Government affair application platform based on artificial intelligence model

Abstract

The application relates to the technical field of government affair intelligentization, in particular to a government affair application platform based on an artificial intelligence model, which comprises a semantic recognition module, a label attribution module, a rule screening module, a path generation module and a path verification module.In the application, the subject-predicate structure of recognized noun phrases and approval verbs in government affair approval texts is identified, the semantic boundary is defined, the structural recognition accuracy is enhanced, the keyword and field item comparison adopts a field-level consistency mode, the label attribution accuracy is improved, the rule node screening introduces field value comparison logic, the matching result pertinence and effectiveness are improved, in the path generation, a logical clear path structure is established through sorting index and field dependency recognition, field conflict and path confusion are avoided, field value consistency verification ensures that the path content is highly consistent with the approval text, the path is effectively available, automatic processing from text analysis, field attribution to path verification is realized, and the efficiency of the government affair approval process is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent government technology, and in particular to a government application platform based on an artificial intelligence model. Background Technology

[0002] The field of intelligent e-government technology encompasses the extensive application of information technology and automation in government public management, aiming to improve administrative efficiency, standardize business processes, and optimize public services. Its core content includes the digitalization of government information collection, management, and service delivery, covering multiple aspects such as government approvals, public affairs management, and resource allocation. Technical means mainly cover data analysis, decision support, and human-computer interaction, and are widely applied in data-driven administrative management scenarios. Relying on e-government systems and service-oriented architectures, it integrates internal and external information resources of government departments, providing information support for policy formulation and implementation, and represents an important direction for the informatization transformation of modern government management systems.

[0003] Among them, the government application platform based on artificial intelligence models refers to a system platform specifically designed for processing government data and supporting government decision-making, built by integrating artificial intelligence models. It encompasses semantic recognition, structured processing, and intelligent classification of government data; knowledge extraction and mapping for policy affairs; behavior prediction and process optimization modeling based on historical government data; archiving and distribution of government documents using text classification models from machine learning models; intent recognition and response rule matching for public inquiries using natural language understanding models; and the construction of a government affairs knowledge graph through entity recognition and graph reasoning technologies to assist in decision support. The platform typically employs deep learning models for semantic analysis, trains classifiers using labeled data from the government domain, and combines database retrieval for information matching and distribution, ultimately achieving data-driven processing of government business.

[0004] In the current government document approval process, the lack of a semantic parsing mechanism for subject-predicate relationships in the document's syntactic structure often leads to semantic misinterpretations or omissions of keywords when dealing with long or nested sentences, resulting in inaccurate modeling of approval intent. Field attribution processing still primarily relies on keyword-field name similarity matching, lacking a one-to-one structured comparison method. This leads to ambiguity and error rates in label selection, especially in high-concurrency scenarios, frequently resulting in system warnings of field attribution mismatches. During rule node filtering, static field binding is often used, lacking a bidirectional consistency judgment mechanism at the field value level. This causes rule judgment results to deviate from the specific business context, leading to rule selection bias. The path structure generation fails to incorporate field sorting and field dependency models, resulting in frequent path redundancy and field conflicts. The system flowchart exhibits node disorder, impacting the efficiency and reliability of the approval process. The lack of consistency verification procedures between field values ​​and actual input content causes path mismatch, resulting in deviations between approval results and user expectations. For example, in the scenario of regional land use approval, due to the lack of field value verification mechanism, the fields in the approval path often do not match the application materials, leading to approval termination or rejection, which seriously affects approval efficiency and user experience. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing an administrative application platform based on an artificial intelligence model.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A government affairs application platform based on an artificial intelligence model includes:

[0007] The semantic recognition module obtains the approval request statement in the government approval text, extracts the first noun phrase, locates the first approval-related verb after it, confirms whether it constitutes a government entity structure, and generates a government semantic recognition fragment.

[0008] The tag attribution module extracts keywords based on the government semantic recognition fragment, compares the keyword field text with the tag field library, confirms the tag required for the current statement, and generates approval field tag items.

[0009] The rule filtering module, based on the approval field label items, combined with the restrictive field items and field values ​​in the approval rule node structure, compares whether the field names and field values ​​of the option fields and rule fields are consistent, filters out rule nodes that meet the conditions, and generates a set of matching government rule nodes;

[0010] The path generation module extracts the original sorting index value of the nodes based on the matching government affairs rule node set, determines that the node with the smallest sorting index value is the main node of the path, performs field dependency judgment on the node fields, and maps it into a path structure flow to generate an artificial intelligence-driven path structure graph.

[0011] Based on the AI-driven path structure graph, the path verification module performs text consistency comparison of field values, determines and marks available approval paths, and constructs a government intelligent approval path graph.

[0012] As a further embodiment of the present invention, the government affairs semantic recognition fragment includes semantic roles, semantic frameworks, and semantic categories; the approval field label items include label type, field name, and field affiliation; the matching government affairs rule node set includes node number, field constraints, and rule type; the artificial intelligence-driven path structure graph includes main path nodes, field dependencies, and path sorting; and the government affairs intelligent approval path graph includes path verification results, field value hit status, and path graph status.

[0013] As a further aspect of the present invention, the semantic recognition module includes:

[0014] The noun phrase extraction submodule obtains sentence structure information based on the approval request text, extracts the phrase structure that includes the noun core word and modifiers, and performs structural verification on the noun phrase structure based on grammatical position and components to determine whether it has complete main components and modifier-dependent relationships, and generates phrase structure features.

[0015] The approval verb localization submodule locates the first verb in the text that is in the dictionary set after the phrase structure feature value, based on the phrase structure feature and the set of verbs set in the government approval semantic dictionary. It then performs position matching and part-of-speech recognition on the located verbs to generate an approval verb position sequence.

[0016] The subject-predicate relationship judgment submodule determines whether the noun phrase and the approval verb constitute a subject-predicate combination relationship semantically based on the position sequence of the approval verb and the phrase structure features, filters subject-predicate structures that meet the combination requirements, and generates government semantic recognition fragments.

[0017] As a further aspect of the present invention, the tag attribution module includes:

[0018] The keyword extraction submodule extracts the core word structure based on the government semantic recognition fragment, reads the labeled nouns, noun phrases and directional phrases in the structure, identifies the semantic importance and word frequency ranking of the terms, performs filtering operations in combination with semantic categories, and generates a keyword sequence list.

[0019] The field comparison submodule compares the text content of each keyword with the field item according to the keyword sequence list and the standard tag field library built into the government system, calculates the keyword field similarity, filters keyword field combinations with similarity greater than the field matching benchmark value, and establishes a tag field matching sequence.

