Master data management method, system and device, medium and program product

Abstract

The invention discloses a master data management method, system and device, a medium and a program product, and the method comprises the steps: firstly receiving master data input by a user, then carrying out the entity matching of the master data input by the user and data in a master database through an entity matching model, and obtaining a data entity matching result; and finally, performing data entity management operation on the main data input by the user according to the data entity matching result. The entity matching model is constructed by adopting a first pre-training language model based on a multi-modal fusion strategy and a second pre-training language model based on a hierarchical graph attention network, and information between vocabularies and texts can be effectively captured, so that context information is effectively captured, and the semantics and contexts of the vocabularies are understood; by adopting the entity matching model constructed by the method to carry out entity matching, the accuracy of entity matching can be improved, so that the consistency and accuracy of data in the whole main data system are ensured.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a master data management method, system, device, medium and program product. Background Art

[0002] Entity matching, also known as entity alignment (EA) or entity resolution, is one of the most basic and critical technologies in knowledge fusion. The goal of entity matching is to identify entities from different knowledge graphs that describe the same real-world objects. The data about business entities is called master data, which provides contextual information for enterprise business transactions and analysis. It mainly includes the core data of key business entities that are shared, used and maintained in large quantities across multiple systems, applications and business processes within an organization. The purpose is to ensure the consistency, accuracy and reliability of data within the organization. By adopting a master data management system, basic data resources can be provided for the existing systems of the enterprise, so that basic data can be fully applied in various business areas such as enterprise finance, centralized material management, sales, and production.

[0003] The main process of entity matching is to analyze whether the newly added data entities match the entities of each data in the main data system, sort them according to whether they match and the degree of match (confidence), and finally process the data by adding or merging. However, the entity matching model built based on neural networks (such as convolutional neural networks CNN or recurrent neural networks RNN) used in existing related technologies cannot effectively capture the information between words and texts, and cannot effectively capture contextual information and understand the semantics and context of words. Therefore, using the entity matching model built in the existing way for entity matching will result in low accuracy of entity matching, and thus cannot ensure the consistency and accuracy of data in the entire main data system. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a master data management method, system, device, medium and program product, which can improve the accuracy of entity matching and ensure the consistency and accuracy of data in the master data management system.

[0005] To solve the above technical problems, an embodiment of the present invention provides a master data management method, including:

[0006] Receive master data input by the user;

[0007] Performing entity matching on the main data input by the user and the data in the main database through an entity matching model to obtain a data entity matching result; wherein the entity matching model is constructed by using a first pre-trained language model based on a multimodal fusion strategy and a second pre-trained language model based on a hierarchical graph attention network;

[0008] A data entity management operation is performed on the master data input by the user according to the data entity matching result.

[0009] Preferably, the entity matching model is trained by the following steps:

[0010] Acquire first training data;

[0011] Preprocessing the first training data to obtain first entity matching pair data;

[0012] Inputting the first entity matching pair data into the entity matching model for model training;

[0013] Wherein, during the training process, data enhancement is performed on the first entity matching pair data through a first pre-trained language model based on a multimodal fusion strategy to obtain second training data;

[0014] Performing entity matching on the second training data by using a second pre-trained language model based on a hierarchical graph attention network to obtain second entity matching pair data;

[0015] The second entity matching pair data is input into the entity matching model as the third training data, and the entity matching model is retrained until a preset stop condition is met, thereby obtaining a trained entity matching model.

[0016] Preferably, the step of performing data enhancement on the first entity matching pair data by using a first pre-trained language model based on a multimodal fusion strategy to obtain second training data includes:

[0017] Inputting the first entity matching pair data into the first pre-trained language model, and serializing the first entity matching pair data;

[0018] Injecting domain knowledge into the serialized first entity matching data;

[0019] The first entity matching data after domain knowledge injection is enhanced to obtain second training data.

