Domain adaptive data governance standard recommendation system based on multi-modal large model and NLU-HGSC

Abstract

The invention discloses a field adaptive data governance standard recommendation system based on a multi-modal large model and an NLU-HGSC. The system comprises a standard management module, a directory preprocessing module, an NLU-HGSC module and a data storage module. The standard management module is used for collecting and analyzing standards; the directory preprocessing module is used for collecting and preprocessing metadata to be matched; and the NLU-HGSC module is used for calculating the similarity between heterogeneous graph structures generated according to the standard management module and the directory preprocessing module, and recommending a corresponding data standard to the directory data in the data storage module as a basis for data management. According to the NLU-HGSC heterogeneous graph similarity calculation technology, training is not needed, even if a standard is newly added or the standard is modified, the recommendation result can be adjusted in a self-adaptive mode, and the model does not need to be retrained along with changes of standard data.

Description

Technical Field

[0001] The present invention belongs to the field of computer information technology and relates to a domain-adaptive data governance standard recommendation system based on a multimodal large model and NLU-HGSC. Background Art

[0002] In the work of data governance, the determination of data governance standards is crucial, providing an important basis for subsequent data quality and data consistency. With the explosive growth of data scenarios and the diversification of fields, the traditional method of manually formulating data governance standards has gradually exposed some limitations, such as low efficiency, difficulty in adapting to data needs in different fields, delayed standard formulation and updating, and low accuracy of recommended results.

[0003] In actual standard governance operations, we face the following challenges: First, manually formulating data governance standards is inefficient. Second, data from different systems involve many fields, and there may even be multiple fields in one system. This diversity of fields also leads to the diversification of data formats and data types, making the formulation of data governance standards complicated and time-consuming. Traditional methods are difficult to efficiently output relevant standards suitable for the field. Third, data standards in various fields are constantly updated and changing. Manually formulating data governance standards is inefficient and difficult to meet the timeliness of the standards.

[0004] Patent No. CN 118012890 A "A matching method for data fields and data standards and a readable storage medium" describes a method for matching data to be processed and standard data based on a language model. In steps S202 to S205, a method for matching based on data fields and standards is described. Although the bloodline graph relationship of the data to be matched (i.e., the source information of the data) is taken into account and the features of the data to be matched can be supplemented, the more critical data standards and graph structure information contained in the data to be explored itself are ignored.

[0005] In patent number CN 115794899 A "Data standard matching method, device, equipment, medium and product", S110 proposes to obtain multiple features (name, data format and remark information) of a field data item for matching calculation. Compared with the previous patent, more features are introduced for standard matching, and a step of matching corresponding standards through data item synonyms is introduced. However, compared with the heterogeneous graph constructed by multiple entities such as fields, standard names, table data items, field data items, and field data item features proposed in this patent, patent CN 115794899 A only considers the local information in this heterogeneous graph, namely the matching of field data items and field data item features, and does not fully consider the table name and field information, thereby affecting the accuracy of the matching.

[0006] Patent No. CN 114840576 A "Data Standard Matching Method and Device" proposes a method based on field and table name decomposition, and determines the degree of matching with the standard according to the number of associations between the decomposition results and the standard library. Although the matching degree of table name and field name information is considered at the same time, the table name and field name are placed at the same level for matching, the hierarchical relationship is lost, and ultimately the accuracy of the match is affected.

[0007] Patent No. CN 115374108 B "A data standard generation and automatic mapping method based on knowledge graph technology" is a mapping method for data standards. First, a matching method based on knowledge graph is proposed. S201 describes that the graph structure consists of standard number, standard English name, standard name, standard classification, standard type, creator, status and operation. This structure focuses more on the basic attribute information of the standard and omits the multi-level content information of the standard itself (such as fields in the standard, field types, etc.). Secondly, in S301, a rule-based method is described for mapping data standards and metadata feature data sets, and generating training sets and verification sets based on the mapping results to train the mapping model. However, this method based on the mapping model obtained through training has a significant limitation: it lacks the ability to adapt to standard mapping. Whenever a new standard is introduced or the content of an existing standard changes, the model needs to be retrained because the model learns past knowledge and cannot automatically adapt to new or modified content.

[0008] Patent No. CN 117951336 A "Enterprise-level data standard processing method, device, equipment and medium" describes a matching method based on multiple steps and multiple rules, many of which involve manual intervention and lack of self-adaptation. The specific composition of the data items and data standards to be processed in step S202 is not publicly described, and there is no clear description of how to determine the business scope (field) in step S301.

