Entity matching method and device based on research direction, equipment and storage medium

Through the combination of 2GRAM and BERT models, automatic matching and standardized storage of research direction information is achieved, the problems of low efficiency and poor accuracy caused by human judgment are solved, and the efficiency and accuracy of acquisition and storage of research direction information are improved.

CN119025668APending Publication Date: 2024-11-26STATE GRID INFORMATION & TELECOMM BRANCH +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411109959.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-13
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

In the prior art, artificial judgment leads to irregular and inconsistent expression of research direction information, low efficiency and low accuracy, making it difficult to effectively extract and store research direction information in expert information.

Method used

The pre-configured 2GRAM model and BERT model are used to segment and text vectorize the matching research direction words and academic topic words. Through similarity comparison and vector similarity calculation, the target academic topic words are automatically determined to realize research direction information. standardized storage.

Benefits of technology

It improves the efficiency and accuracy of information acquisition in research directions, reduces the dependence of manual judgments, and improves the efficiency and accuracy of standardized storage.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119025668A_ABST
    Figure CN119025668A_ABST
Patent Text Reader

Abstract

The invention discloses an entity matching method and device based on a research direction, equipment and a storage medium. The method comprises the following steps: obtaining a to-be-matched research direction word; performing segmentation processing on the to-be-matched research direction word and the academic subject word in the academic subject word library by adopting a 2GRAM model to obtain a first double-pair sequence corresponding to the to-be-matched research direction word and a second double-word pair sequence corresponding to the academic subject word; performing similarity comparison on each second double-pair sequence and the first double-pair sequence to determine a plurality of academic subject terms to be matched from an academic subject term library; and sequentially carrying out text vectorization processing on the to-be-matched research direction words and the to-be-matched academic subject words by adopting a BERT model, calculating the similarity between the obtained to-be-matched academic subject vectors and the to-be-matched research direction vectors, and determining one to-be-matched academic subject word as a target academic subject word. According to the method, the entity matching efficiency in the research direction is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of natural language processing technology, and in particular to an entity matching method, apparatus, device and storage medium based on a research direction. Background Art

[0002] In the process of building an expert information database using expert information from public resources, different experts may have different ways of expressing the same research direction. When extracting entities from the research direction information in the expert information, we may face the problem of non-standard and non-uniform expression of the research direction information, resulting in a wide variety of extracted research direction words.

[0003] In the prior art, users are usually required to search for research direction words extracted from entities, and after manual judgment, unify the research direction words with different expressions into standardized expressions, so as to store the standardized expressions as research direction words extracted from expert information in a unified format. This method is not only inefficient, but also has low accuracy in manual judgment. Summary of the invention

[0004] Based on the above problems, the present application provides a research direction-based entity matching method, apparatus, device and storage medium, the purpose of which is to improve the efficiency and accuracy of obtaining research direction information.

[0005] The embodiments of the present application disclose the following technical solutions:

[0006] In a first aspect, the present application provides an entity matching method based on a research direction, including:

[0007] Obtain the research direction words to be matched from the entity extraction results of the expert information;

[0008] The pre-configured 2GRAM model is used to segment the research direction words to be matched and each academic subject word in the academic subject word library to obtain a first pair sequence corresponding to the research direction words to be matched and a second pair sequence corresponding to each academic subject word;

[0009] According to the second two-word pair sequence corresponding to each academic subject word, a similarity comparison is performed with the first two-word pair sequence corresponding to the research direction word to be matched, and a plurality of academic subject words to be matched are determined from the academic subject word library;

[0010] Using a preconfigured BERT model to perform text vectorization processing on the research direction words to be matched, to obtain a research direction vector to be matched, and performing text vectorization processing on the academic subject words to be matched, to obtain an academic subject vector to be matched;

[0011] By calculating the similarity between each of the academic subject vectors to be matched and the research direction vector to be matched, an academic subject word to be matched is determined from the multiple academic subject words to be matched as the target academic subject word corresponding to the research direction word to be matched.

[0012] Optionally, as in the above method, the preconfigured 2GRAM model is used to segment the research direction words to be matched and each academic subject word in the academic subject word library to obtain a first pair of sequences corresponding to the research direction words to be matched and a second pair of sequences corresponding to each academic subject word, including:

[0013] Input the research direction words to be matched and each academic subject word in the academic subject word library into the preconfigured 2GRAM model;

[0014] Through the sliding window in the preconfigured 2GRAM model, traverse each character in the research direction word to be matched, obtain multiple first double-word pairs and generate a first double-pair sequence corresponding to the research direction word to be matched;

[0015] Through the sliding window in the preconfigured 2GRAM model, each character in each academic subject word is traversed to obtain multiple second bigram pairs corresponding to each academic subject word and generate a second bigram pair sequence corresponding to each academic subject word.

[0016] Optionally, as in the above method, the second two-word pair sequence corresponding to each of the academic subject words is respectively compared with the first two-word pair sequence corresponding to the research direction word to be matched, and a plurality of academic subject words to be matched are determined from the academic subject word library, including:

[0017] Calculate the Jaccard distance between the second bigram pair sequence corresponding to each academic subject word and the first bigram pair sequence corresponding to the research direction word to be matched, and obtain the similarity between each academic subject word and the research direction word to be matched;

[0018] A plurality of academic subject words corresponding to similarities greater than or equal to a preset threshold are selected from the academic subject word library as academic subject words to be matched.

