CN114661832B - Multi-mode heterogeneous data storage method and system based on data quality - Google Patents

Abstract

The invention relates to a multi-mode heterogeneous data storage method and a system based on data quality, comprising the following steps: 1) The original text data is stored in a distributed mode in an original database in a key-value format; 2) Carrying out data modeling on the original multimedia data, and carrying out distributed storage in a file database in a file form; 3) Converting the key-value data into relational data, and constructing a relational database; 4) Constructing a graph database according to the relation among the entities in the relation database; 5) Carrying out data modeling on the activity data of the entity in a chained structure to construct a chained database; 6) Converting the multimedia data into text data, and respectively storing the text data in a multimedia database and an original database according to data types; 7) Linking entity data of each sub-database by constructing a multi-level index structure; 8) And constructing a log file maintenance system of the multi-mode database aiming at the data integration method and each sub-database. The method can greatly reduce the time required for inquiring the data and ensure the efficiency of related personnel when using the data.

Description

A multi-modal heterogeneous data storage method and system based on data quality

Technical Field

The present invention belongs to the field of computer technology, and in particular relates to a multimodal database heterogeneous storage method and system based on data quality.

Background Art

Currently, users generate a large amount of user behavior data on different network platforms. These data are no longer single text or image data, but multimodal data including text, images, videos, etc. from different platforms, including structured data, semi-structured data and unstructured data. Structured data refers to data that can be represented and stored using a relational database and is in a two-dimensional form. The general characteristics are: data is in the form of lines, a line of data represents the information of an entity, and the attributes of each line of data are the same; semi-structured data is a form of structured data. It does not conform to the data model structure associated with relational databases or other data tables, but contains relevant tags to separate semantic elements and stratify records and fields. Therefore, it is also called a self-describing structure; unstructured data is data without a fixed structure. Various documents, pictures, videos, audio, etc. are all unstructured data. For this type of data, it is generally stored directly as a whole.

In recent years, with the emergence of massive multimodal data, the cost of data storage has increased. How to build a good and efficient multimodal database has become a problem that most people in the computer industry need to solve together.

Summary of the invention

In order to solve the above technical problems, the present invention provides a multimodal heterogeneous data storage method and system based on data quality.

In order to achieve the above purpose, the present invention adopts the following technical solutions:

A multimodal heterogeneous data storage method and system based on data quality, characterized by comprising the following steps:

1) The present invention targets the original data (including original text data and original multimedia data) from the Internet data source, and stores the original text data in a distributed manner in the original database in a key-value format;

2) Data modeling is performed for the original multimedia data from the Internet, and distributed storage is performed in the file database in the form of files;

3) Convert the original text data into relational data through data integration methods such as event extraction, entity linking, and incomplete data filling, and then model the relational data to build a relational database;

4) Model the entities that have associations with each other in the relational database and the relationships between entities to build a graph database;

5) The activity data of each entity in the relational database has typical time series characteristics. The activity data is modeled in a chain structure to build a chain database;

6) Converting the video data and audio data in the multimedia data into text data through a data conversion method, and storing them in the multimedia database in the form of files, and storing them in the original database in the form of key-value;

7) Optimize different distributed databases based on data quality, link the entity data of each sub-database by building a multi-level index structure to ensure data consistency;

8) Construct a log file maintenance system for the multimodal database based on the data integration method and each sub-database.

On the other hand, the present invention provides a multimodal heterogeneous data storage system based on data quality, including: an original database: used to store original data from Internet data, and the storage format is: key-value format; a relational database: used to convert the key-value data in the original database into relational data, and then model and store it; a graph database: used to graph and store related entities and the relationships between entities in the relational database; a multimedia database: used to store video data and audio data converted into text format; a chain database: used to store the chain structure of each entity activity data in the relational database.

A computer-readable storage medium cluster stores a computer program, which, when executed by a processor, implements five distributed sub-databases described in a multimodal heterogeneous data storage method based on data quality.

Furthermore, the specific method for storing the original data in step 1) is as follows:

1.1) Use MongoDB database system as the database system for key-value data storage. Save the relevant data crawled from the Internet in the form of JSON files and store them in the MongoDB database. MongoDB automatically generates a unique key value for each piece of data as a unique identifier, and each specific piece of data can be located through the key value;

1.2) Use MongoDB Replica Set in MongoDB as a distributed storage solution for the original database. Set the aforementioned computer-readable storage medium cluster to have 1 master node, 1 replica node, and 1 arbitration node according to the MongoDB Replica Set distribution rule. The master node receives all requests, the replica node maintains the same data set as the master node and can participate in the master election, and the arbitration node votes for the master election.

