A method and related equipment for standardizing institution names based on knowledge graph

By using the knowledge graph representation learning method in the organization name specification and using TransR and RESCAL models to standardize the prediction of organization name, the problem of low accuracy of traditional methods is solved and higher standardized accuracy is achieved.

CN118070788BActive Publication Date: 2024-11-22INST OF MEDICAL INFORMATION CHINESE ACAD OF MEDICAL SCI
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410253730.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-06
Publication Date
2024-11-22
Estimated Expiration
2044-03-06

AI Technical Summary

Technical Problem

The traditional institutional name normative method based on word vector model has low accuracy, resulting in inaccurate results of inaccurate organization name normative.

Method used

Using a representation learning method based on knowledge graph, the first and second prediction models are constructed using the translation distance model TransR and the semantic matching model RESCAL, respectively, to standardize the prediction of the organization name, and finalize it with the results of the two models.

Benefits of technology

It significantly improves the accuracy of organization name specifications, achieves more accurate standardized results, and enhances the accuracy of prediction through rich knowledge mining and correlation reasoning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118070788B_ABST
    Figure CN118070788B_ABST
Patent Text Reader

Abstract

The present application discloses a method for standardizing institution names based on a knowledge graph and related equipment, wherein the method includes: obtaining the bibliographic information of the literature to be processed that at least includes the non-standard institution name; constructing the model input information of the first triplet containing at least the missing head entity, the tail entity, the missing head entity and the relationship in the first triplet are the non-standard institution name and the corresponding standard institution name to be predicted and the relationship between the two; inputting the model input information into the first and second prediction models, so that the two models predict the standard institution name; according to the prediction results of the two models, determining the standard institution name corresponding to the non-standard institution name; the two models use different knowledge graph representation learning algorithms to obtain the vector representation of entities and relationships, and predict the standard institution name based on the learned vector representation. The present application effectively improves the accuracy of institution name standardization processing by adopting the knowledge graph representation learning method to standardize institution names.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the field of natural language processing and artificial intelligence technology, and in particular, relates to a method for standardizing institution names based on a knowledge graph and related equipment. Background Art

[0002] The institution names in the bibliographic information are usually filled in by the authors, and are often non-standard and irregular. Therefore, it is necessary to standardize and normalize the institution names in the bibliographic information to facilitate the query and use of the literature.

[0003] Traditional technology generally provides an institution name standardization method based on a word vector model. This method has the problem of low standardization processing accuracy, which can easily make the final institution name standardization result inaccurate. Summary of the invention

[0004] In view of this, the present application discloses a method and related equipment for standardizing institution names based on a knowledge graph, which effectively improves the accuracy of institution name standardization processing by adopting a knowledge graph representation learning method to standardize institution names.

[0005] The specific plan is as follows:

[0006] A method for standardizing institution names based on knowledge graph, including:

[0007] Obtaining bibliographic information to be processed; the bibliographic information includes at least one entity, and the at least one entity includes a non-standard institution name to be converted into a standard institution name;

[0008] Construct model input information of a first triplet containing at least a missing head entity; the tail entity, the missing head entity and the relationship in the first triplet are respectively the non-standard institution name, the standard institution name to be predicted corresponding to the non-standard institution name, and the relationship between the non-standard institution name and the standard institution name; if the document bibliographic information includes multiple entities, the model input information also includes a corresponding number of second triples, and the head entity, the tail entity and the relationship in each second triplet are respectively the non-standard institution name, other entities other than the non-standard institution name in the document bibliographic information, and the relationship between the other entities and the non-standard institution name;

[0009] Inputting the model input information into a first prediction model and a second prediction model respectively, so that the first prediction model and the second prediction model respectively predict the standard institution name for the non-standard institution name, and obtaining a first prediction result output by the first prediction model and a second prediction result output by the second prediction model;

[0010] Determining a standard institution name corresponding to the non-standard institution name according to the first prediction result and the second prediction result;

[0011] Among them, the first prediction model and the second prediction model respectively use different knowledge graph representation learning algorithms to learn vector representations of entities and relationships, and predict the standard institution name corresponding to the non-standard institution name based on the learned vector representations of entities and relationships.

[0012] Optionally, the at least one entity also includes at least one of an institution address, a zip code, an author, and a research field corresponding to the non-standard institution name.

[0013] Optionally, the first prediction model and the second prediction model respectively predict the standard institution name for the non-standard institution name, including:

[0014] The first prediction model determines the missing head entity in the first triplet by link prediction based on the model input information and the vector representation of entities and relationships obtained by the first knowledge graph representation learning algorithm to obtain the first prediction result;

[0015] The second prediction model is based on the model input information and the vector representation of entities and relationships obtained by a second knowledge graph representation learning algorithm, and determines the missing head entity in the first triplet through a link prediction method to obtain the second prediction result.

[0016] Optionally, the first prediction result includes a first set of standard institution names, the first set of standard institution names includes each candidate standard institution name of topk1 before the score predicted by the first prediction model for the non-standard institution name; the second prediction result includes a second set of standard institution names, the second set of standard institution names includes each candidate standard institution name of topk2 before the score predicted by the second prediction model for the non-standard institution name; k1 and k2 are respectively integers greater than 0;

[0017] The determining, according to the first prediction result and the second prediction result, a standard institution name corresponding to the non-standard institution name includes:

[0018] According to at least one of the scores and the ranking order of the scores corresponding to each candidate standard organization name in the first set of standard organization names, and at least one of the scores and the ranking order of the scores corresponding to each candidate standard organization name in the second set of standard organization names, determine the candidate standard organization name that meets at least one of the score condition and the ranking condition, and obtain the standard organization name corresponding to the non-standard organization name.

[0019] Optionally, the first knowledge graph representation learning algorithm is a translation distance model TransR algorithm, and the second knowledge graph representation learning algorithm is a semantic matching model RESCAL algorithm.