[0019] Optionally, as in the above method, the Jaccard distance between the second bigram pair sequence corresponding to each academic subject word and the first bigram pair sequence corresponding to the research direction word to be matched is calculated to obtain the similarity between each academic subject word and the research direction word to be matched, including:

[0020] Calculate the intersection and union between the second bigram pair sequence corresponding to each academic subject word and the first bigram pair sequence corresponding to the research direction word to be matched;

[0021] Calculate the Jaccard distance between each academic subject word and the research direction word to be matched according to the intersection and the union;

[0022] The Jaccard distance between each academic subject word and the research direction word to be matched is converted into similarity.

[0023] Optionally, as in the above method, the preconfigured BERT model is used to perform text vectorization processing on the research direction words to be matched to obtain the research direction vector to be matched, and the academic subject words to be matched are performed text vectorization processing to obtain the academic subject vector to be matched, including:

[0024] Using the word segmenter in the preconfigured BERT model to decompose the research direction words to be matched into a plurality of first basic units, and decompose the academic subject words to be matched into a plurality of second basic units;

[0025] Inputting the first basic unit into an encoder in a BERT model for text vectorization processing to generate a vector representation corresponding to the first basic unit;

[0026] Inputting the second basic unit into an encoder in a BERT model for text vectorization processing to generate a vector representation corresponding to the second basic unit;

[0027] The vector representations corresponding to the multiple first basic units in the research direction words to be matched are merged to obtain the research direction vector to be matched corresponding to the research direction words to be matched, and the vector representations corresponding to the multiple second basic units in the academic subject words to be matched are merged to obtain the academic subject vector to be matched corresponding to the academic subject words to be matched.

[0028] Optionally, as in the above method, by calculating the similarity between each of the academic subject vectors to be matched and the research direction vector to be matched, determining an academic subject word to be matched from the multiple academic subject words to be matched as the target academic subject word corresponding to the research direction word to be matched, comprises:

[0029] The cosine distance is used to calculate the vector similarity between each of the academic subject vectors to be matched and the research direction vector to be matched, so as to obtain the matching similarity scores corresponding to each of the academic subject words to be matched;

[0030] The academic subject word with the highest matching similarity score is selected from the multiple academic subject words to be matched as the target academic subject word corresponding to the research direction word to be matched.

[0031] Optionally, the method as described above further includes:

[0032] The research direction words to be matched, the target academic subject words and the matching similarity scores corresponding to the target academic subject words are displayed as output results in the form of structured data.

[0033] In a second aspect, the present application provides an entity matching device based on a research direction, comprising:

[0034] An acquisition module is used to obtain the research direction words to be matched from the entity extraction results of the expert information;

[0035] A segmentation module is used for segmenting the research direction words to be matched and each academic subject word in the academic subject word library by using a preconfigured 2GRAM model to obtain a first pair sequence corresponding to the research direction words to be matched and a second pair sequence corresponding to each academic subject word;

[0036] A similarity comparison module is used to compare the similarity of the second two-word pair sequence corresponding to each academic subject word with the first two-word pair sequence corresponding to the research direction word to be matched, and determine a plurality of academic subject words to be matched from the academic subject word library;

[0037] A vectorization processing module, used for performing text vectorization processing on the research direction words to be matched using a preconfigured BERT model to obtain a research direction vector to be matched, and performing text vectorization processing on the academic subject words to be matched to obtain an academic subject vector to be matched;

[0038] The determination module is used to determine an academic subject word to be matched from the multiple academic subject words to be matched as the target academic subject word corresponding to the research direction word to be matched by calculating the similarity between each academic subject vector to be matched and the research direction vector to be matched.

[0039] In a third aspect, the present application provides an electronic device, the device comprising: a processor, and a memory communicatively connected to the processor;

[0040] The memory stores computer-executable instructions;

[0041] The processor executes the computer-executable instructions stored in the memory to implement the entity matching method based on research direction described in any of the above embodiments.

[0042] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer execution instructions, and when the computer execution instructions are executed by a processor, they are used to implement the entity matching method based on research direction described in any of the above embodiments.

[0043] This application uses a pre-configured 2GRAM model to segment the acquired research direction words to be matched and each academic subject word in the academic subject word library, so as to obtain the two-word pair sequence corresponding to the research direction words to be matched and each academic subject word; the two-word pair sequence corresponding to the research direction words to be matched and each academic subject word is compared for similarity, and at least one academic subject word to be matched is determined, thereby screening out at least one academic subject word with a high similarity to the research direction words to be matched. In addition, a pre-configured BERT model is used to perform text vectorization processing on the research direction words to be matched and the academic subject words to be matched, and vector similarity calculation is performed through the semantic representation of the obtained research direction words to be matched and the academic subject words to be matched in the vector space, thereby determining the target academic subject words, realizing fast entity matching processing, and improving the efficiency and accuracy of standardized storage of research directions. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] 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 or the description of the prior art 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 labor.

[0045] Figure 1 A schematic diagram of an embodiment of a method for entity matching based on a research direction provided in an embodiment of the present application;

[0046] Figure 2 A flowchart of another embodiment of an entity matching method based on a research direction provided in an embodiment of the present application;

[0047] Figure 3 A flowchart of another embodiment of an entity matching method based on a research direction provided in an embodiment of the present application;

[0048] Figure 4 A flowchart of another embodiment of an entity matching method based on a research direction provided in an embodiment of the present application;

[0049] Figure 5 A flowchart of another embodiment of an entity matching method based on a research direction provided in an embodiment of the present application;

[0050] Figure 6 A schematic diagram of the structure of an embodiment of an entity matching device based on a research direction provided by the present application;

[0051] Figure 7 A schematic structural diagram of an embodiment of an electronic device provided in the present application. DETAILED DESCRIPTION

[0052] As described above, in the current method, standardized expressions require users to search for research direction words extracted from entities, and after manual judgment, research direction words with different expressions are uniformly expressed as standardized expressions.