[0009] In summary, the existing technical solutions have the following problems:

[0010] 1. Missing matching information: For example, CN 118012890 A ignores the data standard and the graph structure information contained in the data to be explored; although CN 115794899 A mentions field data items and data item features, it still ignores the graph structure information of the field, standard, and table name; although CN 114840576 A considers table name and field information, it matches the two at the same level and loses the hierarchical relationship; CN 115374108 B only focuses on the basic information of the standard, and does not pay attention to the multi-level content information of the standard itself. In summary, the existing technical solutions have certain deficiencies in the description of the matching object, which ultimately affects the accuracy of the matching.

[0011] 2. Lack of adaptive capability for domain matching: As described in CN 115374108 B, the model obtained through training cannot automatically adapt to newly added standards or domain types, resulting in the need to retrain the model after each change; although CN117951336 A mentions determining the business scope (domain matching), it does not propose a specific technical solution.

[0012] 3. Lack of standard acquisition and parsing technology: A high-accuracy standard recommendation system relies on high-quality standard data, but the existing solutions generally lack a description of the standard data acquisition and parsing process. The description of the standard data during the matching process can reveal the lack of matching information. Incomplete data during the parsing process is the main reason why "missing matching information" ultimately affects the "low standard matching accuracy". For example, CN 117951336 A mentions obtaining public data standards within the scope after determining the business scope, but does not explain how to parse these standard files. Similarly, CN 115794899 A and CN114840576 A also do not mention the acquisition and parsing process of standard data. In CN 115374108 B, although S102 proposes a method based on data frequency and importance analysis, S101 does not clearly explain how to parse metadata. Summary of the invention

[0013] In view of the above-mentioned deficiencies in the prior art, the present invention provides a domain-adaptive data governance standard recommendation system based on a multimodal large model and NLU-HGSC.

[0014] The present invention comprises a standard management module, a catalog preprocessing module, an NLU-HGSC module and a data storage module;

[0015] The standard management module is used for collecting and analyzing standards;

[0016] The directory preprocessing module is used to collect and preprocess the metadata to be matched;

[0017] The NLU-HGSC module is used to calculate the similarity between the heterogeneous graph structures generated by the standard management module and the directory preprocessing module, and recommend corresponding data standards for the directory data in the data storage module as a basis for data governance.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. The present invention adopts multimodal large model technology combined with layout analysis technology to parse standard information from standard files. Compared with the existing technical solutions, the standard data parsed by the present invention has richer node information and hierarchical information, providing data support for subsequent high-precision matching calculations.

[0020] 2. The present invention adopts the few-sample learning technology of multimodal large model technology to achieve adaptive domain classification. Even if new domain classification is added, the model does not need to be retrained, and the purpose of domain classification can be achieved by simply adjusting the prompt.

[0021] 3. The present invention adopts NLU-HGSC, a heterogeneous graph similarity calculation technology based on natural language understanding, to recommend corresponding data standards for all tables and fields in the database to be managed, thereby improving the accuracy of matching. Since NLU-HGSC heterogeneous graph similarity calculation technology does not require training, even if new standards are added or the standards are modified, the recommended results will be adjusted adaptively, and there is no need to retrain the model as the standard data changes, and it has strong standard matching adaptive capabilities.

[0022] In summary, compared with traditional technologies, the present invention can extract richer standard information from standard files of various formats, has higher matching accuracy, and has domain adaptation capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A system structure diagram of an embodiment of the present application;

[0024] Figure 2 This is a diagram of the implementation process of the standard management module of the embodiment of the present application;

[0025] Figure 3 A schematic diagram of the similarity between levels of an embodiment of the present application. DETAILED DESCRIPTION

[0026] The present invention will be further described below in conjunction with the accompanying drawings and examples.

[0027] like Figure 1As shown, the domain-adaptive data governance standard recommendation system based on multimodal large model and NLU-HGSC provided in the embodiment of the present application is mainly composed of four modules: standard management module, directory preprocessing module, NLU-HGSC module and data storage module. The system aims to recommend corresponding data standards for all tables and fields in the library to be governed through NLU-HGSC (heterogeneous graph similarity computation based on natural language understanding, i.e. a heterogeneous graph similarity computation technology based on natural language understanding). Provide authoritative standard information such as national standards, local standards, industry standards, enterprise standards and group standards as an important basis for subsequent data governance work. Through this technology, standard files can be parsed more accurately, the scope of the field to which the recommended data belongs can be adaptively recommended, and suitable data standards can be recommended for various data scenarios.

