Data management method and system

Abstract

The invention provides a data management method and system. The method comprises the following steps: receiving multi-modal data from different data sources; performing entity extraction on the multi-modal data to obtain a plurality of entities and entity data of each entity; each entity and the entity data of each entity are input into a label classification large model, a label of each entity is obtained, and the label classification large model is obtained based on multi-level label system training; according to the multi-level label system and the label of each entity, relation data of each entity is determined, and the relation data is used for showing the entity association relation and the hierarchy of the current entity; and according to a data storage standard, the relational data of each entity and the entity data of each entity, performing hierarchical structured processing on each entity to obtain an entity storage table and a relational storage table.

Description

Technical Field

[0001] The present application relates to the technical field of data management, and in particular to a data management method and system. Background Art

[0002] The data management solutions widely adopted in the current market are generally built on general data management tools and platforms. However, when performing data management tasks, these general data management platforms mainly stay at the level of directly physically storing data. This direct storage method has significant limitations, that is, it fails to effectively capture and reflect the inherent correlation and dependency between data. This neglect of data correlation not only limits the in-depth analysis and utilization value of data, but may also lead to the omission or misunderstanding of key information. In addition, given that data often comes from a variety of data sources, there is a problem of inconsistent data standards between these data from different sources. This inconsistency in data standards greatly increases the complexity and challenge of data standardization management, which seriously affects the efficiency and quality of data integration, analysis and utilization. Therefore, the field of data management urgently needs a new solution that can overcome the above defects and achieve efficient and standardized data management. Summary of the invention

[0003] The embodiments of the present application provide a data management method and system.

[0004] According to a first aspect of the present application, a data management method is provided, the method comprising:

[0005] Receive multimodal data from different data sources;

[0006] Performing entity extraction on the multimodal data to obtain multiple entities and entity data of each entity;

[0007] Input each entity and the entity data of each entity into a label classification model to obtain a label for each entity, wherein the label classification model is trained based on a multi-level label system;

[0008] Determine the relationship data of each entity according to the multi-level label system and the label of each entity, the relationship data is used to show the entity association relationship and hierarchy of the current entity;

[0009] According to the data storage standard, the relational data of each entity and the entity data of each entity, each entity is hierarchically structured to obtain an entity storage table and a relational storage table.

[0010] According to an embodiment of the present application, the multi-level tag system is a tag architecture sorted by level; accordingly, determining the relationship data of each entity according to the multi-level tag system and the tag of each entity includes:

[0011] Matching the label of each entity with the multi-level label system to obtain the level of each entity;

[0012] Obtaining logical data of the level of each entity, where the logical data is used to show other tags connected to the current entity and the connection relationship with other tags;

[0013] The logical data of each entity is matched with the labels of other entities to obtain the entity association relationship of each entity.

[0014] According to one embodiment of the present application, the data storage standard includes field record requirements and format requirements. The field record requirements are to record entities according to the data type of the entity, and the format requirements include naming specifications, format standards and description rules that need to be met when storing entities.

[0015] According to an implementation manner of the present application, the data type includes a business type, a technical type, and an operation type; accordingly, the entity is recorded according to the data type of the entity, including:

[0016] For business-type entities, record the business semantics of the entity;

[0017] For technical entities, record the technical attributes of the entity;

[0018] For entities of operation type, record the operation information of the entity in the process.

[0019] According to an implementation manner of the present application, hierarchical structuring is performed on each entity according to the data storage standard, the relationship data of each entity, and the entity data of each entity to obtain an entity storage table and a relationship storage table, including:

[0020] Standardizing the entity data of each entity according to the data storage standard to obtain standardized data of each entity;

[0021] According to the hierarchy of each entity shown in the relational data of each entity, each entity and its standardized data are saved to the initial entity table to obtain the entity storage table;

[0022] According to the hierarchy of each entity shown in the relationship data of each entity, the relationship data of each entity is saved to the initial relationship table to obtain the relationship storage table.

[0023] According to an embodiment of the present application, the multimodal data includes at least one picture; accordingly, the entity extraction of the multimodal data to obtain multiple entities and entity data of each entity includes:

[0024] Input each picture into the image recognition model to obtain multiple basic features of each picture, and use the multiple basic features as the entity of each picture;

[0025] Extract entity data for each entity from multimodal data.