[0053] After research, the inventor provides an entity matching method, device, equipment and storage medium based on research direction to solve the technical problems in the prior art that research direction words with different expressions are artificially unified into standardized expression words, thereby reducing the efficiency of research direction standardization and the accuracy of artificially judged standardized expression words is low.

[0054] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0055] See also Figure 1 , which is a flow chart of an embodiment of an entity matching method based on a research direction provided in an embodiment of the present application. The method includes:

[0056] S101: Obtain the research direction words to be matched from the entity extraction results of the expert information.

[0057] The entity extraction result of the expert information is the relevant information of the expert in the academic field obtained from public resources through entity extraction processing. In this embodiment, the research direction word to be matched is the research direction corresponding to the expert to be entity matched.

[0058] S102: Use the pre-configured 2GRAM model to segment the research direction word to be matched and each academic subject word in the academic subject word library to obtain a first pair sequence corresponding to the research direction word to be matched and a second pair sequence corresponding to each academic subject word.

[0059] Among them, the academic thesaurus contains all standard academic thesaurus terms.

[0060] In the present embodiment, the academic subject words in the academic subject word library are used as the standardized vocabulary to carry out entity matching with the research direction words to be matched. Through the sliding window in the 2GRAM model, each character in the research direction words to be matched is traversed to segment the research direction words to be matched according to the sliding window size to obtain the double-word pair sequence corresponding to the research direction words to be matched; each character in each academic subject word is traversed to segment each academic subject word according to the sliding window size to obtain the double-word pair sequence corresponding to each academic subject word.

[0061] For ease of explanation and understanding, this solution distinguishes the two types of binary pair sequences formed after the above segmentation. The binary pair sequence corresponding to the research direction word to be matched is called the first binary pair sequence, and the binary pair sequence corresponding to each academic subject word is called the second binary pair sequence.

[0062] For example, if the research direction word to be matched is "weather change and prediction", the corresponding first two-word pair sequence obtained after segmentation is ["weather", "phenomenon change", "change", "transformation and", "and prediction", "prediction"]; if the academic subject word is "weather prediction", the corresponding second two-word pair sequence obtained after segmentation is ["weather", "phenomenon prediction", "prediction"].

[0063] S103: Perform similarity comparison on the second two-word pair sequence corresponding to each academic subject word and the first two-word pair sequence corresponding to the research direction word to be matched, and determine multiple academic subject words to be matched from the academic subject word library.

[0064] In this embodiment, for each academic subject word, after calculating the similarity and comparing the second two-word pair sequence corresponding to the academic subject word with the first two-word pair sequence corresponding to the research direction word to be matched, multiple academic subject words with higher similarity in the comparison results are screened out from the academic subject word library as academic subject words to be matched. Specifically, by presetting a similarity threshold, academic subject words with a similarity greater than or equal to the similarity threshold are screened out as academic subject words to be matched for further matching.

[0065] S104: Use the preconfigured BERT model to perform text vectorization processing on the research direction words to be matched, so as to obtain the research direction vector to be matched, and perform text vectorization processing on the academic subject words to be matched, so as to obtain the academic subject vector to be matched.

[0066] Among them, the BERT model is a pre-trained language model. In this embodiment, the pre-configured BERT model (Bidirectional Encoder Representations from Transformers, BERT) is used to decompose the research direction words to be matched and the academic subject words to be matched, generate basic units that can be encoded, and perform text vectorization and merging processing on the decomposed basic units corresponding to each academic subject word to be matched and the research direction words to be matched, so as to obtain the research direction vector to be matched corresponding to the research direction words to be matched and the academic subject vector to be matched corresponding to the academic subject words to be matched.

[0067] Specifically, the English dataset can be processed by the bert-base-uncase model for text vectorization, while the Chinese dataset can be processed by the Chinese-bert-wwm model for text vectorization, thereby adapting to the corresponding language features.

[0068] S105: By calculating the similarity between each academic subject vector to be matched and the research direction vector to be matched, an academic subject word to be matched is determined from multiple academic subject words to be matched as a target academic subject word corresponding to the research direction word to be matched.

[0069] In this embodiment, the similarity between each academic subject vector to be matched and the research direction vector to be matched is calculated, and all academic subject vectors to be matched are sorted to obtain an academic subject word to be matched with the highest similarity from multiple academic subject words to be matched as the target academic subject word corresponding to the research direction word to be matched. Among them, vector similarity calculation is usually performed by calculating the distance between two vectors. The closer the distance between the two vectors, the more similar the two vectors are, including Euclidean distance, Manhattan distance, Chebyshev distance, Minkowski distance, standardized Euclidean distance, Lang's distance and Mahalanobis distance, etc.

[0070] For example, the academic subject term to be matched is acidic fibroblast growth factor, and the target academic subject term to be matched is human fibroblast growth factor. Both are fibroblast growth factors. Among them, "acidic" and "human" may just refer to different subtypes of growth factors. The academic subject term to be matched is upper gastrointestinal bleeding, and the target academic subject term to be matched is upper gastrointestinal tract. Both phrases are related to the upper gastrointestinal tract. The academic subject term to be matched is geographic information service, and the target academic subject term to be matched is geographic information system. These two terms are closely related. A system (GIS) can provide multiple services, including but not limited to geographic information services. The academic subject term to be matched is situ stress measurement, and the target academic subject term to be matched is residual stress measurement. Both stress measurements are important concepts in materials science and engineering, and both involve stress measurement.