Furthermore, the specific method of multimedia data modeling in step 2) is as follows: multimedia data, including video, audio, and image data, are stored in a distributed file system according to specific rules; wherein the specific rules refer to determining the distributed file system node where the multimedia data is stored according to the data source of the multimedia data. That is, multimedia data, including video, audio, or image data, is stored in a distributed file system according to the storage node corresponding to the data source.

Furthermore, the specific method for constructing the relational data model in step 3) is as follows:

3.1) Use data integration methods, including event extraction, entity linking, and incomplete data filling, to convert raw text data into structured data; among them, event extraction mainly uses specific rules to annotate raw text data to form a data set, uses the data set to train the event extraction model, and stores the results in a structured form; entity linking mainly disambiguates certain specific entities between the results of event extraction and the database, and stores the disambiguated structured data; incomplete data filling mainly uses missing data filling methods to fill in the missing parts of the converted structured data to ensure data integrity;

3.2) Use MySQL database system as the database system for relational data storage. Store the structured data after data integration in the relational database MySQL;

3.3) Use MySQLCluster as a distributed storage solution for relational databases. Set up a computer-readable storage medium cluster as described above with 1 management node, 2 data nodes, and 1 application node according to the MySQLCluster distribution rule. The management node manages related configuration files, the data node stores data in a distributed manner, and the application node performs operations such as reading and writing.

Furthermore, the specific method for constructing the graph database storage model in step 4) is as follows:

4.1) Use HBase as the underlying graph data storage solution. Extract the elements with specific relationships and their relationships in the relational database and store them in HBase. The data in HBase is stored in the form of rows through rowkey. Set the aforementioned computer-readable storage medium cluster to have 1 master node, 1 slave node and 1 backup node according to the HBase distribution rule;

4.2) Use Neo4j as a graph database visualization query solution. Use Hive to export some data from HBase and store it in Neo4j to build a knowledge graph that can meet different query requirements. Establish a mapping between HBase and Hive, restore HBase data to relational database data, and establish relationships between the data through Neo4j;

4.3) The entities in the relational database are modeled through the relationships between entities, forming a series of nodes and edges to represent them, where entities are represented as nodes and relationships are represented as edges, and are visualized through Neo4j.

Furthermore, in the step 5), the chain database exists in the form of a consortium chain and a private chain, wherein the consortium chain stores structured data, and the private chain stores semi-structured and unstructured data, including text, pictures, videos, etc.; the consortium chain uses MySQL database to store structured data, and the private chain uses HDFS to store semi-structured and unstructured data.

Furthermore, the specific method for storing multimedia data in step 6) is as follows:

6.1) Crawl relevant multimedia data from the Internet according to multimedia data sources, including video, audio, image, text, etc.; design multimedia data index tables according to data attributes, and locate the specific location of multimedia data through the index table according to attributes such as data source, data type, storage node, storage path, file name, etc., and store the index table in the form of structured data in a relational database;

6.2) Design a data conversion storage model to convert video data into text data through the "video->audio->text" process; convert audio data into text data through the "audio->text" process; convert image data into text data through the "image->text" process; and store them in the original database and multimedia database.

Furthermore, in step 7), a method for defining data quality includes: accuracy, completeness, consistency, and relevance:

7.1) Accuracy refers to ensuring the accuracy of data conversion through indicators such as the accuracy of conversion models and methods in data integration methods such as event extraction, data filling, data consistency detection and conversion;

7.2) Integrity means that for the original text data of the same entity, there are both structured data after data conversion, semi-structured data in key-value format, and unstructured data in the form of documents in the present multimodal heterogeneous data storage system, and they are stored in both the original database and the relational database. On the other hand, for multimodal data, there are both structured data after multimedia data conversion, semi-structured data in key-value format, and unstructured data in the form of multimedia files in the present multimodal heterogeneous data storage system, and they are stored in both the original database and the multimedia database.

7.3) Consistency refers to the consistency test of the data of the same entity in each sub-database through data consistency test and conversion, including consistency of dimension, consistency of expression, consistency of data value, etc.; ensuring that the relevant data stored in each sub-database is consistent with the original text data and the relevant data of the original multimedia data;

7.4) Association refers to associating various sub-databases with entity id or entity name a. The data of the same entity is updated synchronously in various sub-databases, and the data can be traced through the association of entity id or entity name.

