Entity analysis method and device, medium and product

Abstract

The invention discloses an entity analysis method and device, a medium and a product, and relates to the field of entity analysis, and the method comprises the steps: obtaining a data set based on an entity analysis task; each piece of sample data in the data set comprises two entities with different sources and corresponding similar results; generating corresponding sentences for the two entities in each piece of sample data in the data set according to grammatical knowledge; a BERT-Pair-Networks network is trained by using the data set of the generated sentences, and an entity analysis model is obtained; and performing entity analysis on a to-be-analyzed entity by using the entity analysis model. According to the method, the whole sentence can be understood, the strong generalization ability and high expansibility of the entity analysis model are realized, and the accuracy of entity analysis is improved.

Description

Technical Field

[0001] The present application relates to the field of entity resolution, and in particular to an entity resolution method, device, medium and product. Background Art

[0002] Entity resolution (ER) is a research focus in the fields of database, information retrieval, machine learning, natural language processing, etc. It is a prerequisite for data integration. It aims to identify whether tuples from two different sources point to the same object in the real world, or call the two equivalent matches.

[0003] Early methods for solving entity resolution tasks were mainly rule-based methods and traditional machine learning methods. In recent years, deep learning-based methods have been widely used in entity resolution and have achieved ideal results. At present, entity resolution based on deep learning can be divided into two categories: (1) entity resolution models based on attribute features; (2) entity resolution models based on entity types. Among them, entity resolution models based on attribute features include RNN (recurrent neural network) and LSTM (long short-term memory network) combined with attention mechanism. These models can capture semantic and syntactic information to better represent semantic similarity, especially for text attributes. Entity resolution models based on entity types, such as the GraphER model, mainly represent entity records through graph convolutional networks (GCNs), directly extract type comparison features, and softly embed semantic and structural information into graph networks; the GraphER model relies on an end-to-end deep learning framework and solves some challenges faced by attribute-centric methods;

[0004] However, the GraphER model has some shortcomings. First, in terms of feature extraction, although GraphER extracts entity types from the perspective of characters and performs well in capturing word semantics and structural features, it is not good at understanding the overall semantics of a sentence, which helps resolve the ambiguity of entity names and improve the accuracy of entity resolution. Therefore, the GraphER model's processing in feature extraction makes entity resolution inaccurate.

[0005] In terms of scalability, the performance of the GraphER model depends on the quality and structure of the graph structure, which makes the training of the GraphER model relatively complex and requires more computing resources. In particular, as the amount of data increases, the scalability of the model may become a problem.

[0006] In terms of generalization ability, the graph neural network based on the GraphER model may require a lot of work to adjust and optimize the graph structure to adapt to specific tasks when faced with texts in different fields and types.

[0007] In summary, there is an urgent need to provide a new entity resolution method, as the existing GraphER model is not good at understanding the overall sentence semantics and its scalability and generalization capabilities are limited when the amount of data increases or when facing new domain data. Summary of the invention

[0008] The purpose of this application is to provide an entity resolution method, device, medium and product that can understand the entire sentence, achieve strong generalization and high scalability of the entity resolution model, and improve the accuracy of entity resolution.

[0009] To achieve the above objectives, this application provides the following solutions:

[0010] In a first aspect, the present application provides an entity resolution method, the entity resolution method comprising:

[0011] Obtaining a data set based on an entity resolution task; each sample data in the data set includes two entities from different sources and corresponding similar results;

[0012] Generate corresponding sentences from the two entities in each sample data in the dataset based on grammatical knowledge;

[0013] Use the generated sentence dataset to train the BERT-Pair-Networks network to obtain an entity parsing model;

[0014] The entity resolution model is used to perform entity resolution on the entity to be resolved.

[0015] Optionally, generating corresponding sentences from two entities in each sample data in the data set according to grammatical knowledge specifically includes:

[0016] Using formula S i =f g (r i1 , r i2 , ..., r ik , ..., r im )Entity R i Generate sentence S i ;

[0017] Using formula S j =f g (r j1 , r j2 , ..., r jk , ..., r jm )Entity R j Generate sentence S j ;

[0018] Among them, entity R i =(r i1 , ri2 , ..., r ik , ..., r im ), r ik is the kth attribute data of the i-th entity, entity R j =(r j1 ,r j2 ,…,r jk ,…,r jm ), r jk is the kth attribute data of the jth entity, m represents the number of attribute data, and f g A grammatical mapping function for generating sentences using attribute data.