[0071] In this embodiment, the obtained research direction words to be matched and each academic subject word in the academic subject word library are segmented respectively by a pre-configured 2GRAM model to obtain the research direction words to be matched and the two-word pair sequence corresponding to each academic subject word; the two-word pair sequence corresponding to the research direction words to be matched and each academic subject word is compared for similarity, and at least one academic subject word to be matched is determined, thereby screening out at least one academic subject word with a high similarity to the research direction words to be matched. At the same time, the pre-configured BERT model is used to perform text vectorization processing on the research direction words to be matched and the academic subject words to be matched, and the vector similarity calculation is performed through the semantic representation of the obtained research direction words to be matched and the academic subject words to be matched in the vector space, and then the target academic subject words are determined, and fast entity matching processing is realized, so that the obtained target academic subject words are stored in a unified format as the standardized research direction in the expert information, thereby improving the efficiency and accuracy of the standardized storage of research directions.

[0072] See also Figure 2 , which is a flow chart of another embodiment of an entity matching method based on a research direction provided in an embodiment of the present application. Figure 1 , Figure 2 A specific implementation method of text slicing using the 2GRAM model is shown. Figure 4 Introduced in S1021-S1023.

[0073] On the basis of the above embodiment, a specific implementation method of "using a pre-configured 2GRAM model to segment the research direction words to be matched and each academic subject word in the academic subject word library to obtain a first pair of sequences corresponding to the research direction words to be matched and a second pair of sequences corresponding to each academic subject word" in S102 includes:

[0074] S1021: Input the research direction words to be matched and each academic subject word in the academic subject word library into the preconfigured 2GRAM model.

[0075] S1022: Through the sliding window in the preconfigured 2GRAM model, traverse each character in the research direction word to be matched, obtain multiple first bigram pairs corresponding to the research direction word to be matched, and generate a first bigram pair sequence corresponding to the research direction word to be matched.

[0076] In this embodiment, the 2GRAM model is an algorithm based on a statistical language model, which performs a sliding window operation of size 2 on each character in the research direction word to be matched, forming a double-word pair with a length of 2, and summarizes the generated double-word pairs into the first double-pair sequence corresponding to the research direction word to be matched.

[0077] S1023: Through the sliding window in the preconfigured 2GRAM model, traverse each character in each academic subject word to obtain multiple second bigram pairs corresponding to each academic subject word and generate a second bigram pair sequence corresponding to each academic subject word.

[0078] In this embodiment, for each academic keyword, each character in the academic keyword is subjected to a sliding window operation of size 2 to form a double-word pair of length 2, and the generated double-word pairs are summarized into the first double-pair sequence corresponding to the academic keyword.

[0079] Combination Figure 2 As shown, before executing S1021-S1023, refer to the previous embodiment and execute S101 to obtain the research direction words to be matched; after executing S1021-S1023, refer to the previous embodiment and execute S103-S105 to determine the target academic subject words.

[0080] In the present embodiment, by inputting the research direction words to be matched and each academic subject word in the academic subject word library into a preconfigured 2GRAM model, using the sliding window in the 2GRAM model, traversing each character in the research direction words to be matched and each academic subject word respectively, thereby obtaining the first pair sequence corresponding to the research direction words to be matched and the second pair sequence corresponding to each academic subject word. Compared with the prior art, the method of the present application uses the 2GRAM model to slice the research direction words to be matched and the academic subject words, and divides the text into two pairs of words, thereby retaining the association between each character and the adjacent characters, and improving the accuracy of word similarity calculation.

[0081] See also Figure 3 , which is a flow chart of another embodiment of an entity matching method based on research direction provided by an embodiment of the present application. Based on the above embodiment, a specific implementation method of "comparing the second two-word pair sequence corresponding to each academic subject word with the first two-word pair sequence corresponding to the research direction word to be matched, and determining multiple academic subject words to be matched from the academic subject word library" in S103 includes:

[0082] S1031: Calculate the Jaccard distance between the second bigram pair sequence corresponding to each academic subject word and the first bigram pair sequence corresponding to the research direction word to be matched, and obtain the similarity between each academic subject word and the research direction word to be matched.

[0083] Among them, Jaccard distance measures the difference between two sets by calculating the ratio of different elements in the two sets. Jaccard distance is defined as the size of the union minus the size of the intersection divided by the size of the union. In the technical solution of this application, this indicator is used to compare the difference between two twin pair sequences, and then measure the similarity between academic subject words and research direction words to be matched.

[0084] S1032: Select from the academic thesaurus a plurality of academic thesaurus words corresponding to a similarity greater than or equal to a preset threshold as academic thesaurus words to be matched.

[0085] In this embodiment, the preset threshold is a preset similarity. The preset threshold is compared with the academic subject words in each academic subject word library, and multiple academic subject words with similarity greater than or equal to the preset threshold are screened out as academic subject words to be matched, so as to be used for subsequent further similarity calculation and screening, and determine the only target academic subject word.

[0086] Combination Figure 3As shown, before executing S1031-S1032, refer to the previous embodiment and execute S101-S102 to achieve the segmentation of the research direction words to be matched; after executing S1031-S1032, refer to the previous embodiment and execute S104-S105 to achieve the determination of the target academic subject words.