[0020] Optionally, the process of constructing the first prediction model includes:

[0021] Acquire multiple document bibliographic information samples; each document bibliographic information sample includes multiple entities, the multiple entities include a non-standard institution name and a corresponding standard institution name, and also include at least one of an institution address, a zip code, an author, and a research field corresponding to the non-standard institution name;

[0022] Associating the non-standard institution name in each document bibliographic information sample with the corresponding standard institution name, and associating each entity other than the non-standard institution name with the non-standard institution name, respectively, to construct a triple, and obtain a knowledge graph formed by each triple corresponding to each document bibliographic information sample;

[0023] Taking the triples in the knowledge graph as positive samples, negative sampling is performed on at least one of the entities and relations contained in the triples in the knowledge graph to generate negative samples, so as to obtain a training set including at least part of the positive samples and at least part of the negative samples;

[0024] Based on the training set, a projection mapping matrix is ​​learned in a first knowledge graph representation learning algorithm according to a corresponding optimization objective to obtain vector representations of entities and relationships; the first prediction model is a model obtained after the first knowledge graph representation learning algorithm learns the vector representations of entities and relationships;

[0025] Among them, the projection mapping matrix is ​​used to map the entity vector of the entity in the triplet to the space where the relationship is located; the optimization goal corresponding to the first knowledge graph representation learning algorithm is: maximizing the score of the first scoring function for the positive samples in the training set, and minimizing the score of the first scoring function for the negative samples in the training set.

[0026] Optionally, the construction process of the second prediction model includes:

[0027] Acquire multiple document bibliographic information samples; each document bibliographic information sample includes multiple entities, the multiple entities include a non-standard institution name and a corresponding standard institution name, and also include at least one of an institution address, a zip code, an author, and a research field corresponding to the non-standard institution name;

[0028] Associating the non-standard institution name in each document bibliographic information sample with the corresponding standard institution name, and associating each entity other than the non-standard institution name with the non-standard institution name, respectively, to construct a triple, and obtain a knowledge graph formed by each triple corresponding to each document bibliographic information sample;

[0029] Taking the triples in the knowledge graph as positive samples, negative sampling is performed on at least one of the entities and relations contained in the triples in the knowledge graph to generate negative samples, so as to obtain a training set including at least part of the positive samples and at least part of the negative samples;

[0030] Based on the training set, the vector representation of entities and relationships is obtained by optimizing the model parameters based on the third-order tensor according to the corresponding optimization objectives in the second knowledge graph representation learning algorithm; the second prediction model is a model obtained after the second knowledge graph representation learning algorithm learns the vector representation of entities and relationships;

[0031] Among them, the third-order tensor includes a two-dimensional tangent plane corresponding to each relationship in each sample in the training set, and the two-dimensional tangent plane corresponding to each relationship includes a head-tail entity combination vector representation between each head entity and each tail entity provided by the training set; if the relationship exists between the head entity and the tail entity, in the two-dimensional tangent plane corresponding to the relationship, the head-tail entity combination vector representation of the head entity and the tail entity is 1, if the relationship does not exist between the head entity and the tail entity, in the two-dimensional tangent plane corresponding to the relationship, the head-tail entity combination vector representation of the head entity and the tail entity is 0; the second knowledge graph represents that the optimization goal corresponding to the learning algorithm is: to maximize the score of the second scoring function for the positive samples in the training set, and to minimize the score of the second scoring function for the negative samples in the training set.

[0032] Optionally, after obtaining the knowledge graph, the method further includes:

[0033] The knowledge graph is data-fused with a standardized ontology to standardize corresponding entities in triples of the knowledge graph.

[0034] A device for regulating the name of an institution based on a knowledge graph, comprising:

[0035] An acquisition module, used for acquiring bibliographic information to be processed; the bibliographic information includes at least one entity, and the at least one entity includes a non-standard institution name to be converted into a standard institution name;

[0036] A construction module, used to construct model input information of a first triplet containing at least a missing head entity; the tail entity, the missing head entity and the relationship in the first triplet are respectively the non-standard institution name, the standard institution name to be predicted corresponding to the non-standard institution name, and the relationship between the non-standard institution name and the standard institution name; if the document bibliographic information includes multiple entities, the model input information also includes a corresponding number of second triples, and the head entity, tail entity and relationship in each second triplet are respectively the non-standard institution name, other entities other than the non-standard institution name in the document bibliographic information, and the relationship between the other entities and the non-standard institution name;

[0037] A prediction module, used for inputting the model input information into a first prediction model and a second prediction model respectively, so that the first prediction model and the second prediction model respectively predict the standard institution name for the non-standard institution name, and obtain a first prediction result output by the first prediction model and a second prediction result output by the second prediction model;

[0038] A determination module, configured to determine a standard institution name corresponding to the non-standard institution name according to the first prediction result and the second prediction result;

[0039] Among them, the first prediction model and the second prediction model respectively use different knowledge graph representation learning algorithms to learn vector representations of entities and relationships, and predict the standard institution name corresponding to the non-standard institution name based on the learned vector representations of entities and relationships.

[0040] An electronic device, comprising:

[0041] A memory for storing at least one set of computer instructions;

[0042] A processor, configured to implement the method for standardizing institution names based on a knowledge graph as described in any one of the above items by executing the instruction set stored in the memory.

[0043] In summary, the present application provides a method and related equipment for standardizing institution names based on knowledge graphs. The method and related equipment pre-construct a first prediction model and a second prediction model. The two models use different knowledge graph representation learning algorithms to obtain vector representations of entities and relationships, and predict the standard institution name corresponding to the non-standard institution name based on the vector representations of the learned entities and relationships. On the basis of constructing the first prediction model and the second prediction model, for the bibliographic information including at least the non-standard institution name to be standardized, a model input information of at least a first triple containing a missing head entity is constructed; the tail entity, the missing head entity and the relationship in the first triple are the non-standard institution name, the standard institution name to be predicted corresponding to the non-standard institution name, and the relationship between the non-standard institution name and the standard institution name, respectively, and the model input information is input into the first prediction model and the second prediction model respectively, so as to use the first prediction model and the second prediction model to predict the standard institution name for the non-standard institution name, respectively, and obtain the first prediction result output by the first prediction model and the second prediction result output by the second prediction model, and finally determine the standard institution name corresponding to the non-standard institution name according to the first prediction result and the second prediction result.

[0044] It can be seen that this application proposes and implements a solution for institution name standardization using a knowledge graph representation learning method. Compared with the traditional knowledge graph-based institution name standardization method based on a word vector model, the use of a knowledge graph representation learning method for institution name standardization can achieve richer and more comprehensive knowledge mining, association and reasoning, thereby effectively improving the accuracy of institution name standardization, and thus making the final institution name standardization result more accurate. In addition, the embodiment of the present application further improves the accuracy of institution name standardization by using two prediction models to perform joint reasoning / prediction based on knowledge graph representation learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0046] Figure 1 It is a schematic diagram of the first prediction model construction process provided by the present application;

[0047] Figure 2 is a schematic diagram of the second prediction model construction process provided by the present application;

[0048] Figure 3is a schematic diagram of a third-order tensor provided in this application;

[0049] Figure 4 It is a flow chart of the method for standardizing institution names based on knowledge graph provided by this application;

[0050] Figure 5 It is a composition structure diagram of the knowledge graph-based institution name standardization device provided by this application;

[0051] Figure 6 It is a structural diagram of the electronic device provided in this application. DETAILED DESCRIPTION

[0052] 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.