[0020] Preferably, performing entity matching on the second training data by using a second pre-trained language model based on a hierarchical graph attention network to obtain second entity matching pair data includes:

[0021] Inputting the second training data into the second pre-trained language model to construct a hierarchical heterogeneous graph;

[0022] Perform context embedding calculation on the constructed hierarchical heterogeneous graph to obtain context information data;

[0023] The context information data is used to perform data entity matching to obtain second entity matching pair data.

[0024] Preferably, the using the context information data to perform data entity matching to obtain second entity matching pair data includes:

[0025] Comparing entities in the context information data to obtain similarity scores between the entities;

[0026] The entities in the context information data are aggregated according to the similarity scores to obtain second entity matching pair data.

[0027] Preferably, the preprocessing of the first training data to obtain first entity matching pair data includes:

[0028] The first training data is filtered through the blocker to obtain entity candidate pair data;

[0029] The entity candidate pair data is labeled to obtain first entity matching pair data.

[0030] Preferably, after performing data enhancement on the first entity matching pair data through a first pre-trained language model based on a multimodal fusion strategy to obtain second training data, the method further includes:

[0031] The second training data is input into the blocker to perform enhanced training on the blocker.

[0032] To solve the above technical problems, an embodiment of the present invention further provides a master data management system, including:

[0033] A data receiving module, used for receiving master data input by a user;

[0034] A data entity matching module is used to perform entity matching on the main data input by the user and the data in the main database through an entity matching model to obtain a data entity matching result; wherein the entity matching model is constructed by using a first pre-trained language model based on a multimodal fusion strategy and a second pre-trained language model based on a hierarchical graph attention network;

[0035] The data entity management module is used to perform data entity management operations on the master data input by the user according to the data entity matching result.

[0036] To solve the above technical problems, an embodiment of the present invention further provides a master data management device, comprising: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the master data management method as described in any one of the above when executing the computer program.

[0037] To solve the above technical problems, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the master data management methods described above.

[0038] To solve the above technical problems, an embodiment of the present invention further provides a computer program product, including a computer program / instruction, which implements any of the master data management methods described above when executed by a processor.

[0039] Compared with the prior art, the embodiments of the present invention provide a master data management method, system, device, medium and program product. The method first receives the master data input by the user, then performs entity matching on the master data input by the user with the data in the master database through the entity matching model to obtain the data entity matching result, and finally performs data entity management operation on the master data input by the user according to the data entity matching result. Among them, the entity matching model of the present invention is constructed by using a first pre-trained language model based on a multimodal fusion strategy and a second pre-trained language model based on a hierarchical graph attention network, which can effectively capture the information between words and texts, thereby effectively capturing contextual information and understanding the semantics and context of words. The entity matching model constructed by the present invention is used for entity matching, which can improve the accuracy of entity matching, thereby ensuring the consistency and accuracy of data in the entire master data system. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solution of the present invention, the drawings used in the implementation mode will be briefly introduced below. Obviously, the drawings described below are only some implementation modes of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 is a flow chart of a master data management method provided by an embodiment of the present invention;

[0042] Figure 2 Schematic diagram of a system framework for entity matching using a Ditto model provided by an embodiment of the present invention;

[0043] Figure 3It is a schematic diagram of the structure of a hierarchical heterogeneous graph provided by an embodiment of the present invention;

[0044] Figure 4 It is a flowchart of constructing a hierarchical heterogeneous graph provided by an embodiment of the present invention;

[0045] Figure 5 It is a schematic diagram of context embedding using the HierGAT model provided by an embodiment of the present invention;

[0046] Figure 6 Schematic diagram of a system framework for entity matching using a HierGAT model provided by an embodiment of the present invention;

[0047] Figure 7 It is an entity matching training flow chart of an entity matching model provided by an embodiment of the present invention;

[0048] Figure 8 is another flow chart of a master data management method provided by an embodiment of the present invention;

[0049] Fig. 9 is another flow chart of a master data management method provided by an embodiment of the present invention;

[0050] Fig.10 A structural block diagram of a master data management system provided by an embodiment of the present invention;

[0051] Fig.11 It is a schematic diagram of the interaction of various modules in a master data management system provided by an embodiment of the present invention;