[0087] In this embodiment, by calculating the Jaccard distance between the second two-word pair sequence corresponding to each academic subject word and the first two-word pair sequence corresponding to the research direction word to be matched, the similarity between each academic subject word and the research direction word to be matched is obtained, so as to select multiple academic subject words with semantic similarity greater than or equal to a preset threshold value in the academic subject word library as the academic subject words to be matched, for subsequent further matching. Compared with the prior art, the method of the present application calculates the Jaccard distance between each academic subject word and the research direction word to be matched to obtain the similarity. The user selects the academic subject words to be matched that meet the conditions, and quickly screens out the academic subject words with high similarity to the research direction words to be matched in the academic subject word library, thereby reducing the workload of manual screening and improving work efficiency.

[0088] Further, based on the above embodiment, a specific implementation method of "calculating the Jaccard distance between the second bigram pair sequence corresponding to each academic subject word and the first bigram pair sequence corresponding to the research direction word to be matched, and obtaining the similarity between each academic subject word and the research direction word to be matched" in S1031 includes:

[0089] Calculate the intersection and union between the second bigram pair sequence corresponding to each academic subject word and the first bigram pair sequence corresponding to the research direction word to be matched.

[0090] In the present embodiment, each bigram pair of the second bigram pair sequence in an academic subject word is traversed, and the bigram pair is checked whether it also exists in the first bigram pair sequence corresponding to the research direction word to be matched. If a bigram pair appears in both sequences, it is added to the intersection result. After completing the traversal, the intersection result is the set composed of all bigram pairs in the two sequences.

[0091] Create an empty set to store the union set; after adding each bigram pair of the second bigram pair sequence in an academic subject word to the union set, traverse each bigram pair in the first bigram pair sequence corresponding to the research direction word to be matched, and check whether the bigram pair already exists in the union set. If not, add it to the union set until the traversal of the first bigram pair sequence is completed, and the union result is a set consisting of all bigram pairs that are not repeated in the two sequences.

[0092] The Jaccard distance between each academic subject word and the research direction word to be matched is calculated based on the intersection and union.

[0093] In this embodiment, the Jaccard distance is calculated using the following formula based on the intersection and union of the second bigram pair sequence corresponding to each academic subject word and the first bigram pair sequence corresponding to the research direction word to be matched:

[0094]

[0095] Among them, |A∩B| is the number of elements in the intersection; |A∪B| is the number of elements in the union; A and B respectively represent the second bigram pair sequence corresponding to the academic subject words and the first bigram pair sequence corresponding to the research direction words to be matched.

[0096] The Jaccard distance between each academic subject word and the research direction word to be matched is converted into similarity.

[0097] In this embodiment, since the Jaccard distance and the similarity are inversely proportional, the Jaccard distance can be converted into similarity through simple mathematical transformation. Among them, a common method is to directly use 1-J δ (A, B) is used to represent the similarity.

[0098] In this embodiment, the intersection and union between the second two-word pair sequence corresponding to each academic subject word and the first two-word pair sequence corresponding to the research direction word to be matched are calculated; the Jaccard distance between each academic subject word and the research direction word to be matched is calculated based on the intersection and union. Compared with the prior art, the method of the present application can more accurately reflect the similarity between each academic subject word and the research direction word to be matched by comparing the intersection and union size of the second two-word pair sequence of each academic subject word with the first two-word pair sequence corresponding to the research direction word to be matched, reduce the false matching caused by similar words but different meanings, and improve the accuracy of matching.

[0099] See also Figure 4 , which is a flow chart of another embodiment of an entity matching method based on research direction provided in an embodiment of the present application. Based on the above embodiment, a specific implementation method of "using a pre-configured BERT model to perform text vectorization processing on the research direction words to be matched to obtain the research direction vector to be matched, and performing text vectorization processing on the academic subject words to be matched to obtain the academic subject vector to be matched" in S104 includes:

[0100] S1041: Use the word segmenter in the preconfigured BERT model to decompose the research direction words to be matched into multiple first basic units, and decompose the academic subject words to be matched into multiple second basic units.

[0101] Among them, the tokenizer in the BERT (Bidirectional Encoder Representations from Transformers) model is the first step in text processing. Its main function is to decompose the input text into a series of basic units (tokens).

[0102] In this embodiment, the research direction words to be matched and the academic subject words to be matched are respectively passed as input to the word segmenter of the BERT model; the word segmenter decomposes the input research direction words to be matched into a series of first basic units, and decomposes the input academic subject words to be matched into a series of second basic units according to the trained vocabulary and word segmentation rules in the pre-configured BERT model.

[0103] S1042: Input the first basic unit into the encoder in the BERT model for text vectorization processing to generate a vector representation corresponding to the first basic unit.

[0104] Among them, the encoder (Transformer) is composed of multiple encoder layers stacked together.

[0105] In this embodiment, each first basic unit is processed by multiple encoder layers to convert each first basic unit into a vector representation of a fixed dimension.

[0106] S1043: Input the second basic unit into the encoder in the BERT model for text vectorization processing to generate a vector representation corresponding to the second basic unit.

[0107] In this embodiment, each first basic unit is processed by multiple encoder layers to convert each first basic unit into a vector representation of a fixed dimension.

[0108] In the above steps, S1042 and S1043 may be executed one after the other or simultaneously. There is no limitation on the execution order of these two steps.

[0109] S1044: Merge the vector representations corresponding to the multiple first basic units in the research direction words to be matched to obtain the research direction vector to be matched corresponding to the research direction words to be matched, and merge the vector representations corresponding to the multiple second basic units in the academic subject words to be matched to obtain the academic subject vector to be matched corresponding to the academic subject words to be matched.