[0053] An embodiment of the present application provides a method and related equipment for standardizing institution names based on a knowledge graph. The method and related equipment convert document bibliographic information (including non-standard institution names / standard institution names, addresses, authors, zip codes, etc.) into a knowledge graph, use a knowledge graph representation learning method to obtain vector representations of entities and relationships, and normalize the institution name based on the learned entity / relationship vector representations.

[0054] The knowledge graph-based institution name standardization method provided in the embodiment of the present application requires the pre-construction of a first prediction model and a second prediction model as a prerequisite, and specifically implements the standardization of the institution name through the pre-constructed first prediction model and the second prediction model.

[0055] The following first describes the construction process of the first prediction model and the second prediction model.

[0056] Among them, the first prediction model and the second prediction model respectively use different knowledge graph representation learning algorithms to learn vector representations of entities and relationships, and predict the standard institution name corresponding to the non-standard institution name based on the learned vector representations of entities and relationships.

[0057] See also Figure 1 The schematic diagram of the first prediction model construction process shown in FIG. 1 includes:

[0058] Step 101, obtain multiple document bibliographic information samples; each document bibliographic information sample includes multiple entities, and the multiple entities include a non-standard institution name and a corresponding standard institution name, and also include at least one of the institution address, zip code, author, and research field corresponding to the non-standard institution name.

[0059] Specifically, a series of original bibliographic information can be collected as a sample of document bibliographic information, wherein the collected original bibliographic information usually includes a non-standard institution name provided by a user, as well as information such as the institution address, zip code, author and / or research field. In addition, it also includes the standard institution name corresponding to the non-standard institution name.

[0060] Step 102: Associating the non-standard institution name in each document bibliographic information sample with the corresponding standard institution name, and associating each entity other than the non-standard institution name with the non-standard institution name, respectively, to construct a triple, and obtain a knowledge graph formed by each triple corresponding to each document bibliographic information sample.

[0061] Optionally, before constructing triples, data cleaning may be performed on each document bibliographic information sample. The cleaning operations include but are not limited to: removing possessives, removing stop words, converting data to lowercase, removing punctuation, splitting address information by province, city, and district, etc.

[0062] In this embodiment, a triple consisting of a head entity, a relationship, and a tail entity is constructed by associating the non-standard institution name in each document bibliographic information sample with the corresponding standard institution name, and associating each entity other than the non-standard institution name with the non-standard institution name in the document bibliographic information sample, that is, head entity-relationship-tail entity. Exemplarily, each constructed triple is as follows:

[0063] Standard organization name - alias - non-standard organization name;

[0064] Non-standard institution name-address-province;

[0065] Non-standard institution name-address-city;

[0066] Non-standard institution name-address-district;

[0067] Non-standard institution name-zip code-100000.

[0068] For example:

[0069] "dalian university of technology-alias-dalian univ technol";

[0070] "dalian univ technol-address-dalian";

[0071] "dalian univ technol-zip-116023".

[0072] Among them, Dalian University of Technology is the standard institution name, Dalian University of Technology is the non-standard institution name, "Dalian" is the address, and "116023" is the postal code.

[0073] Each piece of document title information forms multiple triples, and a series of triples corresponding to a series of document title information further forms a knowledge graph.

[0074] Step 103: Take the triples in the knowledge graph as positive samples, and perform negative sampling on at least one of the entities and relationships contained in the triples in the knowledge graph to generate negative samples, so as to obtain a training set including at least part of the positive samples and at least part of the negative samples.

[0075] For the constructed knowledge graph, each triple (h, r, t) contained therein is used as a positive sample, and the entity (h, t) and / or relationship (r) in the triple contained in the knowledge graph is negatively sampled to generate a negative sample. On this basis, the positive sample and the negative sample can be merged to obtain a training set for model training to generate a first prediction model. In implementation, optionally, the merged positive / negative sample set can be divided into two parts, respectively as a training set and a test set.

[0076] Step 104: Based on the training set, the projection mapping matrix is ​​learned in the first knowledge graph representation learning algorithm according to the corresponding optimization objective to obtain the vector representation of entities and relationships; the first prediction model is the model obtained after the first knowledge graph representation learning algorithm learns the vector representation of entities and relationships.

[0077] Among them, the projection mapping matrix is ​​used to map the entity vector of the entity in the triplet to the space where the relationship is located; the optimization goal corresponding to the first knowledge graph representation learning algorithm is: maximizing the score of the first scoring function for the positive samples in the training set, and minimizing the score of the first scoring function for the negative samples in the training set.

[0078] The first knowledge graph representation learning algorithm may be, but is not limited to, a translation distance model series algorithm such as TransR, TransE, and TransD.

[0079] When training the translation distance model, the entities and relations of the model (i.e., the entities and relations in each triple in the training set) can be initialized with random vectors, and the head and tail entities of each sample in the training set are represented in the entity space, and the relations are represented in the relation space. For each relation r, the projection mapping matrix Wr is learned, and the entity vectors h and t in each triple in the training set are mapped to the space where the relation r is located through this matrix, and the mapping is obtained. r =hW r , and t r =tW r During the training process, just like neural networks, the vector representations of entities and relationships are learned according to the optimization goal, which is to maximize the scoring function for positive samples while minimizing the scoring function for negative samples. For example, (anhui agricultural university, alias, anhui agr univ) and (anhui agricultural university, alias, anhui acad agr sci) are correct and positive samples, while the latter are wrong and negative samples. When training the model, the scoring function score of the positive samples will be maximized and the scoring function score of the negative samples will be minimized.

[0080] Optionally, the scoring function for translation distance models such as TransR is: Finally, the trained model parameter results are saved in the model, and vector representations of entities and relationships are obtained.