[0052] Fig.12 is another schematic diagram of interaction between modules in a master data management system provided by an embodiment of the present invention;

[0053] Fig.13 It is a structural block diagram of a master data management device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] See also Figure 1 , Figure 1 is a flow chart of a master data management method provided by an embodiment of the present invention, such as Figure 1 As shown, the master data management method includes steps S1 to S3:

[0056] S1. Receive master data input by the user;

[0057] S2. Entity matching is performed on the main data input by the user and the data in the main database through an entity matching model to obtain a data entity matching result; wherein the entity matching model is constructed by using a first pre-trained language model based on a multimodal fusion strategy and a second pre-trained language model based on a hierarchical graph attention network;

[0058] S3. Perform data entity management operations on the master data input by the user according to the data entity matching result.

[0059] Exemplarily, after the user inputs the master data, the preprocessed master data is imported into the entity matching model, and then reasoning calculations are performed to obtain the matching scores between the input master data and each entity in the master database. The matching scores are then sorted from high to low, and finally the top five data entities are displayed. The user can determine whether the input master data already exists in the master database and perform data entity management operations such as adding entities or merging entities, or can set it to automatically determine the entity that best matches the input master data (such as the entity with the highest matching score) for a merge operation.

[0060] In an optional embodiment, the entity matching model is trained by the following steps:

[0061] Acquire first training data;

[0062] Preprocessing the first training data to obtain first entity matching pair data;

[0063] Inputting the first entity matching pair data into the entity matching model for model training;

[0064] Wherein, during the training process, data enhancement is performed on the first entity matching pair data through a first pre-trained language model based on a multimodal fusion strategy to obtain second training data;

[0065] Performing entity matching on the second training data by using a second pre-trained language model based on a hierarchical graph attention network to obtain second entity matching pair data;

[0066] The second entity matching pair data is input into the entity matching model as the third training data, and the entity matching model is retrained until a preset stop condition is met, thereby obtaining a trained entity matching model.

[0067] In an optional embodiment, the step of performing data enhancement on the first entity matching pair data by using a first pre-trained language model based on a multimodal fusion strategy to obtain second training data includes:

[0068] Inputting the first entity matching pair data into the first pre-trained language model, and serializing the first entity matching pair data;

[0069] Injecting domain knowledge into the serialized first entity matching data;

[0070] The first entity matching data after domain knowledge injection is enhanced to obtain second training data.

[0071] The step of performing entity matching on the second training data by using a second pre-trained language model based on a hierarchical graph attention network to obtain second entity matching pair data includes:

[0072] Inputting the second training data into the second pre-trained language model to construct a hierarchical heterogeneous graph;

[0073] Perform context embedding calculation on the constructed hierarchical heterogeneous graph to obtain context information data;

[0074] The context information data is used to perform data entity matching to obtain second entity matching pair data.

[0075] In an optional embodiment, the using the context information data to perform data entity matching to obtain second entity matching pair data includes:

[0076] Comparing entities in the context information data to obtain similarity scores between the entities;

[0077] The entities in the context information data are aggregated according to the similarity scores to obtain second entity matching pair data.

[0078] In an optional embodiment, preprocessing the first training data to obtain first entity matching pair data includes:

[0079] The first training data is filtered through the blocker to obtain entity candidate pair data;

[0080] The entity candidate pair data is labeled to obtain first entity matching pair data.

[0081] In an optional embodiment, after performing data enhancement on the first entity matching pair data by using a first pre-trained language model based on a multimodal fusion strategy to obtain second training data, the method further includes:

[0082] The second training data is input into the blocker to perform enhanced training on the blocker.

[0083] Exemplarily, in an embodiment of the present invention, the first pre-trained language model based on the multimodal fusion strategy is specifically a Ditto model, and the second pre-trained language model based on a hierarchical graph attention network is a HierGAT (Hierarchical Graph Attention Networks) model.

