Layered industry classification method and system based on AI semantic analysis

Abstract

The invention relates to the technical field of hierarchical industry classification, in particular to a hierarchical industry classification method and system based on AI semantic analysis. Comprising the steps of constructing an industrial knowledge base and an intelligent agent; collecting multi-modal data of a to-be-evaluated enterprise, and generating a business association packet of the to-be-evaluated enterprise by the intelligent agent according to the multi-modal data; generating an assessment path set of the to-be-assessed enterprise according to the industrial knowledge base and the business association packet; and constructing a hierarchical industry classification model according to the evaluation path set. The business basic portrait is established by collecting public data of an enterprise, the training mapping set of the to-be-evaluated enterprise is dynamically adjusted, and the semantic analysis model matched with the to-be-evaluated enterprise is quickly established, so that semantic interference of irrelevant industries is effectively shielded, the accuracy of business concept extraction is improved, and the classification accuracy and the adaptability to emerging business forms are improved; and a double-path verification model is constructed based on semantic vector similarity retrieval and business concept map reasoning, so that the business analysis efficiency is improved, and the industrial requirements of multi-standard hierarchical classification are met.

Description

Technical Field

[0001] This application relates to the field of hierarchical industry classification technology, and in particular to a hierarchical industry classification method and system based on AI semantic parsing. Background Technology

[0002] Currently, enterprise industry classification mainly employs two methods: manual interpretation and automated matching based on keyword dictionaries. The manual method relies on expert experience, which has limitations such as high subjectivity, low efficiency, and difficulty in large-scale application. While the automated method improves processing efficiency, its core reliance on a static keyword rule base presents significant shortcomings.

[0003] First, the lack of understanding of the context and deep semantics of business descriptions makes it difficult to effectively handle polysemy, synonyms, and complex business models, thus limiting classification accuracy. Second, the classification granularity is coarse, typically mapping only to broad levels such as industry categories, making it difficult to achieve the refined sub-industry classification crucial for industry analysis and risk management. Third, the system has poor scalability and adaptability, struggling to synchronously and accurately map to both GICS international classification and local industry classification standards. It also cannot flexibly handle diversified enterprises and requires frequent manual maintenance to address the emergence of new business models and concepts. Therefore, existing technologies are insufficient to meet the needs of real-time, accurate, interpretable, and multi-standard hierarchical classification of massive numbers of enterprises for industry analysis. Summary of the Invention

[0004] The purpose of this application is to provide a hierarchical industry classification method and system based on AI semantic parsing to solve the above-mentioned technical problems, aiming to improve the efficiency of business analysis for enterprises and meet the industry needs of multi-standard hierarchical classification.

[0005] In some embodiments of this application, a basic business profile is established by collecting publicly available information about enterprises, the training mapping set of the enterprises to be evaluated is dynamically adjusted, a semantic parsing model adapted to the enterprises to be evaluated is quickly built, semantic interference from irrelevant industries is effectively shielded, the accuracy of business concept extraction is improved, and the classification accuracy and adaptability to emerging business formats are enhanced.

[0006] In some embodiments of this application, a dual-path verification model is constructed based on semantic vector similarity retrieval and business concept graph reasoning. The macro-path grasps the overall tone of the enterprise's business and prevents misjudgments due to ambiguity of individual terms; the micro-path provides specific logical evidence chains to correct potential biases in overall semantic understanding. This improves the robustness of the model and the efficiency of business analysis, meeting the industry's needs for multi-standard hierarchical classification.

[0007] In some embodiments of this application, a hierarchical industry classification method based on AI semantic parsing is provided, including: Build an industry knowledge base and intelligent agents; Collect multimodal data of the companies to be evaluated, and the intelligent agent generates business association packages of the companies to be evaluated based on the multimodal data; Generate an assessment path set for the companies to be assessed based on the industry knowledge base and business association packages; A hierarchical industry classification model is constructed based on the assessment path set.

[0008] In some embodiments of this application, the construction of the intelligent agent includes: Build a pre-trained model; Multiple primary industries are established based on the industry knowledge base; Select the target industries sequentially from all primary industries; Establish a business semantic-tag mapping set for the target industry; The business semantic-label mapping sets of each primary industry are generated sequentially, and an intelligent agent is constructed based on the pre-trained model and all business semantic-label mapping sets.