[0026] According to an embodiment of the present application, the method further includes:

[0027] Generate an image ID for each image;

[0028] Save each picture to a picture storage table based on the picture identifier of each picture;

[0029] According to the picture identifier of each picture, the picture storage table is mapped to the entity storage table and the relationship storage table.

[0030] According to an embodiment of the present application, the method further includes:

[0031] Track the processing of multimodal data and generate lineage information of multimodal data.

[0032] According to an embodiment of the present application, the method further includes:

[0033] According to the data quality monitoring mechanism, quality monitoring is performed on all data in the multimodal data processing process.

[0034] According to a second aspect of the present application, a data management system is provided, the system comprising:

[0035] A receiving module, used for receiving multimodal data from different data sources;

[0036] An entity extraction module, used to extract entities from the multimodal data to obtain multiple entities and entity data of each entity;

[0037] A classification module, used to input each entity and the entity data of each entity into a label classification model to obtain a label for each entity, wherein the label classification model is trained based on a multi-level label system;

[0038] A determination module, used to determine the relationship data of each entity according to the multi-level label system and the label of each entity, the relationship data is used to show the entity association relationship and hierarchy of the current entity;

[0039] The structured processing module is used to perform hierarchical structured processing on each entity according to the data storage standard, the relationship data of each entity and the entity data of each entity to obtain the entity storage table and the relationship storage table.

[0040] The method of the embodiment of the present application receives multimodal data from different data sources; performs entity extraction on the multimodal data to obtain multiple entities and entity data of each entity; inputs each entity and the entity data of each entity into a label classification model to obtain a label for each entity, and the label classification model is trained based on a predefined multi-level label system; determines the relationship data of each entity according to the predefined data hierarchy and the label of each entity, and the relationship data is used to show the entity association relationship and hierarchy of the current entity; hierarchically structures each entity according to the data storage standard, the relationship data of each entity, and the entity data of each entity to obtain an entity storage table and a relationship storage table. By performing entity extraction and label classification on multimodal data and determining the relationship data of each entity, an in-depth exploration of the intrinsic correlation and dependency between data is achieved, avoiding the omission or misunderstanding of key information, and through data storage standards and hierarchical structured processing, standardized data management is achieved, greatly simplifying the process of data integration, analysis and utilization.

[0041] It should be understood that the teachings of the present application are not required to achieve all of the beneficial effects described above, but specific technical solutions can achieve specific technical effects, and other embodiments of the present application can also achieve beneficial effects not mentioned above. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present application will become readily understood. In the accompanying drawings, several embodiments of the present application are shown in an exemplary and non-limiting manner, wherein:

[0043] In the drawings, the same or corresponding reference numerals represent the same or corresponding parts.

[0044] Figure 1 A schematic diagram of the implementation process of the data management method provided in the embodiment of the present application is shown;

[0045] Figure 2 A schematic diagram of the implementation flow of the structured processing operation of the data management method provided in the embodiment of the present application is shown;

[0046] Figure 3 A schematic diagram of the structure of a data management system provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0047] In order to make the purpose, features, and advantages of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0048] Figure 1 A schematic diagram of the implementation process of the data management method provided in an embodiment of the present application is shown.

[0049] refer to Figure 1 , an embodiment of the present application provides a data management method, the method comprising:

[0050] In operation 101 , multimodal data is received from different data sources.

[0051] In one embodiment of the present application, multimodal data for clothing AIGC (Artificial Intelligence Generated Content) is mainly received, that is, the multimodal data is mainly clothing data. Among them, the multimodal data source may include design software, user-generated content (matching pictures uploaded by users), manufacturing factory system (production batch information) and sales platform (product SKU and sales volume), etc.

[0052] Multimodal data generally includes data of multiple modalities, and specifically, multiple modalities may include but are not limited to image modality, text modality, and graphic modality. The embodiment of the present application can receive and manage data of each modality.

[0053] In one embodiment of the present application, based on the importance of the data, multimodal data is divided into metadata and master data. However, since the management method of metadata and master data is the same in this application, no distinction is made between metadata and master data here.

[0054] Operation 102 , performing entity extraction on the multimodal data to obtain a plurality of entities and entity data of each entity.

[0055] In order to better manage data, eliminate data differences between data sources, and improve data quality, data can be converted into entities by materializing the data, and entity data can be used to retain the basic information of the data itself, such as data sources, data relationships, and data indicators. For example, when the multimodal data is a picture of a skirt, the skirt picture can be extracted into entities such as skirt, red, and wool, and all basic information about the skirt picture, such as source, various indicators, etc., are associated with the entity as entity data.