[0020] The tag confirmation submodule determines whether the keywords and field items form a completely consistent attribution relationship in semantics based on the matching value and structural position of each field item in the tag field matching sequence. It performs field item position verification operation in conjunction with the original approval statement to confirm whether there is a phrase combination in the text that is completely consistent with the field item. It retains field items whose comparison value is equal to the comparison relationship strength threshold, obtains all attribution field items and integrates them into an attribution list, and generates approval field tag items.

[0021] As a further aspect of the present invention, the rule filtering module includes:

[0022] The field option extraction submodule obtains the form structure corresponding to the approval field label item, extracts the key-value pair set consisting of all field names and optional field values ​​in the structure, standardizes the key names in combination with the original records of the field table, establishes an effective mapping relationship between field names and values, and generates a field option mapping set.

[0023] The rule field comparison submodule obtains the field restrictions of all nodes in the approval rule structure based on the field option mapping set, splits the field items and values ​​in each rule node to form a set of field restriction items, calculates the field consistency evaluation value, judges whether the field items are completely consistent with the rule field items according to the structural position, and filters the combination of all consistent node field items to obtain the field comparison matching set.

[0024] The rule node filtering submodule determines whether each set of matching items constitutes a complete constraint combination in the rule node structure based on the field comparison matching set. For items with completely identical field values, it retains the node ID and field path, sets the field name and value matching as the rule filtering condition, summarizes all rule node identifiers that meet the matching conditions, obtains and stores the corresponding structural information, and establishes a matching government rule node set.

[0025] As a further aspect of the present invention, the path generation module includes:

[0026] The sequential master node extraction submodule extracts the original sorting index value of each node in the rule structure based on all node numbers in the matching government affairs rule node set, generates a node sequence table in ascending order, identifies the node with the smallest sorting value as the starting node of the path, and records the unique number, path index number and structural position information corresponding to the node, extracts all the restricted field items configured in the node, and generates the path master node field set.

[0027] The field dependency identification submodule analyzes whether there are dependency features such as inclusion, hierarchy, and affiliation between field names and field values ​​based on all field items in the path main node field set, determines whether field items form a parent-child hierarchical structure, filters out sub-fields with dependency relationships, retains field items with independent attributes, and establishes a field redundancy removal structure table.

[0028] The path structure mapping submodule establishes a mapping relationship between field items and node structures based on the field items retained in the field deduplication structure table. It generates an independent path structure unit for each field item, and the field item serves as a path node in the graph structure. It determines the order of the field items in the path and constructs a directed edge structure according to the order and dependency connection relationship between the fields. It synthesizes a continuous field jump structure path and establishes an artificial intelligence-driven path structure graph.

[0029] As a further aspect of the present invention, the path verification module includes:

[0030] The field extraction and comparison submodule obtains the content of each field node in the AI-driven path structure diagram, extracts the combination structure of field name and field value, extracts a list of field pairs of the same form from the approval field tag items, unifies the field format and performs preprocessing, converts all field names into standard field names, performs a one-to-one comparison operation on the field pairs, determines whether there are completely identical items in the tag field values ​​of the path field values, establishes a matching record table and outputs the comparison identifier, and generates a field matching status sequence.

[0031] The path consistency judgment submodule determines whether all field nodes in the artificial intelligence path structure diagram are successfully matched with the approval field label item based on the field matching status sequence. If all field nodes are successfully matched, the path is marked as a consistent path. If there are mismatched nodes, the missing field item and node number are recorded, a path availability identifier is established, the path structure status is updated, and a path consistency status judgment result is generated.

[0032] Based on the path consistency status judgment results, the path graph construction submodule summarizes and processes all structure graphs marked as consistent paths, extracts the field structure, node number sequence and path structure graph number of each consistent path, and reconstructs the graph structure identifier according to the path number. The logical order of the fields is used as the path primary key, and the connection relationship of the structure graph is used as the edge set information to establish a government intelligent approval path graph.

[0033] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0034] In this invention, the semantic boundaries are clarified and the accuracy of structure recognition is enhanced by identifying the subject-predicate structure of noun phrases and approval verbs in government approval texts. The comparison of keywords and field items adopts a field-level consistency method to improve the accuracy of tag attribution. The rule node filtering introduces field value comparison logic to improve the relevance and effectiveness of matching results. In path generation, a logically clear path structure is established through sorting index and field dependency recognition to avoid field conflicts and path confusion. Field value consistency verification ensures that the path content is highly consistent with the approval text, ensuring the path is effective and usable. All links form a closed-loop linkage process, realizing automated processing from text parsing and field attribution to path verification, thereby improving the efficiency of the government approval process. Attached Figure Description

[0035] Figure 1 This is a platform flowchart of the present invention;

[0036] Figure 2 This is a flowchart of the semantic recognition module of the present invention;

[0037] Figure 3 This is a flowchart of the label attribution module of the present invention;

[0038] Figure 4 This is a flowchart of the rule filtering module of the present invention;

[0039] Figure 5 This is a flowchart of the path generation module of the present invention;

[0040] Figure 6 This is a flowchart of the path verification module of the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0042] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the 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, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0043] Please see Figure 1 Government application platforms based on artificial intelligence models include:

[0044] The semantic recognition module obtains the approval request statements submitted by government personnel in the government approval text, extracts the first noun phrase with a noun core word and modifiers, locates the first approval-type verb in the verb set in the government approval semantic dictionary (such as approval, review, verification, etc.), compares whether the subject-predicate combination relationship between the verb and the noun phrase meets the requirements for constructing the approval behavior, and if it does, it confirms that it constitutes a government entity structure and generates a government semantic recognition fragment;

[0045] The tag attribution module extracts core keywords based on government semantic recognition fragments, and combines them with the tag field library built into the government system (such as standard fields such as unit name, approval item, approval type, etc.) to perform field text comparison operations on the keywords and tag field items to determine whether there are any field items that are completely identical to the keywords. If so, it is confirmed as the tag required for the current statement and an approval field tag item is generated.

[0046] The rule filtering module obtains the government form field option data corresponding to the approval field label item, performs key-value pair mapping of the option field, and combines the restrictive field items and field values ​​in the approval rule node structure (such as the region = XX district, project type = infrastructure) to compare whether the field name and field value of the option field and the rule field are consistent. All rule nodes that meet the condition of completely consistent field values ​​are retained, and a matching government rule node set is generated.