[0009] In some embodiments of this application, generating the business association package of the enterprise to be evaluated includes: A basic business profile of the company to be evaluated is established based on the preprocessing results of the multimodal data; Based on the business profile, a fit value is generated between the company to be evaluated and each primary industry. The anchor industry for the company to be evaluated is selected based on all matching values; Define the business semantics-tag mapping set anchored to the industry as the target mapping set; Based on the basic business profile, a correction strategy for the target mapping set is set, and a training mapping set is generated based on the correction results; Construct a semantic parsing model based on the pre-trained model and training mapping set; Business-related packages are generated based on semantic parsing models and multimodal data.

[0010] In some embodiments of this application, the correction strategy for setting the target mapping set includes: Generate multiple keyword subsets based on the target mapping set; Generate basic business profiles and first-level association values ​​for each keyword subset; Preset correlation threshold A1; If A1>ai (i=1, 2…n), generate the removal instruction for the i-th key subset; Where n is the number of key subsets in the target mapping set; Generate a training mapping set based on the complete elimination instruction.

[0011] In some embodiments of this application, the generation of the business association package includes: Semantic parsing models filter multimodal data; Set multiple key nodes according to the screening results; Generate semantic sub-packages for each key node; Select target semantic packages in sequence according to all semantic sub-packages; Generate label sub-components for the target semantic package according to the semantic parsing model; The label sub-component package includes label categories, text content, and initial weight values; Generate label sub-components for each semantic sub-package in sequence; Generate a business association package according to all label sub-components.

[0012] In some embodiments of the present application, the generation of the evaluation path set of the enterprise to be evaluated includes: Generate multiple classification nodes according to the industrial knowledge base; Judge whether each classification node is a candidate node in sequence; Construct a candidate node list according to all candidate nodes; Select multiple anchor nodes from all classification nodes according to the business association package; Generate multiple first-level sub-paths according to all anchor nodes; Construct the evaluation path set of the enterprise to be evaluated according to all first-level sub-paths and the candidate node list.

[0013] In some embodiments of the present application, the construction of the candidate node list includes: Select target classification nodes from all classification nodes in sequence; Generate the first-level similarity value b between the target classification node and the business association package; Generate the first-level similarity values between each classification node and the business association package in sequence; Preset a similarity value threshold B1; If B1 < bi (i = 1, 2... m), set the i-th classification node as a candidate node; Where, bi is the first-level similarity value between the i-th classification node and the business association package; m is the number of classification nodes; Construct a candidate node list according to all candidate nodes.

[0014] In some embodiments of the present application, the selection of multiple anchor nodes includes: Aggregate all label files in the business association package; Set multiple first-level business labels according to the aggregation result and generate the comprehensive weight value of each first-level business label; Preset a weight value threshold C1; If C1 < ci, (i = 1, 2... r), set the i-th first-level business label as an anchor business label; Where ci is the comprehensive weight value of the i-th primary business tag, and r is the number of primary tags; Select multiple anchor nodes from all category nodes based on all anchor business tags.

[0015] In some embodiments of this application, a hierarchical industry classification system based on AI semantic parsing is provided, including: The central control unit is used to build an industry knowledge base and intelligent agents; The data acquisition unit is used to collect multimodal data of the company to be evaluated. The central control unit includes: The first processing module is used to build an intelligent agent, which generates a business association package of the enterprise to be evaluated based on multimodal data. The second processing module is used to generate an evaluation path set for the company to be evaluated based on the industry knowledge base and business association package. The third processing module is used to construct a hierarchical industry classification model based on the evaluation path set; The first processing module is also used for: Build a pre-trained model; Multiple primary industries are established based on the industry knowledge base; Select the target industries sequentially from all primary industries; Establish a business semantic-tag mapping set for the target industry; The business semantic-label mapping set of each primary industry is generated sequentially, and an intelligent agent is constructed based on the pre-trained model and all business semantic-label mapping sets. The first processing module is also used for: A basic business profile of the company to be evaluated is established based on the preprocessing results of the multimodal data; Based on the business profile, a fit value is generated between the company to be evaluated and each primary industry. The anchor industry for the company to be evaluated is selected based on all matching values; Define the business semantics-tag mapping set anchored to the industry as the target mapping set; Based on the basic business profile, a correction strategy for the target mapping set is set, and a training mapping set is generated based on the correction results; Construct a semantic parsing model based on the pre-trained model and training mapping set; Business-related packages are generated based on semantic parsing models and multimodal data.