[0084] It should be noted that the Ditto model is an entity matching model based on a pre-trained language model. When used for entity matching, it usually includes two modules: a blocker and a matcher, and mainly integrates optimization technologies such as serialization, domain knowledge injection, and data enhancement. The uniqueness of Ditto lies in its multimodal fusion strategy. Ditto is pre-trained on a large-scale multimodal dataset and can then be fine-tuned for specific downstream tasks. This fusion method improves the model's ability to understand cross-modal information.

[0085] Understandably, entity matching focuses on determining whether two data items refer to the same entity.

[0086] Specifically, in general, if two data item sets D and D′ are input, then after entity matching, the output needs to be Where (e, e′)∈M represents the same real-world entity. The data item e is in the form of e={(attri,vali)}, where attri represents the attribute name and vali represents the attribute value in text form.

[0087] Therefore, in entity matching using the Ditto model, the goal of the blocker is to quickly identify a small subset of high-recall candidate pairs D×D′ (i.e., a large portion of the actual matching pairs are this subset); the goal of the matcher is to accurately predict, given a pair of entries, whether they refer to the same real-world entity.

[0088] It should be noted that serialization is the conversion of data into a format that can be transmitted or stored. In the Ditto model, the serialization method is to concatenate the attribute names and attribute values ​​of the data items one by one, and add special tokens to indicate the beginning of the attribute name and the beginning of the attribute value respectively.

[0089] It should be noted that domain knowledge injection is achieved by preprocessing the input sequence (i.e. serializing the data entries) and then injecting domain knowledge into the Ditto model to emphasize which information may be important. This follows the intuition that when human workers make a match / unmatch decision on two data entries, they usually look for text spans that contain key information before making a final decision. For example, span types, the types of labeled spans are a kind of domain knowledge that can be provided to the Ditto model, and can also help the Ditto model avoid mismatches; another example is span normalization, which is a second kind of domain knowledge that can be passed to the Ditto model, that is, rewriting grammatically different but equivalent spans into the same string, so that they will have the same embedding and it is easier for the Ditto model to detect that two spans are the same.

[0090] It should be noted that data augmentation is a commonly used technique in computer vision to generate additional training data from existing examples through simple transformations such as cropping, flipping, rotation, padding, etc. Data augmentation not only adds more training data, but the enhanced data also allows the Ditto model to learn to make predictions unaffected by these transformations, making the trained entity matching model more robust.

[0091] See also Figure 2 , Figure 2 is a schematic diagram of a system framework for entity matching using a Ditto model provided in an embodiment of the present invention, such as Figure 2 As shown, the system includes three modules: a blocker, a Ditto, and a matcher, among which the part in the large dotted box is the Ditto model. First, the two input data item sets, Table A and Table B, are screened by the blocker to obtain several candidate pairs (matched data pairs in Table A and Table B), and the candidate pairs are further processed to obtain samples and labels. The samples and labels are then subjected to serialization, domain knowledge injection, aggregation, and data enhancement steps in turn. Finally, after the above processing steps, a "matching pair" is generated by the matcher. The matching pair is the data pair that is finally determined to be matched in Table A and Table B. In addition, the matching pair data is used to train or enhance the machine learning model to improve the performance of the model. Among them, after data enhancement, it can train the blocker and provide a stronger blocking effect.

[0092] It should be noted that the HierGAT model is also an entity matching model based on a pre-trained language model. When used for entity matching, it usually also includes two modules: blocker and matcher, but it mainly integrates components such as hierarchical heterogeneous graph construction, context embedding calculation, entity aggregation and entity comparison.

[0093] It should be noted that the principles of the blocker and the match in the HierGAT model are similar to those in the above-mentioned Ditto model, and will not be elaborated here.

[0094] It should be noted that in order to represent the structural information of entities, a new hierarchical heterogeneous graph (HHG) is designed in the HierGAT model, represented by G = (V, R), where V and R are the node set and edge (relationship) set respectively. Figure 3 , Figure 3 is a schematic diagram of a hierarchical heterogeneous graph provided by an embodiment of the present invention, such as Figure 3 As shown in the figure, the hierarchical heterogeneous graph consists of three types of nodes: entity nodes e (such as e1, e2, e3), attribute nodes a (such as a1~a6) and vocabulary nodes t (such as t1~t8). Each type forms a layer, which is mainly divided into three layers, namely the entity layer, the attribute layer, and the token layer, which capture the entity-entity relationship, the attribute-entity relationship, and the token-entity relationship respectively. Among them, each node in a hierarchical heterogeneous graph is represented by a semantic vector of the same length, and the vector of the labeled node is its word embedding.