[0081] See also Figure 2 The schematic diagram of the second prediction model construction process shown in FIG. 1 includes:

[0082] Step 201, obtain multiple document bibliographic information samples; each document bibliographic information sample includes multiple entities, and the multiple entities include a non-standard institution name and a corresponding standard institution name, and also include at least one of the institution address, zip code, author, and research field corresponding to the non-standard institution name.

[0083] Step 202: Associating the non-standard institution name in each document bibliographic information sample with the corresponding standard institution name, and associating each entity other than the non-standard institution name with the non-standard institution name, respectively, to construct a triple, and obtain a knowledge graph formed by each triple corresponding to each document bibliographic information sample.

[0084] Step 203: Take the triples in the knowledge graph as positive samples, and perform negative sampling on at least one of the entities and relationships contained in the triples in the knowledge graph to generate negative samples, so as to obtain a training set including at least part of the positive samples and at least part of the negative samples.

[0085] Among them, steps 201-203 are the same as steps 101-103, and the details can be found in the above description of steps 101-103, which will not be described in detail.

[0086] Step 204: Based on the training set, the vector representation of entities and relationships is obtained by optimizing the model parameters based on the third-order tensor according to the corresponding optimization objectives in the second knowledge graph representation learning algorithm; the second prediction model is the model obtained after the second knowledge graph representation learning algorithm learns the vector representation of entities and relationships.

[0087] Among them, the third-order tensor includes a two-dimensional tangent plane corresponding to each relationship in each sample in the training set, and the two-dimensional tangent plane corresponding to each relationship is a series of entity vector representations, including the head and tail entity combination vector representations between each head entity and each tail entity provided by the training set; if the relationship exists between the head entity and the tail entity, in the two-dimensional tangent plane corresponding to the relationship, the head and tail entity combination vector representation of the head entity and the tail entity is 1; if the relationship does not exist between the head entity and the tail entity, in the two-dimensional tangent plane corresponding to the relationship, the head and tail entity combination vector representation of the head entity and the tail entity is 0.

[0088] The optimization objective corresponding to the second knowledge graph representation learning algorithm is to maximize the score of the second scoring function for the positive samples in the training set, and to minimize the score of the second scoring function for the negative samples in the training set.

[0089] The second knowledge graph representation learning algorithm may be, but is not limited to, a series of algorithms of semantic matching models such as RESCAL, DistMult, and ComplEx.

[0090] Semantic matching models such as RESCAL are based on tensor decomposition models. The tensor decomposition model regards the entire sample set as a three-dimensional adjacency matrix (third-order tensor), see Figure 3 The schematic diagram of the third-order tensor shown in the figure shows that the two-dimensional tangent plane is represented by the serial entity vectors, and the third dimension is the relationship predicate between the entities, that is, m is defined as the number of all relations in the sample set, n is the number of head entities / tail entities in the sample set, and each relationship R m Corresponds to a two-dimensional tangent plane (two-dimensional adjacency matrix) in the third-order tensor. m If there is a relationship between the head entity and the tail entity, the corresponding element value of the tensor is 1. If not, the corresponding element value is 0. m and n are integers greater than 1.

[0091] Optionally, the scoring function of the semantic matching model such as RESCAL is: X k ≈AR k AT ,k=1,...,m. Model parameters A, R k It is calculated by constrained minimization, where A is the entity matrix, R k is a relation matrix. Semantic matching models such as RESCAL train models by optimizing model parameters based on third-order tensors according to corresponding optimization objectives, and finally save the trained parameter results into the model to obtain vector representations of entities and relations.

[0092] The embodiment of the present application improves the accuracy of standard organization name prediction by constructing two prediction models, namely a first prediction model and a second prediction model, and combining the two prediction models to predict standard organization names based on a knowledge graph representation learning algorithm.

[0093] Furthermore, the applicant's research found that the first prediction model and the second prediction model respectively adopt the translation distance model TransR algorithm and the semantic matching model RESCAL algorithm, that is, the model algorithm combination of "TransR+RESCAL", which can achieve better results in predicting standard institution names for non-standard institution names and has a higher prediction accuracy. Based on this, in implementation, the model algorithm of the first prediction model can preferably be TransR, and the model algorithm of the second prediction model can preferably be RESCAL.

[0094] In an optional embodiment, after obtaining the knowledge graph and before constructing positive and negative sample sets for model (first prediction model, second prediction model) training based on the knowledge graph, the knowledge graph can also be converted into a standard RDF (Resource Description Framework) representation format to facilitate subsequent data reading and processing based on the knowledge graph.

[0095] In an optional embodiment, after obtaining the knowledge graph, before constructing a set of positive and negative samples for model (first prediction model, second prediction model) training based on the knowledge graph, the knowledge graph may be first data-fused with a standardized ontology to standardize the corresponding entities in the triples of the knowledge graph.

[0096] Optionally, the knowledge graph can be fused with standardized administrative division ontology and postal code ontology to optimize the vector representation in the next step of model training.

[0097] The following example illustrates the standardized administrative division ontology of a certain region: "china-include-hebei", "Hebei-include-shijiazhuang", and maps the address entity in the knowledge graph with the address entity name ("china", "hebei", "shijiazhuang") in the administrative division ontology. Specifically, but not limited to, the Jaccard similarity calculation method can be used to calculate the similarity between each address entity in the knowledge graph and each address entity in the administrative division ontology, and the knowledge graph entities (i.e., entities in the knowledge graph) whose similarity between entities reaches a similarity threshold (for example, the threshold is set to 0.9) are fused with the entities in the administrative division ontology. After the fusion, the entities in the knowledge graph are represented by the corresponding entities in the administrative division ontology, thereby standardizing the entities in the triples of the knowledge graph, thereby facilitating the optimization of the vector representation in the next step of model training.

[0098] On the basis of constructing the first prediction model and the second prediction model, see Figure 4 The flowchart of the method for standardizing the institution name based on the knowledge graph is shown in the embodiment of the present application. The method for standardizing the institution name based on the knowledge graph includes the following processing steps:

[0099] Step 401, obtaining document bibliographic information to be processed; the document bibliographic information includes at least one entity, and the at least one entity includes a non-standard institution name to be converted into a standard institution name.

[0100] The document bibliographic information to be processed may include, in addition to the non-standard institution name to be converted into the standard institution name, at least one of the institution address, zip code, author, and research field corresponding to the non-standard institution name but not limited to.

[0101] Step 402: construct model input information of a first triplet containing at least a missing head entity; the tail entity, the missing head entity, and the relationship in the first triplet are respectively the non-standard institution name, the standard institution name to be predicted corresponding to the non-standard institution name, and the relationship between the non-standard institution name and the standard institution name.