[0019] Optionally, the training process of the BERT-Pair-Networks network specifically includes:

[0020] Use the pre-trained language representation model in the BERT-Pair-Networks network to train the sentence S i and sentence S j Encode and get the sentence encoding E i and sentence encoding E j ;

[0021] Use the pooling strategy in the BERT-Pair-Networks network to encode the sentence E i and sentence encoding E j Mapped to a fixed-dimensional vector U i and a fixed-dimensional vector U j ;

[0022] Use the MLP network in the BERT-Pair-Networks network to convert the fixed-dimensional vector U i and a fixed-dimensional vector U j Classification is performed to obtain a classification result; the classification result includes that two entities are similar or two entities are not similar.

[0023] Optionally, the pooling strategy is three mapping functions: average pooling, maximum pooling or CLS.

[0024] Optionally, the MLP network in the BERT-Pair-Networks network is used to convert the fixed-dimensional vector U i and a fixed-dimensional vector U j Perform classification and obtain classification results, including:

[0025] Using the formula O = softmax(W t (U i , U j ,|U i -U j|))Determine the classification result;

[0026] Among them, softmax is the classification function, W t is the training parameter, W t ∈R 3n×d , R is a set of real numbers, n represents the dimension of sentence embedding, d is the number of classification labels, for entity resolution tasks, d = 2, O∈{0,1} is the classification result, 1 means that two entities are similar, and 0 means that two entities are not similar.

[0027] Optionally, the objective function for training the BERT-Pair-Networks network is:

[0028] ||s a -s p ||+ε<||s a -s q ||;

[0029] Among them, s a 、s p 、s q is the embedding vector of the sentence, ||.|| is the sentence metric, ε is the boundary, a is the given original sentence, p is the given positive example, and q is the given negative example.

[0030] In a second aspect, the present application provides an entity resolution device, the entity resolution device comprising:

[0031] A data set acquisition module is used to acquire a data set based on an entity resolution task; each sample data in the data set includes two entities from different sources and corresponding similar results;

[0032] The sentence generation module is used to generate corresponding sentences from the two entities in each sample data in the dataset according to grammatical knowledge;

[0033] The entity resolution model determination module is used to train the BERT-Pair-Networks network using the generated sentence data set to obtain the entity resolution model;

[0034] The entity resolution module is used to perform entity resolution on the entities to be resolved using the entity resolution model.

[0035] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the entity resolution method.

[0036] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the entity resolution method when executed by a processor.

[0037] In a fifth aspect, the present application provides a computer program product, including a computer program, which implements the entity resolution method when executed by a processor.

[0038] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0039] The present application provides an entity resolution method, device, medium and product, which generates corresponding sentences from two entities in each sample data in a data set according to grammatical knowledge, and effectively integrates attribute data from the perspective of sentences by generating sentences, so as to have a more accurate semantic understanding of the attribute data; the entity resolution model obtained by BERT-pair-Networks training, and then the entity resolution task is completed based on BERT-pair-Networks, and the advantages of the pre-trained language representation model (Bidirectional Encoder Representations from Transformers, BERT) in natural language are utilized to achieve strong generalization ability and high scalability of the entity resolution model. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 This is an application environment diagram of an entity resolution method in one embodiment of the present application;

[0042] Figure 2 A flowchart of an entity resolution method provided in one embodiment of the present application;

[0043] Figure 3 Schematic diagram of the BERT-Pair-Networks network structure. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions 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 ordinary technicians in this field without creative work are within the scope of protection of this application.

[0045] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0046] The entity resolution method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the entity to be parsed to the server 104. After the server 104 receives the entity to be parsed, for the entity to be parsed, the server 104 obtains a data set based on the entity parsing task; generates corresponding sentences from the two entities in each sample data in the data set according to grammatical knowledge; trains the BERT-Pair-Networks network with the data set of the generated sentences to obtain an entity parsing model; and uses the entity parsing model to perform entity parsing on the entity to be parsed. The server 104 can feedback the obtained entity parsing result to the terminal 102. In addition, in some embodiments, the entity parsing method can also be implemented separately by the server 104 or the terminal 102, such as the terminal 102 can directly perform entity parsing on the entity to be parsed, or the server 104 can obtain the entity to be parsed from the data storage system and perform entity parsing on the entity to be parsed.

[0047] In an exemplary embodiment, Figure 2 As shown, an entity resolution method is provided, which is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate, including the following S201 to S204. Among them:

[0048] S201, obtaining a data set based on an entity resolution task; each sample data (R i ,R j ,T) includes two entities from different sources (R i and R j ) and the corresponding similar result T;

[0049]

[0050] In a specific embodiment, a training data set is constructed based on a device management business of a certain unit as shown in Table 1, and attribute data of device entities in the training data set is shown in Table 2.

[0051] Table 1

[0052] Entity Type Entity Type Are the two entities similar? Circuit Detector Test table 1 Test table Testing bench 0 Testing bench Circuit Detector 0 …… …… ……

[0053] Table 2

[0054]

[0055] S202, generating corresponding sentences from two entities in each sample data in the data set according to grammatical knowledge;

[0056] S203, using the dataset of generated sentences to train the BERT-Pair-Networks network to obtain an entity parsing model;

[0057] BERT-Pair-Networks is a BERT-based model that uses paired sentence embeddings to handle sentence pair regression tasks such as Semantic Textual Similarity (STS).