[0056] Among them, the entity extraction process of the data can adopt a conventional entity extraction method, such as using a neural network-based entity extraction model to extract entities, which will not be repeated in this application.

[0057] Operation 103 , input each entity and the entity data of each entity into a label classification model to obtain a label for each entity. The label classification model is trained based on a multi-level label system.

[0058] Based on the business needs and data needs in the field, with business as the main body, and referring to factors such as technology and management, a multi-level label system is configured in advance. Among them, the labels in the multi-level label system correspond to various keywords in the field. For example, in the field of clothing, labels can include color, material, and scene. The hierarchical logic between labels, such as the scene is a high-level label according to business needs, and materials and colors may be configured as low-level labels according to business needs and data needs. The labels are arranged according to their own hierarchy to form a multi-level label system.

[0059] In this embodiment of the present application, the label classification model is trained based on a multi-level label system, and the label classification model can fully understand the labels and the hierarchical logic of the labels. The training of the label classification model can refer to the conventional large model training process, which will not be repeated here.

[0060] For example, taking the entities including red and evening dress, after the entities and entity data are input into the label classification model, the label classification model obtains the label of each entity by analyzing the entities and the corresponding entity data. For example, red corresponds to the low-level color label, and evening dress corresponds to the high-level scene label.

[0061] Operation 104 : determining the relationship data of each entity according to the multi-level tag system and the tag of each entity, where the relationship data is used to show the entity association relationship and hierarchy of the current entity.

[0062] In one embodiment of the present application, the multi-level label system is a label architecture sorted by level; accordingly, the relationship data of each entity is determined based on the multi-level label system and the label of each entity, including: matching the label of each entity with the multi-level label system to obtain the level of each entity; obtaining the logical data of the level of each entity, the logical data is used to show other labels connected to the current entity and the connection relationship with other labels; matching the logical data of each entity with the labels of other entities to obtain the entity association relationship of each entity.

[0063] The multi-level label system is a label architecture obtained by connecting multiple labels in hierarchical order. By matching the label of each entity with the multi-level label system, the position of each entity's label in the multi-level label system can be obtained, and the level of each label's position can be obtained, that is, the level of each entity can be obtained.

[0064] After that, the logical data of the position of each tag is obtained, and other tags connected to the position of each tag can be obtained. The tags of other entities obtained by multimodal data processing are matched with the logical data of the current entity. When the tags of other entities exist in the logical data of the current entity, the existing entity is determined to be an entity that has an association relationship with the current entity. The operation is repeated for each entity to obtain the entity association relationship of each entity.

[0065] The entity association relationship includes not only the connected entities of each entity, but also the connection relationship with the connected entities, which is a hierarchical relationship. For example, when the red entity and the evening dress entity are the connected entities of the skirt entity, the entity association relationship includes the connected entities red and evening dress, and also includes the hierarchical relationship between the skirt and the evening dress, and the hierarchical relationship between the skirt and the color.

[0066] Operation 105 , hierarchical structure processing is performed on each entity according to the data storage standard, the relationship data of each entity, and the entity data of each entity to obtain an entity storage table and a relationship storage table.

[0067] In one embodiment of the present application, the data storage standard includes field record requirements and format requirements. The field record requirements are to record entities according to the data type of the entity, and the format requirements include naming specifications, format standards and description rules that need to be met when storing entities.

[0068] In one embodiment of the present application, data types include business types, technical types and operation types; accordingly, entities are recorded according to their data types, including: for business type entities, recording the business semantics of the entities; for technical type entities, recording the technical attributes of the entities; and for operation type entities, recording the operation information of the entities in the process.

[0069] In order to ensure the subsequent standardized management of data, data storage standards are designed in advance. The data storage standards mainly include field recording requirements and format requirements.

[0070] Since entities usually belong to different data types, and different data types have different definition standards, field records are required to record entities according to their data types. Among them, for data in the clothing field, data types are usually divided into business type, technical type and operation type. For business type data, it is necessary to record its business semantics, such as clothing categories (such as "men's clothing" or "women's clothing") and field meanings (such as "color represents the main color"). For technical type entities, it is necessary to record their technical properties, such as table structure, field data representation type (such as string VARCHAR, integer INT) and data storage path. For operation type entities, it is necessary to record their operation information in the process, such as data source, generation process, loading time, etc.