[0095] See also Figure 4 , Figure 4 is a flowchart of constructing a hierarchical heterogeneous graph provided by an embodiment of the present invention, such as Figure 4As shown in Figure 1, a set of entities (e1 and e2) is used as the input of the pairwise entity matching model, or 1+N entities are used as input. For each input entity, the keywords (set of keys) and the corresponding values ​​(tokens for value) in the entity are extracted, and the entity is converted into [<key,[word]> ] pairs, where each keyword (such as title and desc) will become an attribute node a, and the different vocabulary values ​​corresponding to each keyword will become vocabulary nodes t (such as the vocabulary values ​​corresponding to the keyword title in entity e1 are apache and spark, and the vocabulary values ​​corresponding to the keyword desc in entity e1 are cluster and framework; the vocabulary values ​​corresponding to the keyword title in entity e2 are adobe and spark, and the vocabulary values ​​corresponding to the keyword desc in entity e2 are web and framework), and the entity in the hierarchical heterogeneous graph [<key,[word]> ] establishes connections between attribute nodes and vocabulary nodes, and the embeddings of entity nodes are connected by the embeddings of their corresponding attribute nodes (such as the vocabulary nodes web and framework are connected to the attribute node desc, and then the attribute node desc is connected to the corresponding entity node e2).

[0096] It should be noted that in the HierGAT model, the context embedding component first generates context embeddings by considering three levels of context, namely, word level context, attribute level context, and entity level context, and then uses the context embeddings to enrich the original word embeddings to generate word + context (Word plus Context, WPC) embeddings.

[0097] Specifically, see Figure 5 , Figure 5 is a schematic diagram of context embedding using the HierGAT model provided by an embodiment of the present invention, such as Figure 5 As shown in the figure, word + context embedding is divided into vocabulary layer embedding, attribute layer embedding and entity layer embedding. The vocabulary layer embedding performs vocabulary level context embedding through the self-attention layer (Self Attention), the attribute layer embedding performs attribute level context embedding through the graph attention layer (Graph Attention), and the entity layer embedding performs entity level context embedding through the graph attention layer (GraphAttention).

[0098] See also Figure 6 , Figure 6 is a schematic diagram of a system framework for entity matching using a HierGAT model provided by an embodiment of the present invention, such as Figure 6As shown in the figure, the system includes three modules: a blocker, HierGAT, and a matcher, among which the part in the dotted box is the HierGAT model. First, the two input data item sets, Table A and Table B, are screened by the blocker to obtain several candidate items (matched data pairs in Table A and Table B), and the candidate items are then queried by HierGAT to generate entity embedding and entity similarity embedding, that is, hierarchical heterogeneous graph construction, context embedding calculation, entity aggregation, and entity comparison steps are performed, and finally the matching entity pairs are determined by the matcher. Among them, in the entity aggregation step, word + context embedding is hierarchically aggregated to derive entity embedding based on hierarchical heterogeneous graph; entity comparison hierarchically compares the similarity of entities at the attribute level and entity level to derive entity similarity embedding.

[0099] It should be noted that the embodiment of the present invention performs fusion improvement on the above-mentioned first pre-trained language model based on the multimodal fusion strategy (taking the Ditto model as an example) and the second pre-trained language model based on the hierarchical graph attention network (taking the HierGAT model as an example), and obtains an entity matching model provided by the embodiment of the present invention, and trains the entity matching model, and applies the trained entity matching model to a master data management method provided by an embodiment of the present invention.