[0028] In the embodiments of the present application, the standard management module is responsible for the collection and analysis of standards. First, through the standard collection submodule, relevant standard documents are collected from multiple sources, and the release, implementation and update status of the standards are paid attention to; then, the standard documents are analyzed using layout analysis technology, and the table in the image format is further parsed into a text format using a multimodal large model, providing a data basis for the similarity graph calculation module in the subsequent NLU-HGSC module.

[0029] Furthermore, Figure 2 As shown, the standard management module includes two submodules: standard collection and standard parsing. First, through the standard information collection function, the system automatically collects standard files and pays attention to the implementation status and update status of the standards. If there is a newly released standard, the standard file will be automatically downloaded, and it will be determined whether it is recommended based on the implementation status. If the standard file is updated, the latest standard file is obtained and a new standard library is added. The formats of these standard files include but are not limited to PDF, Word, JPG, JPEG, PNG, BMP, HTML, XLSX, JSON, etc., covering various types such as national standards, local standards, and industry standards.

[0030] Secondly, the collected standard files will be input into the layout analysis unit, which uses target detection technology to parse the content in the standard files. The parsed content includes but is not limited to key-value pairs such as the standard Chinese name, standard English name, release time, implementation time, release department, standard number, standard type, as well as table names, charts, pictures and other information. The table content in the picture format is input into the multimodal large model for table parsing, and the text information such as the field name, field type and field length of the table in the picture is output. By combining layout analysis technology with multimodal large model technologies such as table parsing, rich node information and hierarchical relationships are extracted from standard files in various formats, providing a solid data foundation for subsequent standard matching.

[0031] Finally, the parsing results are stored in the data storage module and input into the NLU-HGSC module to add domain classification and generate a graph structure.

[0032] like Figure 1 As shown, in the embodiment of the present application, the directory preprocessing module includes a metadata collection and data cleaning module for the database to be matched. The main function of this module is to preprocess the data directory to be matched, including the collection and cleaning of metadata. Ultimately, these data will be generated in the heterogeneous graph generation module in the NLU-HGSC module to generate a graph structure suitable for subsequent similarity calculation with the standard heterogeneous graph.

[0033] Furthermore, first, the metadata information in the database is obtained through the metadata collection module, including table names, field names, and field types. Then these data are cleaned, including removing duplicate fields and invalid characters in field names. Subsequently, as the input of the NLU-HGSC module, the corresponding standard data is recommended for each table in the catalog as the basis for governance.

[0034] like Figure 1 As shown, the NLU-HGSC module described in the embodiment of the present application is the core part of the system, which is used to recommend corresponding data standards to the directory data in the database as the basis for data governance. First, the standard data and directory metadata from the standard management module and the directory preprocessing module are classified by domain adaptation module, and then the two inputs are converted into two heterogeneous graph structures by heterogeneous graph generation module, and the domain classification results are added as nodes to their respective graphs. Subsequently, the semantic understanding module is responsible for converting each natural language node in the above two graph structures into a computable mathematical form; finally, the graph calculation module is responsible for calculating the similarity of the semantically encoded graph data and outputting the standard data corresponding to the directory.

[0035] In the embodiment of the present application, the data storage module uses a variety of database technologies, including relational database MySQL and graph database Neo4j. The combined use of these database technologies ensures that the system can efficiently store and retrieve different types of data.

[0036] In some embodiments, the domain adaptation module is responsible for domain classification of table names in the standard management module and the directory preprocessing module. Specifically, the module implements this function through its internal domain adaptation submodule. This submodule is based on the Few-Shot Learning technology of the multimodal large model, and can flexibly perform domain classification without retraining the model. Its core advantage is that by adjusting the parameters of the input prompt, the submodule can quickly adapt to and learn the classification capabilities of the newly added domains, thereby ensuring the high accuracy and coverage of the classification results. This adaptability not only significantly improves the scalability of the system, but also ensures that when faced with new domains and new standards, it can respond quickly and achieve efficient and accurate classification processing without retraining the system.