[0047] The path generation module extracts the original sorting index value and establishes a sequence table based on all rule node numbers in the matching government affairs rule node set. It determines that the node with the smallest sorting index value is the path master node, extracts all restrictive field names and field values ​​included in the corresponding node structure, performs field dependency relationship discrimination operation on the fields (such as the dependency structure identification of parent field-child field), filters out field combinations with subordinate redundancy, and maps them into a path structure flow to generate an AI-driven path structure graph.

[0048] The path verification module is based on an AI-driven path structure graph. It compares the field values ​​in the approval field label items with the field values ​​in the path structure graph to determine whether all the field values ​​in the path structure graph match the field values ​​in the label items. If the consistency is found, the path is marked as usable, and a government intelligent approval path graph is constructed.

[0049] The government affairs semantic recognition fragment includes semantic roles, semantic frameworks, and semantic categories; the approval field label items include label type, field name, and field affiliation; the matching government affairs rule node set includes node number, field constraints, and rule type; the AI-driven path structure graph includes main path nodes, field dependencies, and path sorting; and the government affairs intelligent approval path graph includes path verification results, field value hit status, and path graph status.

[0050] Please see Figure 2 The semantic recognition module includes:

[0051] The noun phrase extraction submodule obtains sentence structure information based on the approval request text submitted by government personnel in the government approval text, extracts the phrase structure that includes the noun core word and the modifier, and performs structural verification on the noun phrase structure according to the grammatical position and components to determine whether it has complete main components and modifier dependency relationships, and generates phrase structure features.

[0052] Based on the approval request statements submitted by government officials in government approval documents, it is necessary to identify the sentence structure of each approval statement one by one, and determine whether its first phrase contains a central noun and its modifiers. In specific operations, the text is first processed into sentences, and the first phrase of each sentence is extracted according to the part-of-speech tagging results. If the first phrase is a structure such as "Application for ×× construction project" or "Approval opinion on ×× matter", it is identified as a noun phrase. Then, the modifiers and core nouns in the phrase are broken down. Taking the approval request for "urban road expansion project" as an example, "urban road" can be identified as the core noun, and "expansion project" as a parallel modifier. A modification tree structure is established, and the position of each word is numbered and the dependency relationship is determined. For example, "urban" is an adjective pre-modifier, and "road" is the main modifier. For nouns, "expansion project" is a post-modifying phrase. This structure is considered valid when its dependency weight is above 0.8. The length of each modifier combination is limited to no more than 6 words. If this limit is exceeded, modifiers are selected based on "part-of-speech importance weight," for example, 1.0 for nouns, 0.9 for gerunds, and 0.7 for adjectives. Each phrase combination is scored, and the highest-scoring item is selected as the valid phrase. Combining common approval request phrases, such as "approval request for ×× project" and "×× construction project plan," a valid structure judgment rule is constructed. Among the selected valid phrases, phrase structure feature values ​​are established based on word frequency, word order, and semantic consistency of the modifier. These feature values ​​include structural parameters such as the number of phrase words, modifier length, and principal component position, as shown in the table below.

[0053] Table 1. Structural Feature Parameters of Noun Phrases

[0054]

[0055] As shown in Table 1, the structure of typical phrases is analyzed, and the feature score is used to determine whether the phrase constitutes a complete modification relationship structure. Phrases with a score of more than 0.85 will be included in the subsequent approval verb matching process, and finally the phrase structure features are obtained.

[0056] The approval verb localization submodule locates the first verb in the text that is in the dictionary set after the phrase structure feature value, based on the phrase structure features and the set of verbs set in the government approval semantic dictionary. It then performs position matching and part-of-speech recognition on the located verbs to generate an approval verb position sequence.

[0057] Based on phrase structure features, the verb immediately following an approval statement needs to be located. This requires calling a government approval semantic dictionary, which pre-defines common approval-related verbs such as "approval," "review," "approval," and "authorization," assigning them semantic classification labels and word frequency priorities. The relationship between verb labels and phrase structure feature values ​​is established through position vectors. If the phrase number is 3, the principal component is in the 2nd position, and the following verb is "approval," it is marked as the first approval verb. Further retrieval is needed to determine if this verb is an approval action verb. If its label classification is V_AP (approval-type verb), the verb is retained and its position index is recorded, while non-approval verbs appearing in the sentence, such as "consider" and "suggest," are excluded. Words such as "explanation" and "description" are identified and removed as non-approval verbs through preset part-of-speech tagging rules. For example, in the sentence structure "It is recommended to approve the implementation of the urban road expansion project", "recommend" is a non-approval verb and should be ignored. Only "approval" is retained as a valid matching term. Then, the position vector value of "approval" is normalized and the relative offset is calculated in combination with the positional relationship of the aforementioned phrase structure. A reasonable offset range is set between 1 and 3. If the interval between the verb and the noun phrase exceeds 3 words, it is considered semantically disconnected and is no longer considered a valid approval verb. This process finally filters out all verbs that meet both position and semantic matching and records their index position, word form and part of speech in the sentence to obtain the approval verb position sequence.

[0058] The subject-predicate relationship judgment submodule determines whether the noun phrase and the approval verb constitute a subject-predicate combination relationship semantically based on the position sequence and phrase structure features of the approval verb, filters the subject-predicate structure that meets the combination requirements, and generates government semantic recognition fragments.

[0059] Based on the positional sequence of the verb "approval" and the structural features of the phrase, the first step is to determine whether there is a subject-predicate relationship between the two in the sentence. This determination relies on a dual reference of positional relationship and modification completeness. The position index of the verb "approval" is compared with the position of the corresponding main component of the phrase, and the presence of a subject-predicate dependency chain in the word class relationship between the two is statistically analyzed. Taking the structure "approval of urban road expansion project" as an example, the main component of the noun phrase "expansion project" is in position 3, and the verb "approval" is in position 4, with a relative position difference of 1. A subject-predicate judgment threshold of 3 is set. If the difference does not exceed the threshold, it is considered a valid subject-predicate structure. Then, the degree of difference between the number of phrase modifiers and the number of verb modifiers is judged. If the difference exceeds the threshold, the structure is considered valid. If the difference exceeds 2, it is considered a mismatched structure. For example, the phrase "urban road expansion project" has 2 modifiers. If the verb "approval" has no modifier, the difference value is 2, which is still within the allowable range, and the match is successful. Conversely, if the difference value is 3, the structure is not confirmed. Further, referring to the frequency statistics of the structure in the actual approval text, a matching degree benchmark value range is established. When the subject-verb difference is in the range of 1-3 and the modifier difference is in the range of 0-2, it is considered a valid combination. The subject-verb structure judgment threshold is set to 2.5. Finally, through the above filtering logic, all semantic pairs that meet the combination requirements are combined into a complete semantic entity. The structure number, combined words and their word order information are uniformly recorded to generate government semantic recognition fragments.