[0016] In some embodiments of this application, the second processing module is further configured to: Multiple classification nodes are generated based on the industry knowledge base; Determine whether each category node is a candidate node in turn; Construct a candidate node list based on all candidate nodes; Select multiple anchor nodes from all category nodes based on the business-related packages; Multiple first-level sub-paths are generated based on all anchor nodes; Construct an evaluation path set for the enterprise to be evaluated based on all first-level sub-paths and candidate node lists.

[0017] Compared with existing technologies, the hierarchical industry classification method and system based on AI semantic parsing proposed in this application have the following advantages: By collecting publicly available information about enterprises to establish basic business profiles, dynamically adjusting the training mapping set of the enterprises to be evaluated, and quickly building a semantic parsing model that is adapted to the enterprises to be evaluated, the semantic interference of irrelevant industries can be effectively shielded, the accuracy of business concept extraction can be improved, and the classification accuracy and adaptability to emerging business formats can be enhanced.

[0018] A dual-path verification model is constructed based on semantic vector similarity retrieval and business concept graph reasoning. The macro-path grasps the overall tone of the enterprise's business and prevents misjudgments caused by ambiguity in individual terms; the micro-path provides specific logical evidence chains to correct potential biases in overall semantic understanding. This improves the model's robustness and business analysis efficiency, meeting the industry's needs for multi-standard hierarchical classification. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a hierarchical industry classification method based on AI semantic parsing in a preferred embodiment of this application. Detailed Implementation

[0020] The specific embodiments of this application will be described in further detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate this application, but are not intended to limit the scope of this application.

[0021] In the description of this application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "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 application 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 application.

[0022] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0023] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0024] like Figure 1 As shown in the preferred embodiment of this application, a hierarchical industry classification method based on AI semantic parsing includes: S101: Building an industry knowledge base and intelligent agents; S102: Collect multimodal data of the enterprise to be evaluated, and generate a business association package of the enterprise to be evaluated based on the multimodal data; S103: Generate an evaluation path set for the company to be evaluated based on the industry knowledge base and business association package; S104: Construct a hierarchical industry classification model based on the evaluation path set.

[0025] Specifically, an industry knowledge base is constructed based on all general industry classification standards (GICS, NAICS, and local standards).

[0026] Specifically, multimodal data includes structured data (such as "business scope" in business registration information, product catalogs, patent classification numbers, and industry classification codes from the China Securities Regulatory Commission) and unstructured data (such as company annual reports and data announcements on official websites), as well as other types of data related to the company's business.

[0027] Specifically, building intelligent agents includes: Build a pre-trained model; Multiple primary industries are established based on the industry knowledge base; Select the target industries sequentially from all primary industries; Establish a business semantic-tag mapping set for the target industry; The business semantic-label mapping sets of each primary industry are generated sequentially, and an intelligent agent is constructed based on the pre-trained model and all business semantic-label mapping sets.

[0028] Specifically, training models that perform robustly on general and commercial corpora are selected as pre-training models, such as pre-training models based on the Transformer architecture.

[0029] Specifically, all industry sectors of various general industry classification standards in the industry knowledge base are selected (such as energy, materials, industrial consumer goods, and finance in the GICS system, and manufacturing, mining, construction, transportation, warehousing and postal services in China's "National Economic Industry Classification"). Multiple first-level industries are generated based on all industry sectors, where each first-level industry represents an industry sector.

[0030] Specifically, a business semantics-business label mapping set for each primary industry is constructed using a large-scale manually labeled "business semantics-business label" dataset. All data in a single business semantics-label mapping set belongs to the same industry sector.