[0110] In this embodiment, the vector representations corresponding to multiple first basic units in the research direction words to be matched are merged according to the merging strategy in the preconfigured BERT model to obtain the research direction vector to be matched corresponding to the research direction words to be matched; and the same method is used to obtain the academic subject vector to be matched corresponding to the academic subject words to be matched.

[0111] The merging strategy may include but is not limited to one or more of average pooling, maximum pooling, weighted pooling, and attention mechanism.

[0112] Combination Figure 4 As shown, before executing S1041-S1044, refer to the previous embodiment and execute S101-S103 to obtain the academic subject words to be matched; after executing S1041-S1044, refer to the previous embodiment and execute S105 to determine the target academic subject words.

[0113] In this embodiment, the research direction words to be matched and the academic subject words to be matched are decomposed into corresponding multiple first basic units and multiple second basic units by the word segmenter in the BERT model, and the text vectorization processing is performed by the encoder of the model to generate the vector representation corresponding to the first basic unit and the vector representation corresponding to the second basic unit; the vector representations corresponding to the first basic unit are merged to obtain the research direction vector to be matched, and the vector representations corresponding to the first basic unit are merged to obtain the academic subject vector to be matched. Compared with the prior art, the method of the present application performs text vectorization processing through the BERT model, which can simultaneously consider the context information of each character in the research direction words to be matched and the academic subject words to be matched, so as to more accurately understand their semantics and capture richer context dependencies, thereby improving the accuracy of the text representation of the research direction words to be matched and the academic subject words to be matched.

[0114] See also Figure 5 , which is a flow chart of another embodiment of an entity matching method based on research direction provided in an embodiment of the present application. Based on the above embodiment, a specific implementation method of "determining an academic subject word to be matched from multiple academic subject words to be matched as a target academic subject word corresponding to the research direction word to be matched by calculating the similarity between each academic subject vector to be matched and the research direction vector to be matched" in S105 includes:

[0115] S1051: Use cosine distance to calculate the vector similarity between each academic subject vector to be matched and the research direction vector to be matched, and obtain the matching similarity score corresponding to each academic subject word to be matched.

[0116] In this embodiment, data preprocessing is performed on all the obtained academic subject terms to be matched and their corresponding vector representations and the research direction vectors to be matched to ensure that the dimensions of all vectors are consistent so as to perform cosine similarity calculation.

[0117] Among them, cosine similarity evaluates the similarity of two vectors by calculating the cosine value of the angle between two vectors in the vector space. The corresponding expression is as follows:

[0118]

[0119] Among them, A and B represent the academic subject vector to be matched and the research direction vector to be matched respectively; ‖A‖ and ‖B‖ represent the modulus of the academic subject vector to be matched and the research direction vector to be matched respectively.

[0120] The calculated cosine value between each academic subject vector to be matched and each research direction vector to be matched is used as the matching similarity score between the academic subject word to be matched and the research direction word to be matched.

[0121] S1052: Select the academic subject word with the highest matching similarity score from multiple academic subject words to be matched as the target academic subject word corresponding to the research direction word to be matched.

[0122] In this embodiment, the matching similarity scores corresponding to all the academic subject words to be matched are compared or sorted, and the academic subject word to be matched with the highest matching similarity score is selected as the target academic subject word corresponding to the research direction word to be matched.

[0123] Combination Figure 5 As shown, after executing S1051-S1052, referring to the previous embodiment, execute S101-S104 to obtain the academic subject words to be matched and perform text vectorization processing on the academic subject words to be matched and the research direction words to be matched.

[0124] In this embodiment, the cosine distance is used to calculate the vector similarity of each academic subject vector to be matched and the research direction vector to be matched, and the matching similarity score corresponding to each academic subject word to be matched is obtained; then the academic subject word to be matched with the highest matching similarity score is selected from multiple academic subject words to be matched as the target academic subject word corresponding to the research direction word to be matched. Compared with the prior art, the method of the present application evaluates the similarity between the academic subject vector to be matched and the research direction vector to be matched by calculating the degree of similarity in direction through the cosine distance, rather than just based on the length of the vector or the specific numerical value. It is particularly effective in capturing the potential association between the academic subject word to be matched and the research direction word to be matched, and can more accurately reflect the semantic similarity between the two, thereby improving the matching accuracy of the academic subject words.

[0125] Furthermore, based on the above embodiment, the method may further include:

[0126] The research direction words to be matched, the target academic subject words and the matching similarity scores corresponding to the target academic subject words are displayed as output results in the form of structured data.

[0127] In this embodiment, the research direction words to be matched, the normalized target academic subject words obtained through entity matching processing, and the similarity matching scores between the target academic subject words and the research direction words to be matched are displayed to the user in the form of structured data as the final output results of the model.

[0128] In this embodiment, the matching similarity scores of the research direction words to be matched, the target academic subject words and the target academic subject words are displayed as output results in the form of structured data. Compared with the prior art, the method of the present application can more intuitively display the degree of connection between the obtained academic subject words and the research direction words to be matched, providing a reference basis for users.

[0129] See also Figure 6 , which is a schematic diagram of the structure of an embodiment of an entity matching device based on a research direction provided by the present application. The device 60 includes: an acquisition module 61, a segmentation module 62, a similarity comparison module 63, a vectorization processing module 64 and a determination module 65.