[0060] Please see Figure 3 The tag attribution module includes:

[0061] The keyword extraction submodule is based on government semantic recognition fragments. It extracts the core word composition structure, reads the labeled nouns, noun phrases and directional phrases in the structure, identifies the semantic importance and word frequency ranking of the terms, and performs a filtering operation based on semantic categories. It prioritizes retaining the word phrases that are closely related to the main components in the structure and removes low-frequency auxiliary terms to generate a keyword sequence list.

[0062] Based on the semantic recognition fragments of government affairs, the first step is to extract keywords with semantic core structures from the fragments. This operation should identify whether the core words are core terms of government affairs, such as "project name," "unit name," and "construction content." Then, by calling the part-of-speech tagging records in the annotated corpus, part-of-speech filtering is performed on the words in each fragment, marking those with the part of speech of noun (NN), proper noun (NR), or institutional noun (NT) as keyword candidates. Next, by combining the dependency relationships in the syntactic analysis records, words with subject-predicate or subject-object relationships are grouped into phrase chunks, such as "Transportation Bureau" with "Approval" and "Plan." The phrase "Transportation Bureau Approval Plan" is used to merge terms with a dependency distance of no more than 2 words. After constructing the phrase combination, the phrase weight is cumulatively calculated based on word frequency records. For example, if the word frequency of "Transportation Bureau" in the corpus is 56 and the frequency of "approval" is 39, then the phrase score is 95. The keyword selection benchmark is set at 50 points. If the combined phrase score is higher than this value, it is retained as a keyword. In specific implementation, if a statement is "This unit submitted an approval report for the road repair project," then "road repair project" and "approval report" are extracted as candidate keywords. After scoring according to the above word frequency and dependency relationship, the list of retained keywords is as follows:

[0063] Table 2 Keyword Screening Results

[0064]

[0065] As shown in Table 2, the keyword "this unit" was removed because its weight score was less than 50. The remaining keywords were used to construct a keyword sequence list, which is the keyword sequence list generated by this submodule.

[0066] The field comparison submodule compares the text content of keywords with field items one by one, based on the keyword sequence list and the standard tag field library built into the government system, using the following formula:

[0067] ;

[0068] Calculate keyword field similarity Filter keyword field combinations with similarity greater than the field matching benchmark value, and establish a tag field matching sequence, where... Representing the The semantic vector values ​​of each keyword. The first field in the representative field table The encoded vector value of the item, The semantic strength of the left neighboring words of the representative field item. The semantic strength of the right neighboring word of the representative field item. The number of words to be compared. This represents the number of terms in the adjacent environment.

[0069] Based on the keyword sequence list, each keyword is compared with the fields in the standard label field library of the government system. These fields are mainly divided into standard fields such as unit name, project type, item classification, and business level. Each field has an encoding and text content. For example, the "unit name" field includes "transportation bureau," "construction bureau," and "development and reform commission." The comparison benchmark is set as the cosine value matching between the field item encoding vector and the keyword encoding vector, and a context matching factor is introduced. First, a vector encoding is generated for keywords such as "road repair project," with its dimensions constructed according to word roots and meanings. For example, the keyword encoding is a vector (0.61, 0.45, 0.12). Then, the encoding of the "road repair" field item in the "project name" field table is compared (0.60, 0.44, 0.13), and the cosine similarity score is calculated as follows:

[0070] ;

[0071] A score greater than 0.85 is considered a successful match. Simultaneously, the semantic strength of the adjacent terms of the keywords in the text is considered. For example, the adjacent terms of "road repair project" are "this unit" and "submit," with semantic strengths of 0.82 and 0.76. The adjacent terms of the field term are "municipal" and "implement," with strengths of 0.80 and 0.74. The difference in strength is calculated and normalized, resulting in a matching additionality value of 0.98. Based on the dual comparison of keyword and field term matching values, a matching sequence is synthesized, forming the matching items shown in the table below:

[0072] Table 3: Tag Field Matching Details

[0073]

[0074] As shown in Table 3, the keyword field combinations with a similarity score greater than 0.85 were finally selected and a tag field matching sequence was formed.

[0075] Keyword field similarity measures the degree of matching between keywords in government semantic recognition segments and standard field items in government systems in terms of both semantic content and context. Its core significance lies in measuring the similarity between keyword ontology and field item in semantic vector space, reflecting the consistency of their linguistic expression. At the same time, it combines the semantic intensity changes of adjacent words in the context of the keyword to assess the degree of fit between its meaning expressed in the actual text and the usage scenario of the field item. Thus, it comprehensively judges whether the keyword should be assigned to a certain standard label field. The closer the similarity value is to 1, the higher the consistency between the keyword and the field item in terms of semantics and context, the stronger the reliability of the assignment, and the more suitable it is for subsequent label confirmation operations.

[0076] The formula's operational logic mainly consists of two parts: first, calculating the matching degree between the keyword semantic vector and the field item vector; and second, a normalized evaluation of the keyword contextual contrast. Firstly, Partially by using keyword vectors With field vector The dot product, divided by the sum of the squares of the two products plus the square root of 1, aims to measure the angular similarity between the two products in semantic space. Its structure is approximately similar to the normalized cosine similarity. The constant 1 added to the denominator is mainly used for numerical stability adjustment to avoid the abnormal influence of local minima. The absolute value of this part is used to avoid the influence of the sign direction on the result, thus enhancing the numerical generalization stability. Secondly, Part of the formula is used to evaluate the semantic strength difference between the left and right neighboring words of a keyword and normalize their symmetry in the context in the form of a relative proportion. The structure uses absolute values ​​to reflect the magnitude of semantic deviation and adopts a summation structure to prevent the denominator from being 0, while suppressing the fraction amplification effect caused by the difference being too small. The entire formula is a combination of two parts summing structure. Its purpose is to comprehensively consider the semantic stability of the context in which the keyword is located while ensuring the similarity of semantic vectors, so as to construct a more robust basis for determining the attribution of the tag field.