[0071] In order to unify the format of data, format requirements are also configured. The format requirements mainly include the naming specifications, format standards and description rules that need to be met when storing entities. Among them, the naming specifications can include camel_case for table names and snake_case for field names, the format standards can include the unified YYYY-MM-DD format for date fields, and the description rules can include field description definitions, such as field descriptions that include name, data type, unit, business rules and example values.

[0072] Through the pre-configured data storage standards, hierarchical structured processing of entities can be achieved. The purpose of hierarchical structured processing is mainly to convert entities into hierarchical structured data, that is, to store entities in data tables. The storage of entities includes the storage of entity data itself, that is, the storage of entities and entity data, and also includes the storage of entity relationship data. An entity storage table can be generated for storage of entity data itself, and a relationship storage table can be generated for storage of relationship data. When storing entity data itself and relationship data, they are stored level by level according to the entity's hierarchy.

[0073] In one embodiment of the present application, an identifier is also set for each entity, and the entity storage table and the relationship storage table are associated through the identifier to facilitate subsequent data query, analysis and modification.

[0074] In this way, the embodiments of the present application, by performing entity extraction and label classification on multimodal data and determining the relationship data of each entity, achieves in-depth exploration of the intrinsic correlation and dependency between data, avoids omission or misunderstanding of key information, and realizes standardized management of data through data storage standards and hierarchical structured processing, greatly simplifying the process of data integration, analysis and utilization.

[0075] Figure 2 A schematic diagram of the implementation flow of the structured processing operation of the data management method provided in an embodiment of the present application is shown.

[0076] refer to Figure 2 In one embodiment of the present application, the above operation 105 performs hierarchical structure processing on each entity according to the data storage standard, the relationship data of each entity and the entity data of each entity to obtain the entity storage table and the relationship storage table, including:

[0077] Operation 201 , standardizing entity data of each entity according to a data storage standard to obtain standardized data of each entity;

[0078] Operation 202 , according to the hierarchy of each entity shown in the relational data of each entity, save each entity and its standardized data to an initial entity table to obtain an entity storage table;

[0079] Operation 203 : According to the hierarchy of each entity shown in the relationship data of each entity, the relationship data of each entity is saved to the initial relationship table to obtain a relationship storage table.

[0080] The corresponding hierarchy is shown in the relational data of each entity. All entities are sorted according to the hierarchy to obtain an entity structure system. After all entities are standardized based on the data storage standard, the entities and entity data can be saved to the initial entity table according to the entity structure system to obtain an entity storage table, and the relational data of the entities can be saved to the initial relational table to obtain a relational storage table.

[0081] In one embodiment of the present application, multimodal data may include at least one picture, and entity extraction is performed on the multimodal data to obtain multiple entities and entity data of each entity, including: inputting each picture into an image recognition model to obtain multiple basic features of each picture, and using the multiple basic features as the entity of each picture; extracting the entity data of each entity from the multimodal data.

[0082] Multimodal data usually includes direct data and data based on files, documents or pictures. For direct data, entity extraction can be performed directly without additional operations, but for data based on files, documents and pictures, additional recognition processes must be added. Among them, entity recognition for pictures can be performed based on image recognition models, which can extract basic features of images, and the basic features are entities. For entity extraction from documents and files, refer to the conventional document and file keyword extraction method. Keywords are entities, which will not be repeated here.

[0083] After extracting the entities in the image, entity-related description data is also extracted from the multimodal data as entity data to better understand the entities.

[0084] In one embodiment of the present application, when the multimodal data includes pictures, a picture identifier is also generated for each picture; each picture is saved to a picture storage table based on the picture identifier of each picture; and according to the picture identifier of each picture, the picture storage table is mapped to an entity storage table and a relationship storage table.

[0085] Since the image itself also contains valid information, it is necessary to store the image itself. The image can be stored by saving it as an entity in the image storage table. In order to facilitate the association and retrieval of the image, when saving the image, a picture identification URL is also generated for the image, and the image is saved in the image storage table through the picture identification. After the image is stored in the image storage table, in order to associate the image with the entity, the image identification of the image is also mapped to the entity storage table and the relationship storage table, so that when querying or modifying based on the image, the unified query and modification of the entity, entity data and entity relationship data can be associated.