[0130] Wherein, the acquisition module 61 is used to obtain the research direction words to be matched from the entity extraction results of the expert information;

[0131] A segmentation module 62 is used for segmenting the research direction word to be matched and each academic subject word in the academic subject word library using a pre-configured 2GRAM model to obtain a first pair sequence corresponding to the research direction word to be matched and a second pair sequence corresponding to each academic subject word;

[0132] A similarity comparison module 63 is used to compare the similarity of the second two-word pair sequence corresponding to each academic subject word with the first two-word pair sequence corresponding to the research direction word to be matched, and determine a plurality of academic subject words to be matched from the academic subject word library;

[0133] A vectorization processing module 64 is used to perform text vectorization processing on the research direction words to be matched using a preconfigured BERT model to obtain the research direction vector to be matched, and to perform text vectorization processing on the academic subject words to be matched to obtain the academic subject vector to be matched;

[0134] The determination module 65 is used to determine an academic subject word to be matched from multiple academic subject words to be matched as the target academic subject word corresponding to the research direction word to be matched by calculating the similarity between each academic subject vector to be matched and the research direction vector to be matched.

[0135] An entity matching device based on a research direction provided in an embodiment of the present application can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar and will not be repeated here.

[0136] Further, on the basis of the above embodiments, the segmentation module 62 is specifically used to input the research direction words to be matched and each academic subject word in the academic subject word library into a preconfigured 2GRAM model; it is also used to traverse each character in the research direction words to be matched through the sliding window in the preconfigured 2GRAM model, obtain multiple first double-word pairs corresponding to the research direction words to be matched and generate a first double-pair sequence corresponding to the research direction words to be matched; traverse each character in each academic subject word through the sliding window in the preconfigured 2GRAM model, obtain multiple second double-word pairs corresponding to each academic subject word and generate a second double-word pair sequence corresponding to each academic subject word.

[0137] An entity matching device based on a research direction provided in an embodiment of the present application can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar and will not be repeated here.

[0138] Further, based on the above embodiments, the similarity comparison module 63 is specifically used to calculate the Jaccard distance between the second two-word pair sequence corresponding to each academic subject word and the first two-word pair sequence corresponding to the research direction word to be matched, so as to obtain the similarity between each academic subject word and the research direction word to be matched; and select multiple academic subject words corresponding to a similarity greater than or equal to a preset threshold from the academic subject word library as the academic subject words to be matched.

[0139] An entity matching device based on a research direction provided in an embodiment of the present application can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar and will not be repeated here.

[0140] Furthermore, based on the above embodiments, the similarity comparison module 63 is also specifically used to calculate the intersection and union between the second two-word pair sequence corresponding to each academic subject word and the first two-word pair sequence corresponding to the research direction word to be matched; calculate the Jaccard distance between each academic subject word and the research direction word to be matched based on the intersection and union; and convert the Jaccard distance between each academic subject word and the research direction word to be matched into similarity.

[0141] An entity matching device based on a research direction provided in an embodiment of the present application can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar and will not be repeated here.

[0142] Further, based on the above embodiments, the vectorization processing module 64 is specifically used to use the word segmenter in the preconfigured BERT model to decompose the research direction words to be matched into multiple first basic units, and decompose the academic subject words to be matched into multiple second basic units; input the first basic unit into the encoder in the BERT model for text vectorization processing to generate a vector representation corresponding to the first basic unit; input the second basic unit into the encoder in the BERT model for text vectorization processing to generate a vector representation corresponding to the second basic unit; merge the vector representations corresponding to the multiple first basic units in the research direction words to be matched to obtain the research direction vector to be matched corresponding to the research direction words to be matched, and merge the vector representations corresponding to the multiple second basic units in the academic subject words to be matched to obtain the academic subject vector to be matched corresponding to the academic subject words to be matched.

[0143] An entity matching device based on a research direction provided in an embodiment of the present application can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar and will not be repeated here.

[0144] Further, based on the above embodiment, the determination module 65 is also used to display the matching similarity scores corresponding to the research direction words to be matched, the target academic subject words and the target academic subject words as output results in the form of structured data.

[0145] An entity matching device based on a research direction provided in an embodiment of the present application can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar and will not be repeated here.

[0146] See also Figure 7, which is a schematic diagram of the structure of an embodiment of an electronic device provided by the present application. The electronic device 70 may include: a processor 71 and a memory 72.

[0147] The processor 71 is in communication with the memory 72 , and the memory 72 is used to store computer-executable instructions. The processor 71 is configured to execute the technical solution in any of the aforementioned method embodiments by executing the computer-executable instructions stored in the memory 72 .

[0148] Optionally, the memory 72 may be independent or integrated with the processor 71. Optionally, when the memory 72 is a device independent of the processor 71, the electronic device 70 may further include: a bus for connecting the above devices.

[0149] The electronic device is used to execute the technical solution in any of the aforementioned method embodiments, and its implementation principle and technical effect are similar and will not be repeated here.

[0150] An embodiment of the present application also provides a computer-readable storage medium, which stores computer execution instructions. When the computer execution instructions are executed by a processor, they are used to implement the method described above. The implementation principles and technical effects are similar and will not be repeated here.

[0151] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and equipment embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments. The device and equipment embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components indicated as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. A person of ordinary skill in the art can understand and implement it without paying creative labor.