[0077] The tag confirmation submodule determines whether the keywords and field items constitute a completely consistent attribution relationship in semantics based on the matching value and structural position of each field item in the tag field matching sequence. It performs field item position verification operation in conjunction with the original approval statement to confirm whether there is a phrase combination in the text that is completely consistent with the field item. Only field items with a comparison value equal to the comparison relationship strength threshold are retained. All attribution field items are obtained and integrated into an attribution list to generate approval field tag items.

[0078] Based on the tag field matching sequence, all field items with scores higher than the set benchmark are subjected to reverse verification of the original text to confirm whether there is a literal structure combination in the original text that is completely consistent with the content of the field item. This process requires extracting complete phrases from the original text and comparing them with the field items. For example, in the original text "submit an approval report for road repair project", if the matching field is "approval report", the phrase range position is located from the original text, and the left and right contexts are judged by window words. Let the window be 3 words and the matching interval be -3 to +3 word units. If a completely matching phrase appears in this interval, it is considered a confirmed tag. If the phrase "road repair project" in the original text paragraph is consistent with the field item "road repair project" and is located in the 2nd to 5th word of the sentence, it can be judged as a successful comparison. The comparison relationship strength threshold is set to 1.0. When the comparison value meets the same standard, that is, the original text phrase = field item text, it is judged as a successful confirmation. Finally, all successfully matched field items are integrated into a unified tag result to obtain the approval field tag item.

[0079] Please see Figure 4 The rule-based filtering module includes:

[0080] The field option extraction submodule obtains the form structure corresponding to the approval field label items, extracts the key-value pair set consisting of all field names and optional field values ​​in the structure, such as "Region", "Project Type", "Application Level", etc., and field values ​​such as "X District", "Infrastructure", "Class A", etc., and standardizes the key names by combining them with the original records of the field table, uniformly identifying "Region" as "Region", standardizing "Project Category" as "Project Type", and performing content deduplication, null value removal and format correction operations on each field value to establish an effective mapping relationship between field names and values, and generate a field option mapping set;

[0081] To obtain the form structure corresponding to the approval field tags, it is necessary to first call the standard government form template bound to each approval node in the approval process configuration, read the definition and drop-down options or fixed enumeration values ​​of each field item, parse the form structure line by line, use the field name as the main key, and extract the content of the field's subordinate options as corresponding value items to construct a key-value pair set. In operation, for example, the value of the field "Region" may be "X District", "Y District", "Z District", and the value of the field "Project Type" may be "Infrastructure", "People's Livelihood", "Planning", etc. It is necessary to read each item and exclude field items with empty content or incorrect format, and perform semantic analysis on the field key names. Standardization is implemented, for example, converting "administrative division" to "region" and unifying "project category" to "project type". Character cleaning is used to remove spaces, punctuation, and meaningless appended words from field names and values. Terms with the same but different spellings in field values, such as "infrastructure" and "basic construction," are merged and unified into "infrastructure," forming a complete field-value mapping structure. Duplicate field items are determined using a combination of "field name + value" as a unique identifier. Furthermore, for structures with more than 5 fields, field item count constraints are applied, retaining only those related to the label field for subsequent comparison, as shown in Table 4.

[0082] Table 4: Sample Table of Field Option Mapping

[0083]

[0084] As shown in Table 4, a standard key-value relationship is formed by combining field names and field values, and finally a field option mapping set is generated.

[0085] The rule field comparison submodule retrieves the field restrictions for all nodes in the approval rule structure based on the field option mapping set. It then breaks down the field items and values ​​in each rule node to form a set of field restrictions, using the following formula:

[0086] ;

[0087] Calculated field consistency evaluation value Based on whether field names match, field values ​​are equal, and field types are consistent, each field item is checked item by item according to its structural position to ensure it is completely identical to the rule field items. The combination of all identical node field items is then filtered to obtain the field comparison matching set. Representing the The character length of the field value in the item approval section. Representing the The character length of the field value in the item rule is the difference between the two, reflecting the degree of difference in the field content at the literal level. Representing the The sum of character codes for the field names in the approval process. Representing the The sum of the character codes of the field names in the item rule, and the difference between them, reflects the similarities and differences in the expression of the field names. For the first The field type number of the approval field (e.g., 1 for text, 2 for numeric, 3 for enumeration, etc.). The field type number corresponding to the rule field is multiplied by the absolute value of the two fields, divided by the difference plus 1, to reflect the degree of consistency and matching of the field types. The total number of fields to be compared represents the total number of field pairs involved in the calculation.

[0088] After obtaining the field option mapping set, it is necessary to obtain the field restrictions corresponding to each rule node in the approval rule structure one by one. Each rule in the rule structure record contains field name, restriction value, type code, etc. The field restrictions are standardized and parsed, and the field name and value are split into structure records. The field names are normalized and compared with the field names in the mapping set for character consistency. For example, the field "Project Type" and the rule field "Engineering Category" have the same field name after semantic standardization. Then, the field value, such as "Infrastructure", is compared with the rule field value, such as "Infrastructure", for text content consistency. If the two are similar in text content, ... If the length, character set encoding, and word roots are completely identical, the field values ​​are marked as consistent. At the same time, the field types are extracted. For example, if the type number of the rule field "Project Type" is 3 and the type number of the mapping field "Project Type" is 3, then the field types are consistent. Suppose the length of the field value "X Zone" is 2, the length of the rule field value "X Zone" is also 2, the total character encoding is 178, which is consistent with 178, and the type number is 1, which is the same as the rule item. Based on the above dimensions, a comparison operation is performed item by item, and the differences in field value length, field name characters, and field type number are counted to establish a set of field consistency evaluation values, which are then calculated using a formula.

[0089] There are three fields to compare:

[0090] Group 1 (Field value length): (Field name encoding): (Field type number): ;

[0091] Group 2 (Field value length): (Field name encoding): (Field type number): ;

[0092] Group 3 (Field value length): (Field name encoding): (Field type number): ;

[0093] Substituting the above data into the formula, the following is an explanation of each item:

[0094] Group 1:

[0095] ;

[0096] Group 2:

[0097] ;

[0098] Group 3:

[0099] ;

[0100] Final results summary:

[0101] ;

[0102] The results indicate that the inconsistency between field names and field values ​​in the second group of field values ​​significantly lowers the overall consistency and affects the matching quality. The field consistency evaluation value is -3.5275. With the comparison benchmark value set to 0 (i.e., a non-negative R value indicates that the match is acceptable), the current field combination cannot be included in the field comparison matching set.