On the other hand, the present invention provides a multi-modal heterogeneous data optimization method based on data quality, including a multi-level index structure and a log file maintenance module;

The multi-level index and dynamic maintenance module includes global index, local index and dynamic maintenance parts:

The global index constructs primary and foreign key indexes between each sub-database of the multimodal database, effectively links each sub-database in the multimodal database, and realizes the query operation of related data;

Local indexing constructs independent index structures in each sub-database of the multimodal database to implement local indexing of the content of each sub-database, including:

The local index module of the original database creates an index for each key of the data and sets the index field as a shard key to improve query efficiency through indexing.

The local index module of the relational database creates indexes for commonly used fields in the data. For example, a commonly used field in a certain entity data is the entity name, and the query efficiency is improved through indexing.

The graph database local index module builds secondary indexes through Apache Phoenix, establishes a mapping between Phoenix and HBase tables, implements operations on HBase tables in Phoenix, and improves query efficiency through indexes.

The local index module of the chain database is mainly divided into name index, sorting establishment, dynamic incremental update and other parts. The name index is established according to specific fields, and the alliance chain is built in chronological order, and the data is dynamically updated incrementally;

The local index module of the multimedia database constructs a local index structure based on the basic information of the multimedia data, including the storage node information, path, file name, extension name and other information of the data, and stores it in the relational database. According to the data source, data type, storage node, storage path, file name and other attributes, the specific location of the multimedia data can be located through the index table.

The log file maintenance module includes log file maintenance for multimodal databases and log file maintenance for data integration. Among them, log file maintenance for multimodal databases includes log file maintenance for relational databases, log file maintenance for graph databases, log file maintenance for chain databases, and log file maintenance for original databases; log file maintenance for data integration includes log file maintenance for event extraction, log file maintenance for entity linking, log file maintenance for incomplete data filling, and log file maintenance for data consistency;

8.1) The log file maintenance of the multimodal database is maintained by each sub-database through the log files of the relevant systems or solutions it uses;

8.2) Data integration log file maintenance refers to recording all data integration operation processes and saving them in the form of log files when data changes occur due to data integration operations during data integration. The content of the log files includes: the time when the database operation occurs, the type of data integration, the category of the operation log, and the unique attribute characteristics of each data integration method. For each type of data integration method, the log content includes: the time when the data integration operation occurs (Timestamp), the data integration type, the level of log record (INFO, WARNING, ERROR, etc.), and for each different type of data integration method, the log content elements specific to each data integration method are designed.

8.3) The data integration methods used in the present invention are divided into event extraction (EE), entity linking (EL), incomplete data filling (DF), and data consistency detection (DC). The log levels are divided into five types: serious errors that cause the application to exit (FATAL), although an error occurs, it does not affect the continued operation of the system (ERROR), potential error situations may occur (WARNING), at the coarse-grained level, emphasizing the entire operation of the application (INFO), and at the fine-grained level, it is very helpful for debugging the application (DEBUG);

8.4) The log file record of event extraction is composed of: [Timestamp][EE][Log level][Event type code][Event ID][Title][Event time];

8.5) The structure of the log file record of the entity link is: [Timestamp][EL][Log level][Entity link type code][Data value corresponding to the unique primary key][Link table name][Data value corresponding to the composite primary key 1 required to be used when linking to the master table][Data value corresponding to the composite primary key 2 required to be used when linking to the master table][Data value corresponding to the composite primary key 3 required to be used when linking to the master table];

8.6) The structure of the log file record filled with incomplete data is: [Timestamp][DF][Log level][Operation content][Operation result];

8.7) The structure of the log file record of data consistency check is: [Timestamp][DC][Log level][Type of data consistency check][Operation content][Operation result].

Beneficial effects created by the present invention: The present invention adopts the above scheme, collects data from different data sources, stores the original data in the original database in the form of key-value, converts the key-value data into relational data through data integration methods such as event extraction, entity linking, and incomplete data filling, stores the data in the relational database in the form of entity-relationship in the graph database, stores the activity data with typical time series characteristics in the relational database in the chain database, stores the multimedia data in the multimedia database, processes the data according to the accuracy, completeness, consistency, and relevance of the data, and finally realizes the heterogeneous storage of data and database optimization. Its advantage is that a variety of database models are designed, and data in different formats are stored according to the storage characteristics of each database to form a multimodal database. The query efficiency is greatly improved through the index structure, and the multimodal database can be fault-recovered through the maintenance of the log file, and the operation process of each data integration method can be viewed. After these steps, a multimodal heterogeneous distributed database system with high data quality is finally obtained. The multimodal database heterogeneous storage method and system of the present invention stores the data into different distributed databases according to different data forms, and optimizes data query, which can greatly reduce the time required for querying data and ensure the efficiency of relevant personnel when using data.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is an architecture diagram of a multimodal heterogeneous data storage system based on data quality;

Figure 2 is a diagram of the original database architecture;

Figure 3 is a diagram of a relational database architecture;

FIG4 is a flowchart of graph database data display;

Fig. 5 is a chain database flow chart;

FIG6 is a flow chart of a global index structure.