[0031] It is understood that in the above embodiments, multiple primary industries are constructed based on different industry classification standards, and a business semantic-label mapping set (including multiple business semantic-business labels) is established for each primary industry to improve the cross-standard compatibility of the hierarchical industry classification model.

[0032] In a preferred embodiment of this application, generating a business association package for the enterprise to be evaluated includes: A basic business profile of the company to be evaluated is established based on the preprocessing results of the multimodal data; Based on the business profile, a fit value is generated between the company to be evaluated and each primary industry. The anchor industry for the company to be evaluated is selected based on all matching values; Define the business semantics-tag mapping set anchored to the industry as the target mapping set; Based on the basic business profile, a correction strategy for the target mapping set is set, and a training mapping set is generated based on the correction results; Construct a semantic parsing model based on the pre-trained model and training mapping set; Business-related packages are generated based on semantic parsing models and multimodal data.

[0033] Specifically, the preprocessing result refers to using a rule-based lightweight model to extract a list of core business phrases from the structured fields in multimodal data, and then building a basic business profile of the company to be evaluated based on the extraction results.

[0034] Specifically, if the business scope involved in the basic profile of an enterprise overlaps with the business scope of the current primary industry, the fit value between the enterprise to be evaluated and the current primary industry is set to 1; otherwise, the fit value is 0. Specifically, all primary industries with a fit value of 1 with the company to be evaluated are set as anchor industries. Since each primary industry is generated by different industry classification standards, the business scope represented by each primary industry may overlap. Therefore, multiple anchor industries are selected.

[0035] Specifically, the business semantic-label mapping sets of each anchored industry are summed to generate the target mapping set.

[0036] Specifically, the correction strategy for the target mapping set is defined, including: Generate multiple keyword subsets based on the target mapping set; Generate basic business profiles and first-level association values ​​for each keyword subset; Preset correlation threshold A1; If A1>ai (i=1, 2…n), generate the removal instruction for the i-th key subset; Where n is the number of key subsets in the target mapping set; Generate a training mapping set based on the complete elimination instruction.

[0037] Specifically, based on the industry classification standards corresponding to the anchored industry, multiple industry groups are generated in the anchored industry (e.g., the energy equipment and services industry group and the oil, gas and consumer fuel industry group in the energy industry). The business semantic-tag mapping set of the anchored industry is filtered and classified according to all industry groups, thereby generating keyword subsets corresponding to each industry group; wherein, all business semantics-business tags in a single keyword subset belong to the same industry group.

[0038] Specifically, a first-level association value is set based on the degree of overlap between the business scope of the basic business profile and the business scope of the industry group corresponding to the keyword subset. The greater the overlap, the larger the corresponding first-level association value. The mapping relationship between the two can be set based on historical parameters.

[0039] Specifically, the correlation threshold can be set based on historical parameters. If the real-time correlation value is less than the preset correlation threshold, it means that the company being evaluated is completely unrelated to the business scope of the industry group corresponding to the current keyword subset. The current keyword subset should then be removed.

[0040] Understandably, in the above embodiments, by collecting publicly available information about enterprises to establish a basic business profile, dynamically adjusting the training mapping set of the enterprises to be evaluated, and quickly building a semantic parsing model adapted to the enterprises to be evaluated, semantic interference from irrelevant industries is effectively shielded, improving the accuracy of business concept extraction, and enhancing classification accuracy and adaptability to emerging business models. This significantly improves the accuracy of semantic parsing and the ability to distinguish between subdivided businesses, avoiding noise and bias generated when generalizing generalized models across industries.

[0041] In a preferred embodiment of this application, generating a business association package includes: Semantic parsing models filter multimodal data; Set multiple key nodes based on the screening results; Generate semantic sub-packages for each key node; Select the target semantic package sequentially based on all semantic sub-packages; Generate tag components for the target semantic package based on the semantic parsing model; The tag sub-package includes tag categories, text content, and initial weight values; Generate the tag components for each semantic sub-package in sequence; Generate a business association package based on all tagged sub-components.