[0103] The field consistency assessment value is a composite numerical indicator used to measure the degree of matching between approval fields and rule fields across multiple structural dimensions. This value is calculated by weighting factors such as differences in field value length, differences in field name character encoding, and consistency in field type encoding. It can comprehensively reflect the similarity between two fields in terms of content form, semantic expression, and data structure. The closer the assessment value is to zero or a positive number, the higher the consistency between the two fields in the above dimensions. The more negative the value, the more significant the structural differences in mismatch. This indicator can be used as a direct criterion for screening and confirming whether to include an approval field in the matching rule node range, and has the dual significance of accuracy and operational stability.

[0104] The calculation logic of this formula aims to comprehensively evaluate the degree of consistency and matching between approval fields and rule fields across multiple dimensions, including field values, field names, and field types. Its structure unfolds through item-by-item scoring and the synthesis of a final evaluation. First, This is used to measure the relative difference between the length of approval field values ​​and the length of rule field values. Adding 1 to the denominator avoids division by zero and keeps the difference within a reasonable range; the smaller the value, the more consistent the field value lengths. Secondly, This reflects the difference in the sum of character encoding of field names. The square root operation stretches the difference while reducing the impact of extreme values, ensuring that cases where field names are not completely identical but their encodings are similar receive a neutral score. Finally... This is used to measure the consistency of field types. If the types are completely consistent, the denominator is 1, and the score is the product itself. If the type difference is large, the denominator increases, thus lowering the score for that item. This structure ensures that the type difference has a strong constraint on the overall score through the combination of multiplication and the absolute value of the difference. The entire formula summarizes the above three indicators through the combination of addition and subtraction, and gives a balanced consideration to the three dimensions of length, name and type, and outputs the consistency evaluation value between fields.

[0105] The rule node filtering submodule is based on the field comparison matching set. It determines whether each set of matching items constitutes a complete restriction combination in the rule node structure. For items with completely identical field values, it retains the node ID and field path. It sets the field name and value to match simultaneously as the rule filtering condition, summarizes all rule node identifiers that meet the matching conditions, obtains and stores the corresponding structural information, and establishes a matching government rule node set.

[0106] Based on the field comparison matching set, each rule node's field restrictions are matched one-to-one with the mapped fields. Let node A's restrictions include "Region = X District", "Project Type = Infrastructure", and "Application Level = Level 1". If the field comparison matching set contains all three of these items, node A is deemed to meet the matching rule requirements. This node is identified in the system as node ID_1023, with the field path being ruleTree / root / ID_1023. If only two items match, the result is not counted. The filtering criterion is that the number of field restrictions equals the number of field matches, indicating a valid node. Finally, all rule nodes conforming to this matching logic are filtered, recorded, and summarized. The structure retains the rule number, node path, number of matched field pairs, and specific content of the field pairs. After integrating this structured information, a matching government affairs rule node set is established, as shown below:

[0107] Table 5. Example Table of Matching Rule Node Sets

[0108]

[0109] As shown in Table 5, only node ID_1023 satisfies the condition that the number of field matches equals the number of restrictions. After summarizing the nodes that meet the conditions, a set of matching government affairs rules nodes is established.

[0110] Please see Figure 5 The path generation module includes:

[0111] The sequential master node extraction submodule extracts the original sorting index value of each node in the rule structure based on all node numbers in the matching government affairs rule node set, generates a node sequence table in ascending order, identifies the node with the smallest sorting value as the starting node of the path, and records the unique number, path index number and structural position information of the node. It also extracts all the restricted fields configured in the master node, including field names and corresponding field values, such as "Affiliated Region = District X", "Project Type = Infrastructure", "Investment Amount = 10 million yuan", etc., and generates the path master node field set.

[0112] To extract the sorting index value from the node IDs in the matching government affairs rule node set, the complete information of each matching node in the rule structure table of the system must first be retrieved. The "Node ID" field and its associated "Sorting Index" are read, and these data are combined into an initial mapping table. An ascending order operation is performed on the sorting index to form a priority list. For example, if node ID_0003 has a sorting value of 5, ID_0007 has a sorting value of 1, and ID_0012 has a sorting value of 8, then the node with the smallest sorting value, ID_0007, is determined to be the primary node. Next, the restrictive field records for that node are extracted from the data structure corresponding to ID_0007, including the field names "Region," "Project Type," and "Investment Amount," with corresponding field values ​​of "District X," "Infrastructure," and "10 million yuan." A standardization conversion operation is performed on each field name to unify the field naming rules and avoid identification errors due to naming differences. The standard field items and their corresponding field values ​​are combined into a key-value structure set, as shown below:

[0113] Table 6 Example of Path Master Node Field Set

[0114]

[0115] As shown in Table 6, the field items and field values ​​form a standard structure list, which ultimately yields the path master node field set.

[0116] The field dependency identification submodule analyzes all field items in the path master node field set to determine whether there are dependency features such as inclusion, hierarchy, or affiliation between field names and field values. It determines whether field items form a parent-child hierarchical structure. For example, if "belonging to region = X district" corresponds to the lower-level field "street = Zhongshan street", it is marked as a field dependency relationship. It combines field items in pairs and makes subordinate judgments based on the field's category, hierarchy definition, and field value range. If a field can be derived from or is restricted by another field, it is considered a child field. It records the dependency path location and parent field identifier. It filters out child fields with dependency relationships in the field combination, retains field items with independent attributes, and establishes a field redundancy structure table.

[0117] After retrieving the fields from the main node field set, the relationships between the fields are determined. The process first categorizes the field names, distinguishing between higher-level fields (e.g., administrative level, project level) and lower-level fields (e.g., street, specific category). For example, the field "Region" is classified as higher-level, and the field "Street" as lower-level. Then, the field values ​​are compared. If the administrative unit covered by the field value "District X" includes "Zhongshan Street," then they are determined to be parent and child structures. If two field items belong to "Region = District X" and "Street = Zhongshan Street" respectively, the "Street" field in the system belongs to the "Region" field. If the "street" field is identified as a subfield, it will be removed during the identification process. Then, it will be determined whether there is a project affiliation relationship between "project type = infrastructure" and "sub-project type = municipal engineering". If the list of "sub-project type" values ​​in the system field definition is "municipal engineering", "landscape", and "drainage and sewage", and its parent field is "project type = infrastructure", then "sub-project type" is also determined to be subordinate to "project type". Redundant field items will be removed from the structure, and only "project type" will be retained. After completing this discrimination process in all field items, the retained field items will form a result list, resulting in a field redundancy removal structure table.