DETAILED DESCRIPTION

The technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. In addition, the embodiment described below is only one embodiment of the present invention, rather than all embodiments.

The present invention provides a multimodal heterogeneous data storage method based on data quality, and its design concept is as follows: collect data from different data sources, save the original text data in the form of key-value, and save the multimedia data in the form of files. Secondly, perform data modeling, convert the original text data into relational data and store it in a relational database through a data integration method, extract the elements with specific relationships in the relational database and the relationships between them, store them in HBase, use Hive to export part of the data in HBase and store them in Neo4j to build a graph that can meet different query requirements, and store the original data and compressed multimedia data in a chain database. Finally, perform database optimization, establish a global index structure and a local index structure, as well as a log file maintenance module, to form a multimodal heterogeneous database.

A data quality-based multimodal heterogeneous data storage system framework designed based on this method is shown in Figure 1. It includes: original database, relational database, graph database, chain database, and multimedia database.

The functions of each database are as follows:

Original database: used to store original data from the Internet, the storage format is: key-value format;

Relational database: used to convert key-value data in the original database into relational data, and then model and store it;

Graph database: used to graph and store related entities and relationships between entities in a relational database;

Multimedia database: used to store video data and audio data converted into text format;

Chain database: A chain structure used to store activity data of each entity in a relational database.

The above system is used to implement a multi-modal heterogeneous data storage method based on data quality, and the steps are as follows:

1) Crawl the original data from the Internet and store it in the original database in the form of key-value;

Specifically include:

1.1) Save the relevant data crawled from the Internet in the form of JSON files and store them in the MongoDB database;

1.2) In the present invention, the original database uses MongoDB Replica Set + Sharding cluster to implement the distributed storage mode of the original database. According to the cluster mode adopted, the database is composed of three nodes, namely: the master node, the replica node and the arbitration node. The original database architecture diagram is shown in Figure 2. From a vertical perspective, the three nodes act as three servers respectively, and each server is configured with a routing process, a configuration server process and a corresponding shard. When performing tasks such as storing or reading database operations, the routing process receives the instructions issued by the client and sends the request instructions to the corresponding shard, and the configuration server is responsible for storing the configuration of the metadata in the database. From a horizontal perspective, the present invention designs three shards, and each shard forms a master, backup and arbitration mode on three nodes by using Replica Set.

2) Convert the original text data into relational data through data integration method and store it in relational database;

Specifically include:

2.1) Convert the original text data into relational data and store it in the MySQL database through data integration methods such as event extraction, entity linking, and incomplete data filling;

2.2) In the present invention, the relational database uses MySQL Cluster cluster to implement the distributed storage model of the relational database. According to the cluster mode adopted, the database is composed of four nodes, namely: 1 management node, 2 data nodes, and 1 application node. The relational database architecture diagram is shown in Figure 3. In the relational database model provided by the present invention, the client performs basic operations of the database by connecting to the application node and stores the data in a structured form. After the client operation is completed, the two data nodes will automatically synchronize and replicate the same data to ensure the security of the data. The management node can monitor the status of other nodes at any time and can add and configure new nodes.

3) Extract the elements with specific relationships in the data in the relational database and the relationships between them, store them in the graph database and display them;

Specifically include:

3.1) Extract the elements with specific relationships in the relational database and the relationships between them and store them in HBase;

3.2) Use Hive to export some data from HBase and store it in Neo4j to build a knowledge graph that can meet different query requirements. Establish a mapping between HBase and Hive, restore HBase data to relational database data, and establish relationships between the data through Neo4j.

3.3) Based on the visualization display of Neo4j, the relevant data is extracted from HBase and stored in the Neo4j graph. FIG4 is a flowchart of graph database data display. The graph database uses four nodes to form a cluster to realize the distributed storage of the graph database. In the graph database model provided by the present invention, the client performs basic query operations on the database through Neo4j and displays the data in a visualized form. The underlying graph data is stored through HBase.