[0102] If the document title information includes multiple entities, the model input information also includes a corresponding number of second triples, and the head entity, tail entity and relationship in each second triple are respectively the non-standard institution name, other entities other than the non-standard institution name in the document title information, and the relationship between the other entities and the non-standard institution name.

[0103] Taking the case where the bibliographic information to be processed includes the non-standard institution name "dalian univ technol", the address "dalian" and the zip code "116023", the following first triplet can be constructed:

[0104] X-alias-dalian univ technol;

[0105] Where X is the name of the standard institution to be predicted corresponding to the non-standard institution name "dalian univ technol".

[0106] And construct the following second triple;

[0107] dalian univ technol-address-dalian;

[0108] Dalian Univ Technol-Zip-116023.

[0109] The constructed first triplet and the second triplet together constitute the model input information.

[0110] Step 403: input the model input information into the first prediction model and the second prediction model respectively, so that the first prediction model and the second prediction model respectively predict the standard institution name for the non-standard institution name, and obtain a first prediction result output by the first prediction model and a second prediction result output by the second prediction model.

[0111] Among them, the first prediction model is based on the model input information and the vector representation of entities and relationships obtained by the first knowledge graph representation learning algorithm, and determines the missing head entity in the first triplet through a link prediction method to obtain the first prediction result.

[0112] Similarly, the second prediction model determines the missing head entity in the first triplet through link prediction based on the model input information and the vector representation of entities and relationships obtained by the second knowledge graph representation learning algorithm to obtain the second prediction result.

[0113] Specifically, in the prediction process, the first prediction model and the second prediction model respectively use all entities existing in the dataset space learned by their models as candidate entities, and fill them into the knowledge gaps to be predicted in turn, that is, to complete the prediction for triples of the type (?, r, t), which is the corresponding head entity prediction in the link prediction. And the score values ​​of the corresponding scoring functions (the first prediction model corresponds to the first scoring function, and the second prediction model corresponds to the second scoring function) are used as evaluation indicators to calculate the probability score of each candidate entity as the correct missing information. The higher the score of the candidate entity, the greater the possibility of it being the target entity of the link prediction result. In the whole process, the knowledge embedding representation learning process (i.e., the model training process) of the first prediction model / the second prediction model shares the same scoring function with the prediction process.

[0114] The first prediction result output by the first prediction model includes a first set of standard institution names, and the first set of standard institution names includes each candidate standard institution name of topk1 predicted by the first prediction model for the non-standard institution name before scoring.

[0115] The second prediction result output by the second prediction model includes a second set of standard organization names, and the second set of standard organization names includes the topk2 candidate standard organization names predicted by the second prediction model for the non-standard organization name; k1 and k2 are integers greater than 0 respectively.

[0116] The first set of standardization organization names and the second set of standardization organization names may be in, but are not limited to, list form respectively.

[0117] Step 404: Determine the standard institution name corresponding to the non-standard institution name according to the first prediction result and the second prediction result.

[0118] Among them, the first prediction model and the second prediction model respectively use different knowledge graph representation learning algorithms to learn vector representations of entities and relationships, and predict the standard institution name corresponding to the non-standard institution name based on the learned vector representations of entities and relationships.

[0119] In this step 404, specifically, based on at least one of the scores and the ranking order of the scores corresponding to each candidate standard organization name in the first set of standard organization names, and at least one of the scores and the ranking order of the scores corresponding to each candidate standard organization name in the second set of standard organization names, a candidate standard organization name that meets at least one of the score conditions and the ranking order of the scores can be determined to obtain the standard organization name corresponding to the non-standard organization name.

[0120] In implementation, optionally, corresponding weights may be assigned to the first set of standard organization names and the second set of standard organization names, respectively. The weight corresponding to the first set of standard organization names specifically represents the weight assigned to each candidate standard organization name in the first set of standard organization names. Similarly, the weight corresponding to the second set of standard organization names specifically represents the weight assigned to each candidate standard organization name in the second set of standard organization names.

[0121] On this basis, according to the scores and / or ranking ranks of scores corresponding to each candidate standard institution name in the first set of standard institution names, the scores and / or ranking ranks of scores corresponding to each candidate standard institution name in the second set of standard institution names, and their respective corresponding weights, the comprehensive scores and / or comprehensive ranking ranks of scores corresponding to each candidate standard institution name in the two sets are calculated, and the candidate standard institution names whose corresponding comprehensive scores and / or comprehensive ranking ranks meet the scoring conditions and / or ranking conditions are screened out and used as the standard institution names corresponding to the non-standard institution names, thereby achieving standardization of non-standard institution names.

[0122] The scoring condition can be set as the comprehensive score corresponding to the candidate standard organization name reaches the set score value, and the ranking condition can be set as the comprehensive ranking corresponding to the candidate standard organization name reaches the set ranking. Preferably, in this embodiment, the scoring condition is set as the comprehensive score corresponding to the candidate standard organization name is the highest score, and the ranking condition is set as the comprehensive ranking corresponding to the candidate standard organization name is the highest ranking.

[0123] In summary, the institution name standardization method based on the knowledge graph provided in the embodiment of the present application pre-constructs a first prediction model and a second prediction model. The two models use different knowledge graph representation learning algorithms to obtain vector representations of entities and relationships, and predict the standard institution name corresponding to the non-standard institution name based on the learned vector representations of entities and relationships. On the basis of constructing the first prediction model and the second prediction model, for the literature bibliographic information including at least the non-standard institution name to be standardized, a model input information of at least a first triple containing a missing head entity is constructed; the tail entity, the missing head entity and the relationship in the first triple are the non-standard institution name, the standard institution name to be predicted corresponding to the non-standard institution name, and the relationship between the non-standard institution name and the standard institution name, respectively, and by inputting the model input information into the first prediction model and the second prediction model respectively, the first prediction model and the second prediction model are used to predict the standard institution name for the non-standard institution name, respectively, to obtain the first prediction result output by the first prediction model and the second prediction result output by the second prediction model, and finally, according to the first prediction result and the second prediction result, the standard institution name corresponding to the non-standard institution name is determined.