[0058] S204, performing entity resolution on the entity to be resolved using the entity resolution model.

[0059] In order to understand the whole sentence in entity resolution, S202 starts from the sentence perspective and effectively integrates attribute data by generating sentences, wherein S202 specifically includes:

[0060] Using formula S i =f g (r i1 , r i2 , ..., r ik , ..., r im )Entity R i Generate sentence S i ;

[0061] Using formula S j =f g (r j1 , r j2 , ..., r jk , ..., r jm )Entity R j Generate sentence S j ;

[0062] Among them, entity R i =(r i1 , r i2 , ..., r ik , ..., r im ), r ik is the kth attribute data of the i-th entity, entity R j =(r j1 , r j2, ..., r jk , ..., r jm ), r jk is the kth attribute data of the jth entity, m represents the number of attribute data, and f g A grammatical mapping function for generating sentences using attribute data.

[0063] In a specific embodiment, in combination with grammatical knowledge, the grammatical structure of the sentence generated based on the attributes is:

[0064] [name] for [function], model [model], price [price] yuan;

[0065] According to the above grammatical structure specifications, each record in the training set is mapped into a sentence, as shown in Table 3.

[0066] Table 3

[0067]

[0068] The training based on BERT-Pair-Networks network training is completed using the dataset shown in Table 3.

[0069] like Figure 3 As shown in the figure, the BERT-pair-Networks network uses the advantages of BERT in natural language to achieve strong generalization and high scalability of the entity resolution model. The training process of the BERT-Pair-Networks network specifically includes:

[0070] S1, using the pre-trained language representation model in the BERT-Pair-Networks network to respectively i and sentence S j Encode and get the sentence encoding E i and sentence encoding E j ;

[0071] Use the BERT model to train two sentences S i , S j Encode them separately, each sentence consists of characters, where c ik Represents sentence S i The kth character in embedding Represents the encoding layer of the BERT model, which performs vector mapping on each character in the sentence.

[0072] S i ={c i1 , c i2 ,...},S j ={c j1 , c j2 , ...};

[0073] E i =f embedding (S i );

[0074] E j =f embedding (S j );

[0075] S2, using the pooling strategy in the BERT-Pair-Networks network to encode the sentence E i and sentence encoding E j Mapped to a fixed-dimensional vector U i and a fixed-dimensional vector U j ;

[0076] U i =f pooling (E i );U j =f pooling (E j );

[0077] Among them, the pooling strategy f pooling It can be three mapping functions: average pooling, maximum pooling or CLS. Average pooling means calculating the average over all dimensions, maximum pooling means calculating the maximum value over each dimension, and CLS means taking the vector of the first character of the sentence.

[0078] S3, using the MLP network in the BERT-Pair-Networks network to convert the fixed-dimensional vector U i and a fixed-dimensional vector U j Classification is performed to obtain a classification result; the classification result includes that two entities are similar or two entities are not similar.

[0079] The two sentence vectors U i , U j and element difference |U i -U j |Use MLP network to perform softmax classification, the formula is:

[0080] O = softmax(W t (U i , U j ,|U i -U j |));

[0081] Among them, softmax is the classification function, W t is the training parameter, W t ∈R 3n×d, R is a set of real numbers, n represents the dimension of sentence embedding, d is the number of classification labels, for entity resolution tasks, d = 2, O∈{0,1} is the classification result, 1 means that two entities are similar, and 0 means that two entities are not similar.

[0082] The objective function for training BERT-Pair-Networks is determined as follows:

[0083] Given the original sentence a, the positive sentence p, and the negative example q, in order to reduce the distance between a and p and increase the distance between a and q, the objective function is to minimize the loss of the following formula:

[0084] max(||s a -s p ||-||s a -s q ||+s,0);

[0085] To ensure p At least better than s q Closer to a , then the objective function is:

[0086] ||s a -s p ||+ε<||s a -s q ||;

[0087] Among them, s a 、s p 、s q is the embedding vector of the sentence, ||.|| is the sentence metric, and ε is the boundary.

[0088] This application is based on the entity parsing model of sentences. All the attributes and semantic structure information of the entity are input into BERT as a new sentence, and BERT is used as an encoder to represent the sentence more deeply. BERT is trained in an unsupervised manner on a large text corpus, performs well in natural language processing, and is good at processing sentence pair tasks. Therefore, this application not only takes advantage of the strong generalization ability in semantic understanding, but also plays a high scalability in multiple fields. The entity parsing model provided in this application can quickly adapt to new fields and tasks with only moderate fine-tuning.