[0152] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An entity matching method based on research direction, characterized in that: include: Obtain the research direction words to be matched from the entity extraction results of the expert information; The pre-configured 2GRAM model is used to segment the research direction words to be matched and each academic subject word in the academic subject word library to obtain a first pair sequence corresponding to the research direction words to be matched and a second pair sequence corresponding to each academic subject word; According to the second two-word pair sequence corresponding to each academic subject word, a similarity comparison is performed with the first two-word pair sequence corresponding to the research direction word to be matched, and a plurality of academic subject words to be matched are determined from the academic subject word library; Using a preconfigured BERT model to perform text vectorization processing on the research direction words to be matched, to obtain a research direction vector to be matched, and performing text vectorization processing on the academic subject words to be matched, to obtain an academic subject vector to be matched; By calculating the similarity between each of the academic subject vectors to be matched and the research direction vector to be matched, an academic subject word to be matched is determined from the multiple academic subject words to be matched as the target academic subject word corresponding to the research direction word to be matched.

2. The method according to claim 1, characterized in that The pre-configured 2GRAM model is used to segment the research direction words to be matched and each academic subject word in the academic subject word library to obtain a first pair sequence corresponding to the research direction words to be matched and a second pair sequence corresponding to each academic subject word, including: Input the research direction words to be matched and each academic subject word in the academic subject word library into the preconfigured 2GRAM model; Through the sliding window in the preconfigured 2GRAM model, traverse each character in the research direction word to be matched, obtain multiple first double-word pairs corresponding to the research direction word to be matched, and generate a first double-pair sequence corresponding to the research direction word to be matched; Through the sliding window in the preconfigured 2GRAM model, each character in each academic subject word is traversed to obtain multiple second bigram pairs corresponding to each academic subject word and generate a second bigram pair sequence corresponding to each academic subject word.

3. The method according to claim 1, characterized in that The second two-word pair sequence corresponding to each of the academic subject words is respectively compared with the first two-word pair sequence corresponding to the research direction word to be matched, and a plurality of academic subject words to be matched are determined from the academic subject word library, including: Calculate the Jaccard distance between the second bigram pair sequence corresponding to each academic subject word and the first bigram pair sequence corresponding to the research direction word to be matched, and obtain the similarity between each academic subject word and the research direction word to be matched; A plurality of academic subject words corresponding to similarities greater than or equal to a preset threshold are selected from the academic subject word library as academic subject words to be matched.

4. The method according to claim 3, characterized in that The step of calculating the Jaccard distance between the second bigram pair sequence corresponding to each academic subject word and the first bigram pair sequence corresponding to the research direction word to be matched, and obtaining the similarity between each academic subject word and the research direction word to be matched, comprises: Calculate the intersection and union between the second bigram pair sequence corresponding to each academic subject word and the first bigram pair sequence corresponding to the research direction word to be matched; Calculate the Jaccard distance between each academic subject word and the research direction word to be matched according to the intersection and the union; The Jaccard distance between each academic subject word and the research direction word to be matched is converted into similarity.

5. The method according to claim 1, characterized in that The preconfigured BERT model is used to perform text vectorization processing on the research direction words to be matched to obtain the research direction vector to be matched, and the academic subject words to be matched are performed text vectorization processing to obtain the academic subject vector to be matched, including: Using the word segmenter in the preconfigured BERT model to decompose the research direction words to be matched into a plurality of first basic units, and decompose the academic subject words to be matched into a plurality of second basic units; Inputting the first basic unit into an encoder in a BERT model for text vectorization processing to generate a vector representation corresponding to the first basic unit; Inputting the second basic unit into an encoder in a BERT model for text vectorization processing to generate a vector representation corresponding to the second basic unit; The vector representations corresponding to the multiple first basic units in the research direction words to be matched are merged to obtain the research direction vector to be matched corresponding to the research direction words to be matched, and the vector representations corresponding to the multiple second basic units in the academic subject words to be matched are merged to obtain the academic subject vector to be matched corresponding to the academic subject words to be matched.

6. The method according to claim 1, characterized in that The method of calculating the similarity between each of the academic subject vectors to be matched and the research direction vector to be matched, and determining an academic subject word to be matched from the plurality of academic subject words to be matched as a target academic subject word corresponding to the research direction word to be matched, comprises: The cosine distance is used to calculate the vector similarity between each of the academic subject vectors to be matched and the research direction vector to be matched, so as to obtain the matching similarity scores corresponding to each of the academic subject words to be matched; The academic subject word with the highest matching similarity score is selected from the multiple academic subject words to be matched as the target academic subject word corresponding to the research direction word to be matched.

7. The method according to claim 6, characterized in that Also includes: The research direction words to be matched, the target academic subject words and the matching similarity scores corresponding to the target academic subject words are displayed as output results in the form of structured data.

8. An entity matching device based on the research direction, characterized in that: include: An acquisition module is used to obtain the research direction words to be matched from the entity extraction results of the expert information; A segmentation module is used for segmenting the research direction words to be matched and each academic subject word in the academic subject word library by using a preconfigured 2GRAM model to obtain a first pair sequence corresponding to the research direction words to be matched and a second pair sequence corresponding to each academic subject word; A similarity comparison module is used to compare the similarity of the second two-word pair sequence corresponding to each academic subject word with the first two-word pair sequence corresponding to the research direction word to be matched, and determine a plurality of academic subject words to be matched from the academic subject word library; A vectorization processing module, used for performing text vectorization processing on the research direction words to be matched using a preconfigured BERT model to obtain a research direction vector to be matched, and performing text vectorization processing on the academic subject words to be matched to obtain an academic subject vector to be matched; The determination module is used to determine an academic subject word to be matched from the multiple academic subject words to be matched as the target academic subject word corresponding to the research direction word to be matched by calculating the similarity between each academic subject vector to be matched and the research direction vector to be matched.

9. An electronic device, characterized in that: The device comprises: a processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 7 when executed by a processor.