[0124] It can be seen that this application proposes and implements a solution for institution name standardization using a knowledge graph representation learning method. Compared with the traditional knowledge graph-based institution name standardization method based on a word vector model, the use of a knowledge graph representation learning method for institution name standardization can achieve richer and more comprehensive knowledge mining, association and reasoning, thereby effectively improving the accuracy of institution name standardization, and thus making the final institution name standardization result more accurate. In addition, the embodiment of the present application further improves the accuracy of institution name standardization by using two prediction models to perform joint reasoning / prediction based on knowledge graph representation learning.

[0125] Corresponding to the above-mentioned method for regulating the name of an institution based on a knowledge graph, the embodiment of the present application also provides a device for regulating the name of an institution based on a knowledge graph, and its composition structure is as follows: Figure 5 As shown, including:

[0126] An acquisition module 501 is used to acquire bibliographic information to be processed; the bibliographic information includes at least one entity, and the at least one entity includes at least a non-standard institution name to be converted into a standard institution name;

[0127] A construction module 502 is used to construct model input information of a first triplet containing at least a missing head entity; the tail entity, the missing head entity and the relationship in the first triplet are respectively the non-standard institution name, the standard institution name to be predicted corresponding to the non-standard institution name, and the relationship between the non-standard institution name and the standard institution name; if the bibliographic information includes multiple entities, the model input information also includes a corresponding number of second triples, and the head entity, the tail entity and the relationship in each second triplet are respectively the non-standard institution name, other entities other than the non-standard institution name in the bibliographic information, and the relationship between the other entities and the non-standard institution name;

[0128] Prediction module 503, used for inputting the model input information into a first prediction model and a second prediction model respectively, so that the first prediction model and the second prediction model respectively predict the standard institution name for the non-standard institution name, and obtain a first prediction result output by the first prediction model and a second prediction result output by the second prediction model;

[0129] A determination module 504, configured to determine a standard institution name corresponding to the non-standard institution name according to the first prediction result and the second prediction result;

[0130] Among them, the first prediction model and the second prediction model respectively use different knowledge graph representation learning algorithms to learn vector representations of entities and relationships, and predict the standard institution name corresponding to the non-standard institution name based on the learned vector representations of entities and relationships.

[0131] In an optional implementation, the at least one entity further includes at least one of an institution address, a zip code, an author, and a research field corresponding to the non-standard institution name.

[0132] In an optional implementation, the first prediction model is specifically used to: determine the missing head entity in the first triple by link prediction based on the model input information and the vector representation of entities and relationships obtained by the first knowledge graph representation learning algorithm, so as to obtain the first prediction result;

[0133] The second prediction model is specifically used to: determine the missing head entity in the first triplet by link prediction based on the model input information and the vector representation of entities and relationships obtained by the second knowledge graph representation learning algorithm, so as to obtain the second prediction result;

[0134] The tail entities and relations in each triplet of the missing head entity are respectively the entities and relations in each information group.

[0135] In an optional embodiment, the first prediction result includes a first standard institution name set, the first standard institution name set includes each candidate standard institution name of topk1 before the score predicted by the first prediction model for the non-standard institution name; the second prediction result includes a second standard institution name set, the second standard institution name set includes each candidate standard institution name of topk2 before the score predicted by the second prediction model for the non-standard institution name; k1 and k2 are respectively integers greater than 0;

[0136] The second determination module 504 is specifically used to: determine the candidate standard organization name that meets at least one of the score condition and the ranking condition based on at least one of the scores and the ranking order of the scores corresponding to each candidate standard organization name in the first set of standard organization names, and at least one of the scores and the ranking order of the scores corresponding to each candidate standard organization name in the second set of standard organization names, and obtain the standard organization name corresponding to the non-standard organization name.

[0137] In an optional implementation, the first knowledge graph representation learning algorithm is a translation distance model TransR algorithm, and the second knowledge graph representation learning algorithm is a semantic matching model RESCAL algorithm.

[0138] In an optional implementation, the knowledge graph-based institution name standardization device of the present application further includes a model building module for building a first prediction model and a second prediction model;

[0139] When constructing the first prediction model, the model construction module is specifically used to:

[0140] Acquire multiple document bibliographic information samples; each document bibliographic information sample includes multiple entities, the multiple entities include a non-standard institution name and a corresponding standard institution name, and also include at least one of an institution address, a zip code, an author, and a research field corresponding to the non-standard institution name;

[0141] Associating the non-standard institution name in each document bibliographic information sample with the corresponding standard institution name, and associating each entity other than the non-standard institution name with the non-standard institution name, respectively, to construct a triple, and obtain a knowledge graph formed by each triple corresponding to each document bibliographic information sample;

[0142] Taking the triples in the knowledge graph as positive samples, negative sampling is performed on at least one of the entities and relations contained in the triples in the knowledge graph to generate negative samples, so as to obtain a training set including at least part of the positive samples and at least part of the negative samples;

[0143] Based on the training set, a projection mapping matrix is ​​learned in a first knowledge graph representation learning algorithm according to a corresponding optimization objective to obtain vector representations of entities and relationships; the first prediction model is a model obtained after the first knowledge graph representation learning algorithm learns the vector representations of entities and relationships;

[0144] Among them, the projection mapping matrix is ​​used to map the entity vector of the entity in the triplet to the space where the relationship is located; the optimization goal corresponding to the first knowledge graph representation learning algorithm is: maximizing the score of the first scoring function for the positive samples in the training set, and minimizing the score of the first scoring function for the negative samples in the training set.

[0145] In an optional implementation manner, when constructing the second prediction model, the model construction module is specifically used to:

[0146] Acquire multiple document bibliographic information samples; each document bibliographic information sample includes multiple entities, the multiple entities include a non-standard institution name and a corresponding standard institution name, and also include at least one of an institution address, a zip code, an author, and a research field corresponding to the non-standard institution name;

[0147] Associating the non-standard institution name in each document bibliographic information sample with the corresponding standard institution name, and associating each entity other than the non-standard institution name with the non-standard institution name, respectively, to construct a triple, and obtain a knowledge graph formed by each triple corresponding to each document bibliographic information sample;

[0148] Taking the triples in the knowledge graph as positive samples, negative sampling is performed on at least one of the entities and relations contained in the triples in the knowledge graph to generate negative samples, so as to obtain a training set including at least part of the positive samples and at least part of the negative samples;

[0149] Based on the training set, the vector representation of entities and relationships is obtained by optimizing the model parameters based on the third-order tensor according to the corresponding optimization objectives in the second knowledge graph representation learning algorithm; the second prediction model is a model obtained after the second knowledge graph representation learning algorithm learns the vector representation of entities and relationships;

[0150] Among them, the third-order tensor includes a two-dimensional tangent plane corresponding to each relationship in each sample in the training set, and the two-dimensional tangent plane corresponding to each relationship includes a head-tail entity combination vector representation between each head entity and each tail entity provided by the training set; if the relationship exists between the head entity and the tail entity, in the two-dimensional tangent plane corresponding to the relationship, the head-tail entity combination vector representation of the head entity and the tail entity is 1, if the relationship does not exist between the head entity and the tail entity, in the two-dimensional tangent plane corresponding to the relationship, the head-tail entity combination vector representation of the head entity and the tail entity is 0; the second knowledge graph represents that the optimization goal corresponding to the learning algorithm is: to maximize the score of the second scoring function for the positive samples in the training set, and to minimize the score of the second scoring function for the negative samples in the training set.