[0118] The path structure mapping submodule establishes a mapping relationship between field items and node structures based on the field items retained in the field deduplication structure table. It generates an independent path structure unit for each field item, with the field item serving as a path node in the graph structure. The node key is constructed using the field item name and field value as identifiers to determine the order of arrangement in the path. A directed edge structure is constructed based on the order and dependency connection relationship between fields, generating the vertex set and edge set of the structure path graph. The continuous field jump structure path is synthesized to establish an artificial intelligence-driven path structure graph.

[0119] A path structure graph is constructed based on the field items and node path relationships in the deduplication structure table. First, a unique node identifier is assigned to each field item, and path nodes are named in the form of "field name + field value". Then, these nodes are numbered according to the extraction order, such as node 1 being "region = X district", node 2 being "project type = infrastructure", and node 3 being "investment amount = 10 million yuan". Next, node connection relationships are established according to the logical order of the fields, connecting node 1 to node 2, node 2 to node 3, generating edge structure information. Then, directional connections are drawn in the path graph according to the node identifiers and edge structures. At the same time, the hierarchical structure of the nodes is encoded and mapped into the graph structure for path tracing and identification. In this process, if there are multiple main fields corresponding to multiple branch nodes, a tree-like subgraph structure is constructed with the main field as the center. Finally, all field items are mapped to form a complete path structure graph, as shown in Table 7.

[0120] Table 7 Field Path Node Connection Table

[0121]

[0122] As shown in Table 7, a clear path relationship is formed between the fields, and finally an artificial intelligence-driven path structure diagram is established.

[0123] Please see Figure 6 The path verification module includes:

[0124] The field extraction and comparison submodule obtains the content of each field node in the AI-driven path structure diagram, extracts the combination structure of field name and field value, organizes it into a field list, extracts a list of field pairs of the same form from the approval field label items, unifies the field format and performs preprocessing, converts all field names to system standard field names, performs a one-to-one comparison operation on the field pairs, determines whether there are completely identical items in the label field values ​​of the path field values, establishes a matching record table and outputs the comparison identifier, and generates a field matching status sequence.

[0125] To obtain the field node data in the AI-driven path structure diagram, it is necessary to traverse each node in the structure diagram one by one, extract the complete content pairs composed of the field name and field value in the field item, standardize this information, and store it in the field comparison buffer area in the form of key-value pairs. For example, the first node in the structure diagram represents "所属区域=X区", the second node represents "项目类型=基建", and the third node represents "投资额=1000万元". Construct three key-value items according to the form of "field name + field value" for the above data structure, and then extract the field value records marked by the system from the approval field label items. The field structure is like "所属区域=X区", "项目类型=基建", "申报等级=一级". To eliminate the influence of input differences on the comparison results, it is necessary to uniformly perform standardization operations on the field items of the two data sources, including unified naming of field names (such as normalizing "行政区域" to "所属区域"), unifying the character set of field values (such as unifying full-width and half-width, removing redundant punctuation marks), and cleaning of field content (such as normalizing "￥1000.00万" to "1000万元"). Compare each field item in the two sets based on the criteria that the field names are exactly the same and the field values are exactly the same. If there is a match in field name but a slight difference in field value, record the difference value and classify it as a "pending judgment field item", and mark the remaining exactly the same field items as "matched items", as shown in Table 8:

[0126] Table 8 Field Comparison Result Table

[0127]

[0128] As shown in Table 8, matching and non-matching records are generated through field comparison, and finally a field matching status sequence is established.

[0129] The path consistency judgment sub-module determines whether all field nodes in the path structure diagram match successfully according to the field matching status sequence. If all field nodes match successfully, mark the path as a consistent path. If there are unmatched nodes, record the unhit field items and node numbers, establish a path availability identifier, and synchronously extract the path diagram number, the number of field nodes, and the number of unmatched items, update the path structure status, and generate a path consistency status judgment result;

[0130] After obtaining the field matching status sequence, the consistency between field items and label items in the path structure graph is judged. First, the total number of field nodes in the structure graph needs to be counted. For example, if there are 3 field nodes in the path graph, each field is compared to see if there is a complete match in the label field value. The judgment criteria are that the field name is consistent and the content of the field value is consistent after standardization. If 2 out of 3 items are completely matched and 1 is inconsistent, the matching count is recorded as 2 and the inconsistency count as 1, and the path is judged as an "inconsistent path". If all 3 items are consistent, it is judged as an "available path". For inconsistent items, the field name, original field value and the reason for the matching failure are recorded. All matching cases are summarized to generate a path consistency record table. The path graph number, total number of fields and inconsistency item list are further integrated to establish a path status record structure. At the same time, the threshold benchmark is set as "path field matching rate ≥ 1" for a usable path. The field matching rate can be calculated by dividing the number of consistent fields by the total number of fields. For example, in the above example, the matching rate is 2 / 3 = 0.667, which is less than 1, and the path status is unavailable. The final output result is the path identifier ID and the availability flag, generating the path consistency status judgment result.

[0131] The path graph construction submodule summarizes all structure graphs marked as consistent paths based on the path consistency status judgment results. It extracts the field structure, node number sequence and path structure graph number of each consistent path, and reconstructs the graph structure identifier according to the path number. The logical order of the fields is used as the path primary key, and the connection relationship of the structure graph is used as the edge set information to establish a government intelligent approval path graph.

[0132] Based on the path consistency status judgment results, a summary and reconstruction operation is performed on all structure graphs marked as available paths. The node number order, field combination information, and unique structure graph number in the available paths are extracted to establish a path set record table. For each path, a path primary key is generated according to the node number order in the structure graph. The path item identifier is constructed by field name + field value, for example, "Region = X District" → "Project Type = Infrastructure" → "Investment Amount = 10 million yuan". This sequence is stored as the path field sequence primary key. Then, according to the node connection relationship in the structure graph, a graph structure edge set is generated and the direction is marked, thereby constructing a graph structure object that can be used for structure visualization. All path structure objects are reorganized and archived according to the path ID number, and summarized to form a graph element set. The vertex attributes and edge connection relationships of the graph elements are encoded, and a draft of the government process path graph structure is drawn and a graph index table is generated, as shown in Table 9.

[0133] Table 9. Structure of the Government Approval Pathway

[0134]

[0135] As shown in Table 9, the path number and the structure diagram sequence constitute the graph record, and finally establish the government intelligent approval path graph.