[0086] In one embodiment of the present application, the secondary developed Apache Atlas (data governance and metadata management platform) is also introduced to expand multimodal support, specifically to provide the following functions externally:

[0087] Add, delete, modify and query functions: For example, a graphical interface can be provided to support users (such as fashion designers and data analysts) to easily query data and dependencies between data, that is, the association between entities.

[0088] Batch operation function: For example, it supports users to batch modify and update all tags corresponding to all pictures, documents, etc.

[0089] Data quality management functions: For example, support for detecting unrelated entities and tags in images, support for maintaining data version history, recording data change trajectories, and support for rollback.

[0090] For example, when a user needs to change a skirt picture in multimodal data, the introduction of the above platform can support the user to select a skirt picture and modify the entity corresponding to the label. After the modification is completed, the platform can automatically update the picture storage table, entity storage table, and relationship storage table.

[0091] In one embodiment of the present application, data storage rules are also configured, and the data storage rules are: for structured data (such as material and color), a relational data repository is used for storage; for unstructured data (such as image files), a distributed file system is used for storage, and the associated structured data is used for indexing.

[0092] In one embodiment of the present application, the processing process of the multimodal data is also tracked, and the lineage information of the multimodal data is generated.

[0093] In order to realize the data lineage function, the entire multimodal data processing process is also tracked to generate multimodal data lineage information. The lineage information records the data generation, processing and consumption process. By tracing the lineage information, it can support abnormal tracing and positioning, core business analysis, task monitoring and alarm functions.

[0094] For example, when the problem of "color deviation of the generated clothing design image" occurs, the lineage information can be traced back to the error in data processing of a certain data field (entity or label), such as the color being incorrectly mapped during the image generation stage.

[0095] In one embodiment of the present application, in order to ensure data security, different data access permissions are assigned to different user roles (such as designers, operators, developers), sensitive data is dynamically masked based on the data protection mechanism, and user access behavior is recorded in real time.

[0096] In one embodiment of the present application, quality monitoring is also performed on all data in the multimodal data processing process according to a data quality monitoring mechanism.

[0097] In order to ensure the quality of data, data quality monitoring tools are introduced to track the parsing, storage and use of multimodal data (such as pictures, texts, design documents, etc.) in real time according to the data quality monitoring mechanism, so as to timely identify errors in the data and automatically correct them to ensure that the data in each link is accurate and reliable. Among them, the data quality monitoring mechanism mainly includes the monitoring mechanism for the accuracy, completeness, consistency and timeliness of the data, and different monitoring mechanisms are set for data of different modes, that is, different accuracy, completeness, consistency and timeliness requirements are set for data of different modes.

[0098] In one embodiment of the present application, in order to be able to correct the data in a timely manner when problems with data quality are monitored, an automatic data correction mechanism is also provided. The automatic data correction mechanism may be manual intervention correction or automated correction.

[0099] In one embodiment of the present application, in order to ensure synchronization of data after correction, an automatic synchronization mechanism is also provided, and the automatic synchronization mechanism is used to synchronize the corrected data when the data correction is completed.

[0100] In one embodiment of the present application, in order to enable the staff to understand the data situation, a monitoring report is also generated and fed back to the staff.

[0101] In one embodiment of the present application, in order to ensure the accuracy of the data quality monitoring mechanism, the data quality monitoring mechanism is continuously iteratively optimized through real-time data.

[0102] Figure 3 A schematic diagram of the structure of a data management system provided in an embodiment of the present application is shown.

[0103] refer to Figure 3 Based on the above data management method, the embodiment of the present application also provides a data management system, which includes:

[0104] A receiving module 10, configured to receive multimodal data from different data sources;

[0105] An entity extraction module 20, used to extract entities from multimodal data to obtain multiple entities and entity data of each entity;

[0106] A classification module 30, for inputting each entity and the entity data of each entity into a label classification model to obtain a label for each entity, wherein the label classification model is trained based on a multi-level label system;

[0107] A determination module 40, for determining the relationship data of each entity according to the multi-level label system and the label of each entity, wherein the relationship data is used to show the entity association relationship and hierarchy of the current entity;

[0108] The structured processing module 50 is used to perform hierarchical structured processing on each entity according to the data storage standard, the relationship data of each entity and the entity data of each entity to obtain an entity storage table and a relationship storage table.