[0100] See also Figure 7 , Figure 7 The entity matching training flow chart of an entity matching model provided by an embodiment of the present invention is constructed by using a first pre-trained language model based on a multimodal fusion strategy (taking the Ditto model as an example) and a second pre-trained language model based on a hierarchical graph attention network (taking the HierGAT model as an example). Figure 7 As shown, an entity matching model provided by an embodiment of the present invention first performs domain knowledge injection and data enhancement operations of the Ditto model, then constructs a hierarchical heterogeneous graph for context embedding calculation, and performs entity aggregation comparison and matching of the HierGAT model.

[0101] Specifically, Figure 7There are two tables on the far left, marked as "Table A" and "Table B", which are the original data sources for data processing (the first training data). The data in Table A and Table B are then screened and filtered through the blocker to ensure that only qualified data enters the subsequent processing steps; after the data is processed by the blocker, candidate pairs are generated, and the candidate pairs are further processed to generate samples and labels (the first entity matching pair data). The samples and labels are the data used to train the model; then the samples and labels are serialized and injected with domain knowledge. The data injected with domain knowledge is enhanced, and the data enhanced (the second training data) is "aggregated", where aggregation is to integrate the data into a unified format or structure In the process, the aggregated data is used to construct a hierarchical heterogeneous graph to represent the hierarchical relationship and heterogeneity between the data. Then, the context embedding calculation is performed based on the constructed hierarchical heterogeneous graph to convert the data into a vector representation containing context information (context information data). Furthermore, the data after the context embedding calculation is subjected to entity comparison, the entities in the data are compared, and the similarity scores between them are calculated. Finally, entity aggregation is performed based on the similarity scores, and similar entities are merged or aggregated to reduce data redundancy. Finally, after the above processing steps, a matching pair (second entity matching pair data) is generated by a matcher, and the matching pair data (as the third training data) is used to further train the entity matching model to improve the performance of the model. Among them, after data enhancement is performed on the candidate pair data to obtain enhanced training data, the enhanced training data (second training data) can also be input into the blocker to perform enhanced training on the blocker to provide a stronger blocking effect.

[0102] It is worth noting that compared with the separate Ditto model and HierGAT model, the entity matching model obtained by fusion improvement in the embodiment of the present invention has better results in some data sets. For example, when the entity matching model in the embodiment of the present invention is trained and verified using the public data set Megallan for entity resolution tasks and the WDC data set (an ultra-large-scale Chinese dialogue data set), the F1 value is greatly improved.

[0103] See also Figure 8 , Figure 8 is another flow chart of a master data management method provided by an embodiment of the present invention, such as Figure 8As shown, when a user enters new data, he can access the database data through the front-end page, and obtain the user form submitted by the front-end page (i.e., the new data entered by the user) through the back-end, and then initiate an entity resolution request to the artificial intelligence machine learning module (AIML module), respond to the entity resolution request initiated by the back-end through the AIML module, and find the most matching data entity through the AIML model (artificial intelligence machine learning model, i.e., the entity matching model in the embodiment of the present invention), determine whether the new data needs to be inserted or merged into the original entity, and send the final entity resolution / matching result to the back-end service, and present it to the front-end page, so that the user can determine whether to perform data insertion or merging. If so, the back-end service further adds, deletes, modifies, and queries the data in the main data database according to the entity resolution / matching result to perform data insertion or merging. In addition, the data in the main data database and the back-end data can be obtained through the AIML module, the training of the entity matching model and the reasoning of the entity matching model can be performed, and the training and reasoning results can be obtained and sent to the back-end, and the entity matching model can be updated at the same time.

[0104] For example, see Fig. 9 , Fig. 9 It is another flow chart of a master data management method provided by an embodiment of the present invention. After the user inputs data, it is first checked whether the input data conforms to the specification. If not, an error message is prompted. If so, it is converted into JSON and input into the AIML module. The AIML module converts the JSON into a data frame (DataFrame), and then splices it with the data in the database. After splicing, it is converted into a word vector. After vectorization, it is imported into the entity matching model for reasoning, and the matching score with each entity in the main database is calculated. It is sorted from high to low according to the matching score. Finally, the top 5 sorted data are displayed to the user on the front end. The user determines whether the data already exists in the main database, that is, whether the entity of the input data is compared with the entity with a higher matching score in the main database to determine whether the two are the same entity. If not, the entity is added. If so, the entity is merged, or the most matching one is automatically merged.