[0089] Based on the same inventive concept, the embodiment of the present application also provides an entity resolution device for implementing the entity resolution method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more entity resolution device embodiments provided below can refer to the limitations on the entity resolution method above, and will not be repeated here.

[0090] In an exemplary embodiment, an entity resolution device is provided, comprising:

[0091] A data set acquisition module is used to acquire a data set based on an entity resolution task; each sample data in the data set includes two entities from different sources and corresponding similar results;

[0092] The sentence generation module is used to generate corresponding sentences from the two entities in each sample data in the dataset according to grammatical knowledge;

[0093] The entity resolution model determination module is used to train the BERT-Pair-Networks network using the generated sentence data set to obtain the entity resolution model;

[0094] The entity resolution module is used to perform entity resolution on the entities to be resolved using the entity resolution model.

[0095] In an exemplary embodiment, a computer device is provided, which may be a server or a terminal. The computer device includes a processor, a memory, an input / output interface (I / O for short) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, an entity resolution method is implemented.

[0096] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0097] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0098] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0099] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0100] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0101] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0102] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. An entity resolution method, characterized in that: The entity resolution method comprises: Obtaining a data set based on an entity resolution task; each sample data in the data set includes two entities from different sources and corresponding similar results; Generate corresponding sentences from the two entities in each sample data in the dataset based on grammatical knowledge; Use the generated sentence dataset to train the BERT-Pair-Networks network to obtain an entity parsing model; The entity resolution model is used to perform entity resolution on the entity to be resolved.

2. The entity resolution method according to claim 1, characterized in that: The method of generating corresponding sentences from two entities in each sample data in the data set according to grammatical knowledge specifically includes: Using formula S i =f g (r i1 , r i2 , ..., r ik , ..., r im )Entity R i Generate sentence S i ; Using formula S j =f g (r j1 , r j2 , ..., r jk ,...r jm )Entity R j Generate sentence S j ; Among them, entity R i =(r i1 ,r i2 ,...,r ik ,...,r im ), r ik is the kth attribute data of the i-th entity, entity R j =(r j1 ,r j2 ,...,r jk ,...,r jm ), r jk is the kth attribute data of the jth entity, m represents the number of attribute data, and f g A grammatical mapping function for generating sentences using attribute data.

3. The entity resolution method according to claim 2, characterized in that: The training process of the BERT-Pair-Networks network specifically includes: Use the pre-trained language representation model in the BERT-Pair-Networks network to train the sentence S i and sentence S j Encode and get the sentence encoding E i and sentence encoding E j ; Use the pooling strategy in the BERT-Pair-Networks network to encode the sentence E i and sentence encoding E j Mapped to a fixed-dimensional vector U i and a fixed-dimensional vector U j ; Use the MLP network in the BERT-Pair-Networks network to convert the fixed-dimensional vector U i and a fixed-dimensional vector U j Classification is performed to obtain a classification result; the classification result includes that two entities are similar or two entities are not similar.

4. The entity resolution method according to claim 3, characterized in that: The pooling strategy includes three mapping functions: average pooling, maximum pooling or CLS.

5. The entity resolution method according to claim 3, characterized in that: The MLP network in the BERT-Pair-Networks network is used to convert the fixed-dimensional vector U i and a fixed-dimensional vector U j Perform classification and obtain classification results, including: Using the formula O = softmax(W t (U i ,U j ,|U i -U j |))Determine the classification result; Among them, softmax is the classification function, W t is the training parameter, W t ∈R 3n×d , R is a set of real numbers, n represents the dimension of sentence embedding, d is the number of classification labels, for entity resolution tasks, d = 2, O∈{0,1} is the classification result, 1 means that two entities are similar, and 0 means that two entities are not similar.

6. The entity resolution method according to claim 3, characterized in that: The objective function for training the BERT-Pair-Networks network is: ||s a -s p ||+ε<||s a -s q ||; Among them, s a 、s p 、s q is the embedding vector of the sentence, ||.|| is the sentence metric, ε is the boundary, a is the given original sentence, p is the given positive example, and q is the given negative example.

7. An entity resolution device, characterized in that: The entity resolution device comprises: A data set acquisition module is used to acquire a data set based on an entity resolution task; each sample data in the data set includes two entities from different sources and corresponding similar results; The sentence generation module is used to generate corresponding sentences from the two entities in each sample data in the dataset according to grammatical knowledge; The entity resolution model determination module is used to train the BERT-Pair-Networks network using the generated sentence data set to obtain the entity resolution model; The entity resolution module is used to perform entity resolution on the entities to be resolved using the entity resolution model.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the entity resolution method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the entity resolution method described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the entity resolution method described in any one of claims 1 to 6 is implemented.

Images