4) storing the original data and the compressed multimedia data in a chain database;

The chain database flow chart of step 4) is shown in FIG5 :

In the consortium chain, the json text file is parsed and its corresponding attributes are stored in the corresponding fields of the corresponding table of the MySQL database. In the private chain, the HDFS distributed file system is used to store the original detailed content of the event. Each event corresponds to the text file, picture, video, etc. of the event in the local file system, which are packaged and compressed into a compressed package, and then the hash value of the compressed package is calculated. The hash value is stored in the corresponding hash field of MySQL, and the compressed package is uploaded to the HDFS distributed file system.

5) According to the same data attributes in each sub-database, a global index is constructed. The present invention links the related data in each database through certain specific field indexes, such as relational database, graph database, chain database, multimedia database and original database, and provides three different functions: first, the data query function of related data is realized; second, through the construction of the index, the traceability function of the original data from other databases to the original database is realized; third, after converting the video, audio and image data in the multimedia data into text data, the copy management function of the relational database, original database and multimedia database for the storage of multiple copies of text data is realized. Through the construction of a multimodal global index structure, the linking and traceability functions of related data are realized. The global index structure flow chart is shown in Figure 6:

5.1) In a relational database, the entity basic information table contains basic information attributes of the entity, such as entity ID, entity name, object ID, etc. The entity ID attribute is used to link the entity in the relational database to the original data in JSON format corresponding to the entity in the original database, thus realizing the traceability from relational data to original data;

5.2) Using the entity ID attribute in each entity business data table in the relational database as a foreign key, referencing the basic information data of the entity in the entity basic information table, to achieve the association query function from the basic information data to the business data;

5.3) Through the entity ID attribute of the entity information table in the relational database, link to the multimedia index table corresponding to the entity stored in the relational database, including: storage node information, path, file name, extension name and other data of video, audio, image and text data, to realize the association modification and deletion function from the relational database to the multimedia database;

5.4) Through the multimedia index table, the storage node information, path, file name, extension and other attribute information of the multimedia files (including: video, audio, image, text data) are combined and associated for use, so as to realize the query function of the multimedia index table for the multimedia files stored in each node;

5.5) The chain database uses the entity name as the keyword and uses the chain structure of alliance chain + private chain to store the event information of the entity. Through the entity name in the entity basic information table in the relational database, the link between the relational data and the chain database is realized, and the data of the entity in the chain database is queried;

5.6) In the graph database, the entity name is used as a keyword, and the link relationship between entities with the entity as the central node is stored and displayed by constructing a triple of entity information. Through the entity name in the entity basic information table in the relational database, the link between the relational data and the graph database is realized, and the data of the entity in the graph database and the association relationship between entities are queried;

5.7) In the data conversion provided by the present invention, in addition to being stored in the multimedia database, the text data converted from the video, audio, and image data is also stored as raw data in the original database in order to provide a richer data interface to the outside world. The text data and the JSON formatted files in the original database are associated through the text ID, thereby realizing the link between the text data in the multimedia database and the raw data in the original database.

Claims (10)

1. The multi-mode heterogeneous data storage method based on the data quality is characterized by comprising the following steps of:

1) The method comprises the steps that original data from Internet data are stored in a distributed mode in an original database in a key-value format; the original data comprises original text data and original multimedia data;

2) Modeling the original multimedia data, and storing the data in a file database in a distributed manner in a file form;

3) Converting the original text data into relational data through event extraction, entity linking and incomplete data filling data integration methods, modeling the relational data, and constructing a relational database;

4) Modeling entities with association relations among the entities in the relational database and the relation among the entities to construct a graph database;

5) The activity data of each entity in the relational database has typical time sequence characteristics, and the activity data is subjected to data modeling in a chained structure to construct a chained database;

6) Converting video data and audio data in the multimedia data into text data through a data conversion method, storing the text data in a multimedia database in a file form, and storing the text data in an original database in a key-value format;

7) According to the data quality, database optimization is carried out on different distributed databases, and entity data of each sub-database are linked through constructing a multi-level index structure, so that the consistency of the data is ensured;

8) And constructing a log file maintenance system of the multi-mode database aiming at the data integration method and each sub-database.