[0037] In some embodiments, the heterogeneous graph generation module utilizes the function of the Neo4j graph database and is responsible for converting the output data of the standard management module, the catalog preprocessing module, and the domain adaptation module into graph data. Specifically, the standard management module outputs all parsed standard information, which includes standard domain classification, table name, field name, type and length information, etc.; the catalog preprocessing module outputs the metadata of the database table in the library to be governed, including the library table name, field Chinese name and field type, etc. The heterogeneous graph generation module maps the entities, relationships and attributes generated by each module to the heterogeneous graph structure through the Neo4j graph processing engine, and makes full use of Neo4j's native graph algorithm and query language (Cypher) to realize the complex association and calculation of nodes and edges. At the same time, during the data import process, the system supports dynamic adjustment of the graph data model to ensure the seamless integration and interoperability of data from different sources and types in the same graph, thereby providing a solid data foundation for subsequent data standard recommendations.

[0038] In certain embodiments, the semantic understanding submodule provides semantic understanding for similarity graph calculations in the NLU-HGSC module, and the semantic understanding submodule is mainly responsible for embedding all nodes in the graph. Specifically, this submodule relies on a deep learning language model to convert natural language elements in the graph into highly abstract and computable mathematical representations, thus laying the foundation for subsequent similarity graph calculations. Through this processing, the semantic understanding submodule can effectively capture and characterize the potential semantic relationships and complex associations between nodes, so that the system can maintain high-precision computing capabilities when processing diverse and complex heterogeneous data, thereby providing solid technical support for the accuracy of automated recommendations of data standards.

[0039] In some embodiments, during the process of calculating the similarity of the heterogeneous graph, the heterogeneous graph generation module divides the standard graph data and the metadata graph data into four levels according to different node types, namely, the domain level, the table name level, the field level and the field type level, and calculates the similarity between each level one by one.

[0040] like Figure 3 As shown in the figure, taking the similarity calculation at the field layer as an example, firstly, each field node in the metadata graph data is compared with all the field nodes in the standard graph data one by one, and the similarity between the fields is calculated by semantic similarity. Then, these similarity results are grouped according to the library table name, and the standard field matching result with the highest similarity in each group is selected as the final similarity value of the field node corresponding to the field in the current standard. Subsequently, each standard table name is grouped, and the mean of the similarities of all fields in the group is calculated, which is used as the overall similarity value of the standard table at the field layer. The similarity calculation method at the field type layer is consistent with that at the field layer. Finally, the semantic similarity of the domain layer and the table name layer is calculated. This hierarchical method of similarity calculation for heterogeneous graph data fully considers the impact of information at different levels on the similarity results, based on accuracy and rationality.

[0041] Furthermore, in order to ensure that the importance of each level can be fully reflected in the overall similarity calculation process, different weights are assigned to each level based on expert experience. This weight allocation strategy aims to improve the reliability of standard recommendations. Specifically, after completing the similarity calculation of each level, the graph calculation module integrates the similarity values ​​of each level by weighted summation to generate a comprehensive similarity score. This score is used to quantitatively evaluate the similarity between the heterogeneous graph structure of each table in the database to be governed and the heterogeneous graph structure of the standard data to ensure that the most matching standard data can be accurately identified. Subsequently, the standard with the highest similarity to the database table to be matched is obtained by taking the maximum value in the group according to the database table name grouping, as the final recommendation result. At the same time, the field corresponding to the most matching standard result is the field to which the database table to be matched belongs. This calculation method not only significantly improves the accuracy of standard recommendations, but also enhances the flexibility and adaptability of the system. In particular, when facing the implementation of new standards and the update of existing standards, the system can automatically adapt to these changes, thereby avoiding the tedious retraining process of the model and further improving the efficiency and intelligence level of the system.

[0042] This application is based on a heterogeneous graph similarity calculation method based on multi-node type and multi-level relationship information, and based on natural language understanding technology. It can accurately identify and match isomorphic relationships between different graph structures through semantic understanding. Even when new standards are implemented, current standards are updated, and new field classifications are added, the standard recommendation results and field classification results can be quickly adjusted, avoiding the disadvantage of the traditional method of needing to retrain the model.

[0043] In summary, the present application relates to automated standard parsing technology and NLU-HGSC standard matching technology. Automated standard parsing technology uses a multimodal large model combined with layout analysis technology to realize standard information extraction, and adopts a heterogeneous graph similarity calculation method based on semantic understanding to recommend high-accuracy standard information. First, various elements such as pictures, tables, table names and other information are extracted from the picture through layout analysis technology. Then, the table in the picture format is parsed into a digital table using multimodal large model technology to provide a high-quality data foundation for subsequent matching calculations. Heterogeneous graph similarity calculation is a method based on graph theory that can identify and match isomorphic relationships between different graph structures. Semantic understanding technology is introduced on the basis of heterogeneous graph similarity calculation, based on deep learning algorithms in natural language processing, to understand the meaning of standard names, standard table names and standard field names. This process further improves the accuracy of heterogeneous graph similarity calculation, and ultimately improves the accuracy of governance standard recommendations.