[0109] In one embodiment of the present application, the system also includes a data governance and metadata management platform for tracking the processing of multimodal data and generating lineage information of multimodal data.

[0110] In one embodiment of the present application, the system also includes a data quality monitoring tool for performing quality monitoring on all data in the multimodal data processing process according to a data quality monitoring mechanism.

[0111] It should be noted that the description of the system in the embodiment of the present application is similar to the description of the method embodiment described above, and has similar beneficial effects as the method embodiment, so it will not be repeated. Figure 1 to Figure 2 The present invention can be understood by referring to the description of any one of the accompanying drawings.

[0112] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this application can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution disclosed in this application can be achieved, and this document is not limited here.

[0113] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A data management method, characterized in that: The method comprises: Receive multimodal data from different data sources; Performing entity extraction on the multimodal data to obtain multiple entities and entity data of each entity; Input each entity and the entity data of each entity into a label classification model to obtain a label for each entity, wherein the label classification model is trained based on a multi-level label system; Determine the relationship data of each entity according to the multi-level label system and the label of each entity, the relationship data is used to show the entity association relationship and hierarchy of the current entity; According to the data storage standard, the relational data of each entity and the entity data of each entity, each entity is hierarchically structured to obtain an entity storage table and a relational storage table.

2. The method according to claim 1, characterized in that The multi-level tag system is a tag architecture sorted by level; accordingly, determining the relationship data of each entity according to the multi-level tag system and the tag of each entity includes: Matching the label of each entity with the multi-level label system to obtain the level of each entity; Obtaining logical data of the level of each entity, where the logical data is used to show other tags connected to the current entity and the connection relationship with other tags; The logical data of each entity is matched with the labels of other entities to obtain the entity association relationship of each entity.

3. The method according to claim 1, characterized in that The data storage standard includes field record requirements and format requirements. The field record requirements are to record entities according to the data type of the entity. The format requirements include naming specifications, format standards and description rules that need to be met when storing entities.

4. The method according to claim 3, characterized in that: The data types include business types, technical types and operation types; accordingly, entity records are performed according to the entity data types, including: For business-type entities, record the business semantics of the entity; For technical entities, record the technical attributes of the entity; For entities of operation type, record the operation information of the entity in the process.

5. The method according to claim 1, characterized in that According to the data storage standard, the relationship data of each entity and the entity data of each entity, each entity is hierarchically structured to obtain an entity storage table and a relationship storage table, including: Standardizing the entity data of each entity according to the data storage standard to obtain standardized data of each entity; According to the hierarchy of each entity shown in the relational data of each entity, each entity and its standardized data are saved to the initial entity table to obtain the entity storage table; According to the hierarchy of each entity shown in the relationship data of each entity, the relationship data of each entity is saved to the initial relationship table to obtain the relationship storage table.

6. The method according to claim 1, characterized in that The multimodal data includes at least one picture; accordingly, the entity extraction of the multimodal data to obtain multiple entities and entity data of each entity includes: Input each picture into the image recognition model to obtain multiple basic features of each picture, and use the multiple basic features as the entity of each picture; Extract entity data for each entity from multimodal data.

7. The method according to claim 6, characterized in that The method further comprises: Generate an image ID for each image; Save each picture to a picture storage table based on the picture identifier of each picture; According to the picture identifier of each picture, the picture storage table is mapped to the entity storage table and the relationship storage table.

8. The method according to claim 1, characterized in that The method further comprises: Track the processing of multimodal data and generate lineage information of multimodal data.

9. The method according to claim 1, characterized in that: The method further comprises: According to the data quality monitoring mechanism, quality monitoring is performed on all data in the multimodal data processing process.

10. A data management system, characterized in that: The system comprises: A receiving module, used for receiving multimodal data from different data sources; An entity extraction module, used to extract entities from the multimodal data to obtain multiple entities and entity data of each entity; A classification module, used to input each entity and the entity data of each entity into a label classification model to obtain a label for each entity, wherein the label classification model is trained based on a multi-level label system; A determination module, used to determine the relationship data of each entity according to the multi-level label system and the label of each entity, the relationship data is used to show the entity association relationship and hierarchy of the current entity; The structured processing module is used to perform hierarchical structured processing on each entity according to the data storage standard, the relationship data of each entity and the entity data of each entity to obtain the entity storage table and the relationship storage table.

Images