[0042] Specifically, based on the collected multimodal data from the constructed semantic parsing model, an initial screening is performed to identify content nodes that may contain business description data. These nodes are then designated as key nodes. Based on the keywords identified in each content node, the required context length is determined, thereby generating semantic sub-packages for each key node.

[0043] Specifically, the semantic parsing model parses the text content within the semantic sub-package, outputs the corresponding business tags, and sets the corresponding initial weight value based on the similarity between the text content and the standard business semantics corresponding to the business tag, as well as the data source of the text content. The greater the similarity and the higher the level of the data source of the text content (for example, annual report content is higher than official website content, and the level of each data source can be preset according to the actual status of the enterprise), the greater the corresponding initial weight value.

[0044] A preferred embodiment of this application generates an evaluation path set for the company to be evaluated, including: Multiple classification nodes are generated based on the industry knowledge base; Determine whether each category node is a candidate node in turn; Construct a candidate node list based on all candidate nodes; Select multiple anchor nodes from all category nodes based on the business-related packages; Multiple first-level sub-paths are generated based on all anchor nodes; Construct an evaluation path set for the enterprise to be evaluated based on all first-level sub-paths and candidate node lists.

[0045] Specifically, a classification evaluation value is generated based on a weighted average of the initial confidence value of the target sub-path and the average of the first-level similarity values ​​of each classification node in the target sub-path. The higher the classification evaluation value, the better. Specifically, multiple classification nodes are generated based on the various industry classification standards set in the industry knowledge base (e.g., GICS level nodes (department, industry group, industry, sub-industry), local industry classification level nodes (category, major category, medium category, minor category)), and a standard description text is associated with each classification node. An industry classification knowledge graph is then constructed based on all classification nodes.

[0046] Specifically, the candidate node list provides macro semantic matching, and all first-level sub-paths provide specific concept evidence. Through preset threshold filtering and hierarchical consistency verification (for example, the "sub-industry" matched must belong to the "industry" matched), finally, an optimal classification path from the root to the leaf, as well as the confidence scores at each level, are output for each classification criterion. For enterprises with diversified businesses, the top N paths with confidence higher than the secondary threshold can be retained to construct an evaluation path set.

[0047] Specifically, the matching results are encapsulated into a standardized data structure (such as JSON format), including: enterprise ID, timestamp, GICS path (including codes, names, and confidence at each layer), local standard classification path, core business concept list, etc., so as to construct a hierarchical industrial classification model for the enterprise to be evaluated.

[0048] Specifically, a candidate node list is constructed, including: Target classification nodes are sequentially selected according to all classification nodes; The first-level similarity value b between the target classification node and the business association package is generated; The first-level similarity values between each classification node and the business association package are sequentially generated; A similarity value threshold B1 is preset; If B1 < bi (i = 1, 2... m), the i-th classification node is set as a candidate node; where bi is the first-level similarity value between the i-th classification node and the business association package; m is the number of classification nodes; A candidate node list is constructed according to all candidate nodes.

[0049] Specifically, according to the standard description text corresponding to the target classification node, the business label associated with the target classification node is determined, and all label sub-components in the business association package are traversed according to the selected associated business label. The text content in each compliant business sub-component (that is, the internal label category is the business label associated with the target classification node) is analyzed and compared with the standard description text of the target classification node. The first-level similarity value is set according to the number of compared business sub-components and the average difference. The more the number of compared business sub-components and the smaller the average difference, the greater the corresponding first-level similarity value.

[0050] Specifically, the similarity value threshold can be set according to historical parameters. If the first-level similarity value between the business association package and the current classification node is greater than the preset similarity value threshold, it indicates that the enterprise to be evaluated may be involved in the business scope corresponding to the current classification node.

[0051] Specifically, multiple anchor nodes are selected, including: All label files in the business association package are aggregated; Set multiple first-level business tags according to the aggregation result, and generate the comprehensive weight value of each first-level business tag; Preset the weight value threshold C1; If C1 < ci, (i = 1, 2... r), set the i-th first-level business tag as the anchored business tag; Where, ci is the comprehensive weight value of the i-th first-level business tag, and r is the number of first-level tags; Select multiple anchored nodes from all classification nodes according to all the anchored business tags.