[0044] Please note that the technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. The above embodiments only express several implementation methods of the present application, and their descriptions are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, without departing from the concept of the present application, several variations and improvements can be made, which all belong to the scope of protection of the present application. Therefore, the scope of protection of the patent in this application shall be based on the attached claims.

Claims

1. A domain-adaptive data governance standard recommendation system based on a multimodal large model and NLU-HGSC, characterized by: It includes standard management module, catalog preprocessing module, NLU-HGSC module and data storage module; The standard management module is used for collecting and analyzing standards; The directory preprocessing module is used to collect and preprocess the metadata to be matched; The NLU-HGSC module is used to calculate the similarity between the heterogeneous graph structures generated by the standard management module and the directory preprocessing module, and recommend corresponding data standards for the directory data in the data storage module as a basis for data governance.

2. The domain-adaptive data governance standard recommendation system based on multimodal large model and NLU-HGSC according to claim 1, characterized in that: The standard management module includes a standard collection module and a standard analysis module; The standard collection module is used to collect relevant standard documents from multiple sources and pay attention to the implementation and update status of standards; The standard parsing module is used to perform element parsing on the standard file.

3. The domain-adaptive data governance standard recommendation system based on multimodal large model and NLU-HGSC according to claim 2 is characterized by: The standard document is analyzed for elements using layout analysis technology, wherein the table content in picture format is parsed into text format using a multimodal large model.

4. The domain-adaptive data governance standard recommendation system based on a multimodal large model and NLU-HGSC according to any one of claims 1 to 3, characterized in that: The directory preprocessing module includes a metadata collection module and a metadata cleaning module; The metadata collection module is used to obtain metadata information in the data storage module, including table name, field name and field type; The metadata cleaning module is used to clean the acquired metadata information, including removing duplicate fields and removing invalid characters in field names.

5. The domain-adaptive data governance standard recommendation system based on a multimodal large model and NLU-HGSC according to any one of claims 1 to 3, characterized in that: The NLU-HGSC module includes a domain adaptation module, a heterogeneous graph generation module, a semantic understanding module and a graph computing module; The domain adaptation module is used to perform domain classification on table names in the standard management module and the directory preprocessing module; The heterogeneous graph generation module is used to convert the output data of the standard management module, the catalog preprocessing module and the domain adaptation module into graph data; The semantic understanding module is used to convert each natural language node in the graph data into an abstract and computable mathematical form; The graph calculation module is used to perform similarity calculation on the graph data output by the semantic understanding module and output standard data corresponding to the directory.

6. The domain-adaptive data governance standard recommendation system based on multimodal large model and NLU-HGSC according to claim 5, characterized in that: The domain adaptation module adopts a small sample learning technology based on a multimodal large model, so that it can quickly adapt to and learn the classification capabilities of newly added domains, thereby ensuring the accuracy and coverage of the classification results.

7. The domain-adaptive data governance standard recommendation system based on multimodal large model and NLU-HGSC according to claim 5, characterized in that: In the similarity calculation process, the standard graph data and metadata graph data are divided into four levels according to different node types, namely domain layer, table name layer, field layer and field type layer, and the similarity between each level is calculated one by one.

8. The domain-adaptive data governance standard recommendation system based on multimodal large model and NLU-HGSC according to claim 7, characterized in that: After completing the similarity calculation at each level, the similarity values ​​at each level are integrated by weighted summation to generate a comprehensive similarity score.

9. The domain-adaptive data governance standard recommendation system based on multimodal large model and NLU-HGSC according to claim 7, characterized in that: The data storage module includes the relational database MySQL and the graph database Neo4j.

10. The domain-adaptive data governance standard recommendation system based on multimodal large model and NLU-HGSC according to claim 9, characterized in that: The heterogeneous graph generation module maps entities, relationships and attributes to a heterogeneous graph structure through the graph processing engine of the graph database Neo4j, and uses Neo4j's native graph algorithm and query language to implement the association and calculation of nodes and edges.

Images