units at the word level, syntactic function level, and discourse structure level. The semantic units at different granularities are aligned and semantically encoded to obtain multi-granular semantic structure representation data that can simultaneously represent word semantic features, sentence functional attributes, and discourse organization relationships. Step S2: Construction of administrative element structure constraint graph. Based on the multi-granularity semantic structure representation data, candidate administrative element semantic fragments in administrative texts are identified under semantic constraints at the syntactic function level and discourse structure level. The administrative element structure constraint graph is constructed using the candidate administrative element semantic fragments as nodes. This includes: identification of candidate administrative element semantic fragments, weak type attribution modeling of candidate administrative elements, generation of logical dependency relationship edges of administrative elements, screening of structural consistency constraint relationship edges, and generation of administrative element structure constraint graph. The identification of semantic fragments of candidate administrative elements is based on multi-granularity semantic structure representation data. Under the semantic constraints of syntactic function level and discourse structure level, the identification of semantic fragments of candidate administrative elements is carried out from administrative texts. The corresponding semantic representation vector, position index and identification information of syntactic function level and discourse structure level are assigned to the semantic fragments of candidate administrative elements to obtain a set of semantic fragments of candidate administrative elements. The weak type attribution modeling of the candidate administrative elements is based on the semantic representation vector of the semantic fragments of the candidate administrative elements and their semantic information at the syntactic function level and discourse structure level. The weak type attribution modeling of each candidate administrative element semantic fragment generates weak attribution score information that represents whether it belongs to the basis element, decision element, implementing subject element or effective condition, or implementation time limit element. The generation of logical dependency relationship edges for administrative elements is based on the semantic similarity, structural distance features, and weak type attribution scoring information between semantic segments of candidate administrative elements. Logical dependency relationship edges are generated between semantic segments of candidate administrative elements, and corresponding relationship types and confidence levels are assigned to the logical dependency relationship edges to obtain an initial set of administrative element relationship edges. The structural consistency constraint relationship edge filtering is based on the logical closure constraint between administrative elements. The initial set of administrative element relationship edges is subjected to structural consistency gating filtering to remove relationship edges that do not participate in the administrative semantic logical structure closure, resulting in a set of administrative element relationship edges that have been structurally constrained and sparsified. The administrative element structure constraint graph is generated by using the semantic fragments of the candidate administrative elements as graph nodes and the relationship edges of the administrative elements after being filtered by structural consistency constraints as graph edges. Step S3: Joint extraction of multi-granularity semantic structure constraint elements. The multi-granularity semantic structure representation data and the administrative element structure constraint diagram are jointly modeled. In the process of administrative element extraction, the administrative element structure constraint diagram is introduced as an explicit structural constraint condition, so that the prediction result of administrative elements must satisfy the administrative semantic logical dependency relationship. Under the structural constraints of the administrative element structure constraint diagram, the semantic fragments of the candidate administrative elements are jointly predicted to obtain the administrative text element extraction results with consistent structure. Step S4: Element verification output. Based on the administrative text element extraction results and combined with the administrative element structure constraint diagram, a structural consistency analysis is performed on the combination of administrative elements. After completing the structural consistency verification, the administrative elements that pass the verification are output, and a structural consistency evaluation result reflecting the rationality and completeness of the combination of administrative elements is generated. The structural consistency evaluation result is used to characterize the credibility of the administrative text element extraction results.

2. The method for optimizing administrative text element extraction based on natural language processing according to claim 1, characterized in that: In step S1, the multi-granularity semantic structure parsing of the administrative text includes the following steps: basic structural preprocessing of the administrative text, word-level semantic unit parsing, syntactic function-level semantic unit parsing, discourse structure-level semantic unit parsing, and structured semantic unit alignment output.

3. The method for extracting and optimizing administrative text elements based on natural language processing according to claim 2, characterized in that: In step S1, the multi-granularity semantic structure representation data includes at least the semantic representation vector of each semantic unit, the granularity identifier information of the semantic unit, and the correspondence information between semantic units of different granularities. The multi-granularity semantic structure representation data serves as the basic input data for subsequent administrative element structure modeling and joint element extraction.

4. The method for extracting and optimizing administrative text elements based on natural language processing according to claim 3, characterized in that: In step S3, the joint extraction of multi-granularity semantic structure constraint elements includes the following steps: construction of multi-source semantic structure joint representation, joint reasoning of structure-aware elements, optimization of structural consistency constraints, determination of administrative element attribution, and generation of administrative text element extraction results.

5. The method for optimizing administrative text element extraction based on natural language processing according to claim 4, characterized in that: In step S3, the construction of the multi-source semantic structure joint representation is used to integrate the semantic and structural information of the administrative text. Specifically, based on the multi-granularity semantic structure representation data obtained in step S1 and the administrative element structure constraint graph constructed in step S2, the semantic fragments of each candidate administrative element are jointly represented and encoded. The semantic representation, syntactic function-level semantic representation and discourse structure-level semantic representation of the candidate administrative elements are spliced ​​or fused to obtain the joint semantic-structural representation data of the candidate administrative elements. The structure-aware element joint reasoning is used to complete the synchronous prediction of candidate administrative elements under the administrative semantic logic constraint. Specifically, it uses the joint semantic-structure representation data as node features and the administrative element structure constraint graph as structural constraint conditions to jointly predict candidate administrative elements during the structure-aware reasoning process, so that the prediction results of each candidate administrative element are jointly affected by the semantic and structural information of its logically dependent nodes. The structural consistency constraint optimization is used to suppress element combinations that do not conform to administrative semantic logic. Specifically, it introduces structural consistency constraint loss in the element prediction process to penalize prediction results that violate administrative element existence constraints, element logical dependency order constraints, or element combination closure constraints, thereby guiding the model to output element combination results that satisfy administrative semantic logical dependency relationships.

6. An administrative text element extraction and optimization system based on natural language processing, used to implement the administrative text element extraction and optimization method based on natural language processing as described in any one of claims 1-5, characterized in that: It includes a structural analysis module, an element constraint module, a joint extraction module, and a stability assessment module.

7. The administrative text element extraction and optimization system based on natural language processing according to claim 6, characterized in that: The structure parsing module is used for multi-granularity semantic structure parsing of administrative texts. Through multi-granularity semantic structure parsing of administrative texts, multi-granularity semantic structure representation data is obtained, and the multi-granularity semantic structure representation data is sent to the element constraint module and the joint extraction module. The element constraint module is used to construct an administrative element structure constraint diagram. Through the construction of the administrative element structure constraint diagram, the element structure constraint relationship is obtained, and the element structure constraint relationship is sent to the joint extraction module. The joint extraction module is used for joint extraction of multi-granularity semantic structure constraint elements. Through joint extraction of multi-granularity semantic structure constraint elements, the administrative text element extraction result is obtained, and the administrative text element extraction result is sent to the stability assessment module. The stability assessment module is used for element verification output, and the structural consistency assessment result is obtained through element verification output.

Images