[0052] Specifically, aggregate each tag file with the same tag category to generate multiple first-level business tags, and each first-level business tag is the same.

[0053] Specifically, set the sum of the initial weight values of all tag sub-parts aggregated according to each first-level business tag as the comprehensive weight value.

[0054] Specifically, the weight value threshold can be set according to historical parameters. If the comprehensive weight value of the current first-level business tag is greater than the comprehensive weight value threshold, it means that the current first-level business tag is a key business concept of the enterprise to be evaluated, and set it as the anchored business tag.

[0055] Specifically, set the classification nodes corresponding to each anchored business tag as the anchored nodes and traverse them in the knowledge graph. Directly associate to possible classification nodes through the "belong to" relationship, and expand through the hypernym and hyponym relationships. Count the hit frequency and path weight of each reached classification node.

[0056] Specifically, obtain all the reached classification nodes to generate multiple first-level sub-paths. Among them, the initial confidence value of each first-level sub-path is set according to the average value of the initial scores corresponding to all the classification nodes included in this path. The larger the average value, the larger the corresponding initial confidence value.

[0057] Specifically, each reached classification node obtains an initial score. The score is jointly determined by the confidence of the "business concept" that triggers it and the confidence of the "belong to" relationship through which it passes. The weight obtained by indirect evidence decays according to the path length.

[0058] It can be understood that in the above embodiments, a dual-path verification model is constructed based on semantic vector similarity retrieval and business concept graph reasoning. The macro path grasps the overall tone of the enterprise's business and prevents misjudgment caused by the ambiguity of individual terms; the micro path provides a specific logical evidence chain to correct possible deviations in the overall semantic understanding. Improve the robustness of the model and the efficiency of business analysis, and meet the industrial requirements of multi-standard hierarchical classification.

[0059] In another preferred embodiment of the hierarchical industry classification method based on AI semantic parsing based on any of the above preferred embodiments, this preferred embodiment provides a hierarchical industry classification method based on AI semantic parsing, including: The central control unit is used to build an industry knowledge base and intelligent agents; The data acquisition unit is used to collect multimodal data of the company to be evaluated. The central control unit includes: The first processing module is used to build an intelligent agent, which generates a business association package of the enterprise to be evaluated based on multimodal data. The second processing module is used to generate an evaluation path set for the company to be evaluated based on the industry knowledge base and business association package. The third processing module is used to construct a hierarchical industry classification model based on the evaluation path set; The first processing module is also used for: Build a pre-trained model; Multiple primary industries are established based on the industry knowledge base; Select the target industries sequentially from all primary industries; Establish a business semantic-tag mapping set for the target industry; The business semantic-label mapping set of each primary industry is generated sequentially, and an intelligent agent is constructed based on the pre-trained model and all business semantic-label mapping sets. The first processing module is also used for: A basic business profile of the company to be evaluated is established based on the preprocessing results of the multimodal data; Based on the business profile, a fit value is generated between the company to be evaluated and each primary industry. The anchor industry for the company to be evaluated is selected based on all matching values; Define the business semantics-tag mapping set anchored to the industry as the target mapping set; Based on the basic business profile, a correction strategy for the target mapping set is set, and a training mapping set is generated based on the correction results; Construct a semantic parsing model based on the pre-trained model and training mapping set; Business-related packages are generated based on semantic parsing models and multimodal data.

[0060] In a preferred embodiment of this application, the second processing module is further configured to: Multiple classification nodes are generated based on the industry knowledge base; Determine whether each category node is a candidate node in turn; Construct a candidate node list based on all candidate nodes; Select multiple anchor nodes from all category nodes based on the business-related packages; Multiple first-level sub-paths are generated based on all anchor nodes; Construct an evaluation path set for the enterprise to be evaluated based on all first-level sub-paths and candidate node lists.

[0061] Based on the first concept of this application, a basic business profile is established by collecting publicly available information about enterprises, the training mapping set of the enterprises to be evaluated is dynamically adjusted, and a semantic parsing model adapted to the enterprises to be evaluated is quickly built. This effectively shields semantic interference from irrelevant industries, improves the accuracy of business concept extraction, and enhances classification accuracy and adaptability to emerging business formats.