[0105] In summary, an embodiment of the present invention provides a master data management method, which first receives master data input by a user, then performs entity matching on the master data input by the user with the data in the master database through an entity matching model to obtain a data entity matching result, and finally performs a data entity management operation on the master data input by the user according to the data entity matching result. Among them, the entity matching model of the present invention is constructed using a first pre-trained language model based on a multimodal fusion strategy and a second pre-trained language model based on a hierarchical graph attention network, which can effectively capture the information between vocabulary and text, thereby effectively capturing contextual information and understanding the semantics and context of vocabulary. The entity matching model constructed by the present invention is used for entity matching, which can improve the accuracy of entity matching, thereby ensuring the consistency and accuracy of data in the entire master data system.

[0106] See also Fig.10 , Fig.10 : is a structural block diagram of a master data management system 200 provided in an embodiment of the present invention, wherein the master data management system 200 comprises:

[0107] The data receiving module 21 is used to receive the main data input by the user;

[0108] The data entity matching module 22 is used to perform entity matching on the main data input by the user and the data in the main database through an entity matching model to obtain a data entity matching result; wherein the entity matching model is constructed by using a first pre-trained language model based on a multimodal fusion strategy and a second pre-trained language model based on a hierarchical graph attention network;

[0109] The data entity management module 23 is used to perform data entity management operations on the master data input by the user according to the data entity matching result.

[0110] In an optional embodiment, the master data management system includes several parts including a front-end page, a back-end service, a master database, an artificial intelligence machine learning module (AIML module), and an AIML model (artificial intelligence machine learning model, i.e., an entity matching model in the embodiment of the present invention) module.

[0111] See also Fig.11 , Fig.11 is a schematic diagram of the interaction of various modules in a master data management system provided by an embodiment of the present invention, such as Fig.11As shown, after the user completes the front-end form, click the submit button to submit the data entered in the form. The front-end POSTs (requests) the data to the back-end. The back-end converts the data in the form into a JSON string (JSON String) and passes it to the AIML module. The AIML module converts the JSON into a data frame, splices it with the data in the database, and then converts the data into a word vector, which is input into the entity matching model and generates an inference result. The inference result is sorted according to the matching score. After saving the inference data, the table is converted into a JSON string (JSON String), and the JSON string of the top 5 most matching data is returned, which is output to the front-end and displayed to the user in a table. The user takes the addition and merging operations according to the table, and the front-end passes it to the back-end through a POST request. The back-end then asks the database to execute SQL (the operating language of the relational database) according to the corresponding request to perform the addition or merging operations of the data entity.

[0112] In addition, see Fig.12 , Fig.12 FIG. 1 is another schematic diagram of the interaction between modules in a master data management system provided by an embodiment of the present invention. Fig.12 As shown, the user can also directly set the training parameters of the entity matching model and perform model training. The AIML module pulls the labeled data (i.e., the training data set) from the main database and learns until convergence. After convergence, the entity matching model with the highest F1-Score (F1 value score) is saved for subsequent use.

[0113] It should be noted that a master data management system provided in an embodiment of the present invention is used to execute all process steps of a master data management method in the above embodiment, and the working principles and beneficial effects of the two correspond one to one, so they will not be described in detail.

[0114] See also Fig.13 , Fig.13 3 is a structural block diagram of a master data management device 300 provided in an embodiment of the present invention, the master data management device 300 includes a processor 31, a memory 32, and a computer program stored in the memory 32 and executable on the processor 31. When the processor 31 executes the computer program, the steps in the above-mentioned master data management method embodiments are implemented, such as steps S1 to S3.

[0115] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory 32 and executed by the processor 31 to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the master data management device 300.