[0151] In an optional embodiment, the knowledge graph-based organization name standardization device provided in the embodiment of the present application also includes an ontology fusion module, which is used to: after obtaining the knowledge graph, perform data fusion on the knowledge graph and the standardized ontology to standardize the corresponding entities in the triples of the knowledge graph.

[0152] As for the knowledge graph-based institution name standardization device provided in the embodiment of the present application, since it corresponds to the knowledge graph-based institution name standardization method provided in the above method embodiment, the description is relatively simple. For relevant similarities, please refer to the description of the above method embodiment, which will not be described in detail here.

[0153] The present application also discloses an electronic device, the composition structure of the electronic device, such as Figure 6 As shown, including at least:

[0154] A memory 10, for storing a computer instruction set;

[0155] A set of computer instructions can be implemented in the form of a computer program.

[0156] The processor 20 is used to implement the method for standardizing the organization name based on the knowledge graph as disclosed in any of the method embodiments above by executing a computer instruction set.

[0157] The processor 20 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices.

[0158] Optionally, the electronic device may also include a display device and / or a display interface and may be connected to an external display device.

[0159] Optionally, the electronic device may further include a camera assembly, and / or be connected to an external camera assembly.

[0160] In addition, the electronic device may also include components such as a communication interface and a communication bus. The memory, the processor and the communication interface communicate with each other through the communication bus.

[0161] The communication interface is used for communication between electronic devices and other devices. The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0162] An embodiment of the present application also provides a computer-readable medium having a computer instruction set stored thereon, which, when executed by a processor, is used to implement a knowledge graph-based institution name standardization method as provided in any of the method embodiments above.

[0163] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0164] For the convenience of description, the above system or device is described by dividing it into various modules or units according to its functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0165] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or the part that makes a creative contribution can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., including several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0166] Finally, it should be noted that, in this article, relational terms such as first, second, third and fourth are used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0167] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for standardizing institution names based on knowledge graph, characterized in that: include: Obtaining bibliographic information to be processed; the bibliographic information includes at least one entity, and the at least one entity includes a non-standard institution name to be converted into a standard institution name; Construct model input information of a first triplet containing at least a missing head entity; the tail entity, the missing head entity and the relationship in the first triplet are respectively the non-standard institution name, the standard institution name to be predicted corresponding to the non-standard institution name, and the relationship between the non-standard institution name and the standard institution name; if the document bibliographic information includes multiple entities, the model input information also includes a corresponding number of second triples, and the head entity, the tail entity and the relationship in each second triplet are respectively the non-standard institution name, other entities other than the non-standard institution name in the document bibliographic information, and the relationship between the other entities and the non-standard institution name; Inputting the model input information into a first prediction model and a second prediction model respectively, so that the first prediction model and the second prediction model respectively predict the standard institution name for the non-standard institution name, and obtaining a first prediction result output by the first prediction model and a second prediction result output by the second prediction model; Determining a standard institution name corresponding to the non-standard institution name according to the first prediction result and the second prediction result; The first prediction model and the second prediction model respectively use different knowledge graph representation learning algorithms to learn vector representations of entities and relationships, and predict the standard institution name corresponding to the non-standard institution name based on the learned vector representations of entities and relationships; The construction process of the first prediction model includes: Acquire multiple document bibliographic information samples; each document bibliographic information sample includes multiple entities, the multiple entities include a non-standard institution name and a corresponding standard institution name, and also include at least one of an institution address, a zip code, an author, and a research field corresponding to the non-standard institution name; Associating the non-standard institution name in each document bibliographic information sample with the corresponding standard institution name, and associating each entity other than the non-standard institution name with the non-standard institution name, respectively, to construct a triple, and obtain a knowledge graph formed by each triple corresponding to each document bibliographic information sample; Taking the triples in the knowledge graph as positive samples, negative sampling is performed on at least one of the entities and relations contained in the triples in the knowledge graph to generate negative samples, so as to obtain a training set including at least part of the positive samples and at least part of the negative samples; Based on the training set, a projection mapping matrix is ​​learned in a first knowledge graph representation learning algorithm according to a corresponding optimization objective to obtain vector representations of entities and relationships; the first prediction model is a model obtained after the first knowledge graph representation learning algorithm learns the vector representations of entities and relationships; The projection mapping matrix is ​​used to map the entity vector of the entity in the triple to the space where the relationship is located; the optimization goal corresponding to the first knowledge graph representation learning algorithm is: maximizing the score of the first scoring function for the positive samples in the training set, and minimizing the score of the first scoring function for the negative samples in the training set; The construction process of the second prediction model includes: Based on the training set, the vector representation of entities and relationships is obtained by optimizing the model parameters based on the third-order tensor according to the corresponding optimization objectives in the second knowledge graph representation learning algorithm; the second prediction model is a model obtained after the second knowledge graph representation learning algorithm learns the vector representation of entities and relationships; Among them, the third-order tensor includes a two-dimensional tangent plane corresponding to each relationship in each sample in the training set, and the two-dimensional tangent plane corresponding to each relationship includes a head-tail entity combination vector representation between each head entity and each tail entity provided by the training set; if the relationship exists between the head entity and the tail entity, in the two-dimensional tangent plane corresponding to the relationship, the head-tail entity combination vector representation of the head entity and the tail entity is 1, if the relationship does not exist between the head entity and the tail entity, in the two-dimensional tangent plane corresponding to the relationship, the head-tail entity combination vector representation of the head entity and the tail entity is 0; the second knowledge graph represents that the optimization goal corresponding to the learning algorithm is: to maximize the score of the second scoring function for the positive samples in the training set, and to minimize the score of the second scoring function for the negative samples in the training set.