[0136] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A government application platform based on artificial intelligence model, characterized in that, The platform includes: The semantic recognition module obtains the approval request statement in the government approval text, extracts the first noun phrase, locates the first approval-related verb after it, confirms whether it constitutes a government entity structure, and generates a government semantic recognition fragment. The tag attribution module extracts keywords based on the government semantic recognition fragment, compares the keyword field text with the tag field library, confirms the tag required for the current statement, and generates approval field tag items. The rule filtering module, based on the approval field label items, combined with the restrictive field items and field values ​​in the approval rule node structure, compares whether the field names and field values ​​of the option fields and rule fields are consistent, filters out rule nodes that meet the conditions, and generates a set of matching government rule nodes; The path generation module extracts the original sorting index value of the nodes based on the matching government affairs rule node set, determines that the node with the smallest sorting index value is the main node of the path, performs field dependency judgment on the node fields, and maps it into a path structure flow to generate an artificial intelligence-driven path structure graph. The path verification module, based on the AI-driven path structure graph, performs text consistency comparison of field values, determines and marks available approval paths, and constructs a government intelligent approval path graph. The rule filtering module includes: The field option extraction submodule obtains the form structure corresponding to the approval field label item, extracts the key-value pair set consisting of all field names and optional field values ​​in the structure, standardizes the key names in combination with the original records of the field table, establishes an effective mapping relationship between field names and values, and generates a field option mapping set. The rule field comparison submodule obtains the field restrictions of all nodes in the approval rule structure based on the field option mapping set, splits the field items and values ​​in each rule node to form a set of field restriction items, calculates the field consistency evaluation value, judges whether the field items are completely consistent with the rule field items according to the structural position, and filters the combination of all consistent node field items to obtain the field comparison matching set. The rule node filtering submodule determines whether each set of matching items constitutes a complete constraint combination in the rule node structure based on the field comparison and matching set. For items with completely identical field values, the node ID and field path are retained. The field name and value are set to match simultaneously as the rule filtering condition. All rule node identifiers that meet the matching conditions are summarized, and the corresponding structural information is obtained and stored to establish a matching government rule node set. The path generation module includes: The sequential master node extraction submodule extracts the original sorting index value of each node in the rule structure based on all node numbers in the matching government affairs rule node set, generates a node sequence table in ascending order, identifies the node with the smallest sorting value as the starting node of the path, and records the unique number, path index number and structural position information corresponding to the node, extracts all the restricted field items configured in the node, and generates the path master node field set. The field dependency identification submodule analyzes whether there are dependency features such as inclusion, hierarchy, and affiliation between field names and field values ​​based on all field items in the path main node field set, determines whether field items form a parent-child hierarchical structure, filters out sub-fields with dependency relationships, retains field items with independent attributes, and establishes a field redundancy removal structure table. The path structure mapping submodule establishes a mapping relationship between field items and node structures based on the field items retained in the field deduplication structure table. It generates an independent path structure unit for each field item, and the field item serves as a path node in the graph structure. It determines the order of the field items in the path and constructs a directed edge structure according to the order and dependency connection relationship between the fields. It synthesizes a continuous field jump structure path and establishes an artificial intelligence-driven path structure graph. The path verification module includes: The field extraction and comparison submodule obtains the content of each field node in the AI-driven path structure diagram, extracts the combination structure of field name and field value, extracts a list of field pairs of the same form from the approval field tag items, unifies the field format and performs preprocessing, converts all field names into standard field names, performs a one-to-one comparison operation on the field pairs, determines whether there are completely identical items in the tag field values ​​of the path field values, establishes a matching record table and outputs the comparison identifier, and generates a field matching status sequence. The path consistency judgment submodule determines whether all field nodes in the AI-driven path structure diagram match the approval field label item successfully based on the field matching status sequence. If all field nodes match successfully, the path is marked as a consistent path. If there are mismatched nodes, the missing field item and node number are recorded, a path availability identifier is established, the path structure status is updated, and a path consistency status judgment result is generated. Based on the path consistency status judgment results, the path graph construction submodule summarizes and processes all structure graphs marked as consistent paths, extracts the field structure, node number sequence and path structure graph number of each consistent path, and reconstructs the graph structure identifier according to the path number. The logical order of the fields is used as the path primary key, and the connection relationship of the structure graph is used as the edge set information to establish a government intelligent approval path graph.

2. The government application platform based on an artificial intelligence model according to claim 1, characterized in that, The government affairs semantic recognition fragment includes semantic roles, semantic frameworks, and semantic categories; the approval field label items include label type, field name, and field affiliation; the matching government affairs rule node set includes node number, field constraints, and rule type; the artificial intelligence-driven path structure graph includes main path nodes, field dependencies, and path sorting; and the government affairs intelligent approval path graph includes path verification results, field value hit status, and path graph status.

3. The government application platform based on an artificial intelligence model according to claim 1, characterized in that, The semantic recognition module includes: The noun phrase extraction submodule obtains sentence structure information based on the approval request text, extracts the phrase structure that includes the noun core word and modifiers, and performs structural verification on the noun phrase structure based on grammatical position and components to determine whether it has complete main components and modifier-dependent relationships, and generates phrase structure features. The approval verb localization submodule locates the first verb in the text that is in the dictionary set after the phrase structure feature value, based on the phrase structure feature and the set of verbs set in the government approval semantic dictionary. It then performs position matching and part-of-speech recognition on the located verbs to generate an approval verb position sequence. The subject-predicate relationship judgment submodule determines whether the noun phrase and the approval verb constitute a subject-predicate combination relationship semantically based on the position sequence of the approval verb and the phrase structure features, filters subject-predicate structures that meet the combination requirements, and generates government semantic recognition fragments.

4. The government application platform based on an artificial intelligence model according to claim 1, characterized in that, The tag attribution module includes: The keyword extraction submodule extracts the core word structure based on the government semantic recognition fragment, reads the labeled nouns, noun phrases and directional phrases in the structure, identifies the semantic importance and word frequency ranking of the terms, performs filtering operations in combination with semantic categories, and generates a keyword sequence list. The field comparison submodule compares the text content of each keyword with the field item according to the keyword sequence list and the standard tag field library built into the government system, calculates the keyword field similarity, filters keyword field combinations with similarity greater than the field matching benchmark value, and establishes a tag field matching sequence. The tag confirmation submodule determines whether the keywords and field items form a completely consistent attribution relationship in semantics based on the matching value and structural position of each field item in the tag field matching sequence. It performs field item position verification operation in conjunction with the original approval statement to confirm whether there is a phrase combination in the text that is completely consistent with the field item. It retains field items whose comparison value is equal to the comparison relationship strength threshold, obtains all attribution field items and integrates them into an attribution list, and generates approval field tag items.

Images