[0062] Based on the second concept of this application, a dual-path verification model is constructed using semantic vector similarity retrieval and business concept graph reasoning. The macro-path grasps the overall tone of the enterprise's business and prevents misjudgments due to ambiguity of individual terms; the micro-path provides specific logical evidence chains to correct potential biases in overall semantic understanding. This improves the robustness of the model and the efficiency of business analysis, meeting the industry's needs for multi-standard hierarchical classification.

[0063] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and substitutions can be made without departing from the technical principles of this application, and these improvements and substitutions should also be considered within the scope of protection of this application.

Claims

1. A hierarchical industry classification method based on AI semantic analysis, characterized in that, The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent.

2. The AI semantic analysis-based hierarchical industry classification method of claim 1, wherein, The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. 3.The AI semantic parsing-based hierarchical industry classification method of claim 2, wherein, The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. 4.The AI semantic parsing-based hierarchical industry classification method of claim 3, wherein, The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. 5.The AI semantic parsing-based hierarchical industry classification method of claim 4, wherein, The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. 6.The AI semantic parsing-based hierarchical industry classification method of claim 5, wherein, The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. 7.The AI semantic parsing-based hierarchical industry classification method of claim 6, wherein, The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge base and an intelligent agent. The application relates to an industry knowledge According to all candidate nodes, a candidate node list is constructed. 8.The AI semantic parsing-based hierarchical industry classification method of claim 6, wherein, The plurality of anchor nodes are selected, including: All label files in the business correlation package are aggregated. According to the aggregation result, a plurality of first-level business tags are set, and a comprehensive weight value of each first-level business tag is generated. A preset weight value threshold C1 is set. If C1<ci(i=1, 2…r), the ith first-level business tag is set as an anchor business tag. Wherein, ci is the comprehensive weight value of the ith first-level business tag, and r is the number of first-level tags. According to all anchor business tags, a plurality of anchor nodes are selected in all classification nodes.

9. A hierarchical industry classification system based on AI semantic parsing, adopting the hierarchical industry classification system based on AI semantic parsing according to any one of claims 1-8, characterized in that, Including: A central control unit for constructing an industrial knowledge base and an intelligent agent; A data acquisition unit for acquiring multi-modal data of an enterprise to be evaluated; The central control unit includes: A first processing module for constructing an intelligent agent, which generates a business correlation package of the enterprise to be evaluated according to multi-modal data; A second processing module for generating an evaluation path set of the enterprise to be evaluated according to the industrial knowledge base and the business correlation package; A third processing module for constructing a hierarchical industrial classification model according to the evaluation path set; The first processing module is further used for: Establishing a pre-training model; Setting a plurality of first-level industries according to the industrial knowledge base; Selecting a target industry in all first-level industries in turn; Establishing a business semantic-tag mapping set of the target industry; Generating a business semantic-tag mapping set of each first-level industry in turn, and constructing an intelligent agent according to the pre-training model and all business semantic-tag mapping sets; The first processing module is further used for: Establishing a business basic portrait of the enterprise to be evaluated according to the preprocessing result of the multi-modal data; Generating a fit value of the enterprise to be evaluated and each first-level industry according to the business basic portrait; Selecting an anchor industry of the enterprise to be evaluated according to all fit values; Setting the business semantic-tag mapping set of the anchor industry as a target mapping set; Setting a correction strategy of the target mapping set according to the business basic portrait, and generating a training mapping set according to the correction result; Constructing a semantic analysis model according to the pre-training model and the training mapping set; Generating a business correlation package according to the semantic analysis model and the multi-modal data. 10.The AI semantic parsing-based hierarchical industry classification system of claim 9, wherein The second processing module is further used for: Generating a plurality of classification nodes according to the industrial knowledge base; Judging whether each classification node is a candidate node in turn; According to all candidate nodes, a candidate node list is constructed. According to the business correlation package, a plurality of anchor nodes are selected in all classification nodes; According to all anchor nodes, a plurality of first-level sub-paths are generated; According to all first-level sub-paths and the candidate node list, an evaluation path set of the enterprise to be evaluated is constructed.

Images