2. The method for storing multi-modal heterogeneous data based on data quality according to claim 1, wherein in the step 1), the specific method for storing the original text data in the key-value format in the original database in a distributed manner is as follows:

2.1 A database system using a MongoDB database system as key-value data storage;

2.2 A distributed storage solution using mongo db REPLICA SET in mongo db as the original database.

3. The method according to claim 1, wherein in the step 2), the multimedia data is stored in the distributed file system according to the type of the data source, i.e. the video, audio or picture data is stored in the storage node corresponding to the data source.

4. The method for storing multi-modal heterogeneous data based on data quality according to claim 1, wherein in the step 3), the specific method for constructing the relational database is as follows:

3.1 The original text data is converted into relational data through three data integration methods of relation extraction, entity linking and incomplete data filling;

3.2 Using MySQL database system as the database system for relational data storage;

3.3 Using MySQL Cluster as a distributed storage solution for relational databases.

5. The method for storing multi-modal heterogeneous data based on data quality according to claim 1, wherein in the step 4), the specific method for constructing the graph database is as follows:

4.1 Using HBase as the underlying graph data storage scheme;

4.2 Using Neo4j as a graph database visualization query scheme;

4.3 Modeling entities in a relational database by relationships between the entities.

6. The method for storing multi-modal heterogeneous data based on data quality according to claim 1, wherein in the step 5), the specific method for constructing the chain database is as follows:

The chained database uses MySQL and HDFS as data storage schemes and stores the MySQL and the HDFS into the chained database; the alliance chain stores structured data by MySQL, and the private chain stores semi-structured data by HDFS.

7. The method for storing multi-modal heterogeneous data based on data quality according to claim 1, wherein the specific method for storing the multi-media data in step 6) is as follows:

6.1 Crawling relevant multimedia data including video, audio, images and texts from the Internet according to the multimedia data source; designing a multimedia data index table according to the data attribute, positioning the multimedia data to a specific position through the index table according to the data source, the data type, the storage node, the storage path and the file name attribute, and storing the index table in a relational database in a structured data form;

6.2 A data conversion storage model is designed, and video data is converted into text data through a video- > audio- > text process; converting the audio data into text data through an audio- > text process; converting the image data into text data through an image- > text process; and stored in the original database and the multimedia database.

8. The method for storing multi-modal heterogeneous data based on data quality according to claim 1, wherein in the step 7), the multi-level index structure is composed of a global index and a local index; the dynamic maintenance process is as follows:

Constructing a main external key index among the global index original database, the relational database, the graph database, the chained database and the multimedia database, and effectively linking all the sub databases to realize the query operation of related data; the local index constructs independent index structures in the databases to realize the local index of the contents of each sub-database;

Each sub-database index module is as follows: the original database local index module establishes an index for each key of the data, sets a fragment key for an index field and improves the query efficiency through the index;

The relational database local index module is used for establishing indexes for common fields in data, performing secondary index construction through Apache Phoenix, establishing mapping with a table in HBase in Phoenix, realizing table operation in Phoenix, and improving query efficiency through indexes;

The chain type database local index module is mainly divided into a name index part, a sorting establishment part and a dynamic increment updating part; establishing a name index according to a specific field, simultaneously constructing a alliance chain according to a time sequence, and dynamically updating data increment;

the local index module of the multimedia database constructs a local index structure from basic information of the multimedia data, including storage node information, path, file name and extension information of the data, and stores the local index structure in the relational database, and can locate specific positions of the multimedia data through an index table according to data sources, data types, storage nodes, storage path and file name attributes.

9. The method for storing multi-modal heterogeneous data based on data quality according to claim 1, wherein in the step 8), the specific method for maintaining the log file is as follows:

The log file maintenance is divided into log file maintenance of a multi-mode database and log file maintenance of data integration; the log file maintenance of the multi-mode database comprises log file maintenance of a relational database, log file maintenance of a graph database, log file maintenance of a chained database and log file maintenance of an original database; the log file maintenance of the data integration comprises log file maintenance of event extraction, log file maintenance of entity links, log file maintenance of incomplete data filling and log file maintenance of data consistency.

10. A data quality based multi-modal heterogeneous data storage system for implementing a data quality based multi-modal heterogeneous data storage method as claimed in any one of claims 1 to 9, comprising:

Original database: the storage format is used for storing the original data from the internet data, and the storage format is as follows: a key-value format;

relational database: the method comprises the steps of converting key-value data in an original database into relational data, and building a model for storage;

graph database: the method comprises the steps of being used for imaging and storing related entities and relations among the entities in a relational database;

multimedia database: for storing video data, audio data converted into text format;

Chain database: and the chain structure is used for storing the activity data of each entity in the relational database.