[0116] The master data management device 300 may include, but is not limited to, a processor 31 and a memory 32. Those skilled in the art may understand that the schematic diagram is merely an example of the master data management device 300 and does not constitute a limitation on the master data management device 300. The master data management device 300 may include more or fewer components than shown in the diagram, or may combine certain components, or different components. For example, the master data management device 300 may also include input and output devices, network access devices, buses, etc.

[0117] The processor 31 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor 31 is the control center of the master data management device 300, and uses various interfaces and lines to connect various parts of the entire master data management device 300.

[0118] The memory 32 can be used to store the computer program and / or module. The processor 31 implements various functions of the master data management device 300 by running or executing the computer program and / or module stored in the memory 32 and calling the data stored in the memory 32. The memory 32 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory 32 can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0119] Wherein, if the module / unit integrated in the master data management device 300 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor 31, the steps of the above-mentioned various method embodiments can be implemented. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0120] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A master data management method, characterized in that: include: Receive master data input by the user; Performing entity matching on the main data input by the user and the data in the main database through an entity matching model to obtain a data entity matching result; wherein the entity matching model is constructed by using a first pre-trained language model based on a multimodal fusion strategy and a second pre-trained language model based on a hierarchical graph attention network; A data entity management operation is performed on the master data input by the user according to the data entity matching result.

2. The master data management method according to claim 1, characterized in that: The entity matching model is trained by the following steps: Acquire first training data; Preprocessing the first training data to obtain first entity matching pair data; Inputting the first entity matching pair data into the entity matching model for model training; Wherein, during the training process, data enhancement is performed on the first entity matching pair data through a first pre-trained language model based on a multimodal fusion strategy to obtain second training data; Performing entity matching on the second training data by using a second pre-trained language model based on a hierarchical graph attention network to obtain second entity matching pair data; The second entity matching pair data is input into the entity matching model as the third training data, and the entity matching model is retrained until a preset stop condition is met, thereby obtaining a trained entity matching model.

3. The master data management method according to claim 2, characterized in that: The step of performing data enhancement on the first entity matching pair data by using a first pre-trained language model based on a multimodal fusion strategy to obtain second training data includes: Inputting the first entity matching pair data into the first pre-trained language model, and serializing the first entity matching pair data; Injecting domain knowledge into the serialized first entity matching data; The first entity matching data after domain knowledge injection is enhanced to obtain second training data.

4. The master data management method according to claim 3, characterized in that: The step of performing entity matching on the second training data by using a second pre-trained language model based on a hierarchical graph attention network to obtain second entity matching pair data includes: Inputting the second training data into the second pre-trained language model to construct a hierarchical heterogeneous graph; Perform context embedding calculation on the constructed hierarchical heterogeneous graph to obtain context information data; The context information data is used to perform data entity matching to obtain second entity matching pair data.

5. The master data management method according to claim 4, characterized in that: The using the context information data to perform data entity matching to obtain second entity matching pair data includes: Comparing entities in the context information data to obtain similarity scores between the entities; The entities in the context information data are aggregated according to the similarity scores to obtain second entity matching pair data.

6. The master data management method according to claim 2, characterized in that: The preprocessing of the first training data to obtain first entity matching pair data includes: The first training data is filtered through the blocker to obtain entity candidate pair data; The entity candidate pair data is labeled to obtain first entity matching pair data.

7. The master data management method according to claim 6, characterized in that: After performing data enhancement on the first entity matching pair data by using a first pre-trained language model based on a multimodal fusion strategy to obtain second training data, the method further includes: The second training data is input into the blocker to perform enhanced training on the blocker.

8. A master data management system, characterized in that: include: A data receiving module, used for receiving master data input by a user; A data entity matching module is used to perform entity matching on the main data input by the user and the data in the main database through an entity matching model to obtain a data entity matching result; wherein the entity matching model is constructed by using a first pre-trained language model based on a multimodal fusion strategy and a second pre-trained language model based on a hierarchical graph attention network; The data entity management module is used to perform data entity management operations on the master data input by the user according to the data entity matching result.

9. A master data management device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the master data management method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the master data management method according to any one of claims 1 to 7.

11. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the master data management method according to any one of claims 1 to 7 is implemented.

Images