2. The method for standardizing institution names based on knowledge graph according to claim 1 is characterized in that: The at least one entity also includes at least one of an institution address, a zip code, an author, and a research field corresponding to the non-standard institution name.

3. The method for standardizing institution names based on knowledge graph according to claim 1 is characterized in that: The first prediction model and the second prediction model respectively predict the standard institution name for the non-standard institution name, including: The first prediction model determines the missing head entity in the first triplet by link prediction based on the model input information and the vector representation of entities and relationships obtained by the first knowledge graph representation learning algorithm to obtain the first prediction result; The second prediction model is based on the model input information and the vector representation of entities and relationships obtained by a second knowledge graph representation learning algorithm, and determines the missing head entity in the first triplet through a link prediction method to obtain the second prediction result.

4. The method for standardizing institution names based on knowledge graph according to claim 1 is characterized in that: The first prediction result includes a first set of standard institution names, which includes topk1 candidate standard institution names predicted by the first prediction model for the non-standard institution name before scoring; the second prediction result includes a second set of standard institution names, which includes topk2 candidate standard institution names predicted by the second prediction model for the non-standard institution name before scoring; k1 and k2 are respectively integers greater than 0; The determining, according to the first prediction result and the second prediction result, a standard institution name corresponding to the non-standard institution name includes: According to at least one of the scores and the ranking order of the scores corresponding to each candidate standard organization name in the first set of standard organization names, and at least one of the scores and the ranking order of the scores corresponding to each candidate standard organization name in the second set of standard organization names, determine the candidate standard organization name that meets at least one of the score condition and the ranking condition, and obtain the standard organization name corresponding to the non-standard organization name.

5. The method for standardizing institution names based on knowledge graph according to claim 4 is characterized in that: The first knowledge graph representation learning algorithm is the translation distance model TransR algorithm, and the second knowledge graph representation learning algorithm is the semantic matching model RESCAL algorithm.

6. The method for standardizing institution names based on knowledge graph according to claim 1 is characterized in that: After obtaining the knowledge graph, it also includes: The knowledge graph is data-fused with a standardized ontology to standardize corresponding entities in triples of the knowledge graph.

7. A device for standardizing institution names based on knowledge graph, characterized in that: include: An acquisition module, used for acquiring bibliographic information to be processed; the bibliographic information includes at least one entity, and the at least one entity includes a non-standard institution name to be converted into a standard institution name; A construction module, used to construct model input information of a first triplet containing at least a missing head entity; the tail entity, the missing head entity and the relationship in the first triplet are respectively the non-standard institution name, the standard institution name to be predicted corresponding to the non-standard institution name, and the relationship between the non-standard institution name and the standard institution name; if the document bibliographic information includes multiple entities, the model input information also includes a corresponding number of second triples, and the head entity, tail entity and relationship in each second triplet are respectively the non-standard institution name, other entities other than the non-standard institution name in the document bibliographic information, and the relationship between the other entities and the non-standard institution name; A prediction module, used for inputting the model input information into a first prediction model and a second prediction model respectively, so that the first prediction model and the second prediction model respectively predict the standard institution name for the non-standard institution name, and obtain a first prediction result output by the first prediction model and a second prediction result output by the second prediction model; A determination module, configured to determine a standard institution name corresponding to the non-standard institution name according to the first prediction result and the second prediction result; The first prediction model and the second prediction model respectively use different knowledge graph representation learning algorithms to learn vector representations of entities and relationships, and predict the standard institution name corresponding to the non-standard institution name based on the learned vector representations of entities and relationships; A model building module, used to build the first prediction model and the second prediction model; When constructing the first prediction model, the model construction module is specifically used to: Acquire multiple document bibliographic information samples; each document bibliographic information sample includes multiple entities, the multiple entities include a non-standard institution name and a corresponding standard institution name, and also include at least one of an institution address, a zip code, an author, and a research field corresponding to the non-standard institution name; Associating the non-standard institution name in each document bibliographic information sample with the corresponding standard institution name, and associating each entity other than the non-standard institution name with the non-standard institution name, respectively, to construct a triple, and obtain a knowledge graph formed by each triple corresponding to each document bibliographic information sample; Taking the triples in the knowledge graph as positive samples, negative sampling is performed on at least one of the entities and relations contained in the triples in the knowledge graph to generate negative samples, so as to obtain a training set including at least part of the positive samples and at least part of the negative samples; Based on the training set, a projection mapping matrix is ​​learned in a first knowledge graph representation learning algorithm according to a corresponding optimization objective to obtain vector representations of entities and relationships; the first prediction model is a model obtained after the first knowledge graph representation learning algorithm learns the vector representations of entities and relationships; The projection mapping matrix is ​​used to map the entity vector of the entity in the triple to the space where the relationship is located; the optimization goal corresponding to the first knowledge graph representation learning algorithm is: maximizing the score of the first scoring function for the positive samples in the training set, and minimizing the score of the first scoring function for the negative samples in the training set; When constructing the second prediction model, the model construction module is specifically used to: Based on the training set, the vector representation of entities and relationships is obtained by optimizing the model parameters based on the third-order tensor according to the corresponding optimization objectives in the second knowledge graph representation learning algorithm; the second prediction model is a model obtained after the second knowledge graph representation learning algorithm learns the vector representation of entities and relationships; Among them, the third-order tensor includes a two-dimensional tangent plane corresponding to each relationship in each sample in the training set, and the two-dimensional tangent plane corresponding to each relationship includes a head-tail entity combination vector representation between each head entity and each tail entity provided by the training set; if the relationship exists between the head entity and the tail entity, in the two-dimensional tangent plane corresponding to the relationship, the head-tail entity combination vector representation of the head entity and the tail entity is 1, if the relationship does not exist between the head entity and the tail entity, in the two-dimensional tangent plane corresponding to the relationship, the head-tail entity combination vector representation of the head entity and the tail entity is 0; the second knowledge graph represents that the optimization goal corresponding to the learning algorithm is: to maximize the score of the second scoring function for the positive samples in the training set, and to minimize the score of the second scoring function for the negative samples in the training set.

8. An electronic device, characterized in that: include: A memory for storing at least one set of computer instructions; A processor, used to implement the method for standardizing institution names based on a knowledge graph as described in any one of claims 1 to 6 by executing the instruction set stored in the memory.