Large-scale multi-language knowledge graph alignment method based on node cutting and multiple division

Abstract

The invention discloses a large-scale multi-language knowledge graph alignment method based on node cutting and multi-division, and the method comprises the steps: 1) obtaining a Chinese and English knowledge graph and an English and Chinese knowledge graph; 2) utilizing the pre-training model to obtain semantic vector representation of the entity; 3) constructing subgraph pairs with similar structures based on node segmentation; 4) obtaining a structure vector representation of the entity by using a structure-based embedding model; 5) fusing the semantic vector representation and the structure vector representation to obtain fusion similarity; and 6) checking the alignment degree between the large-scale multi-language knowledge maps by using the entity alignment evaluation indexes. According to the method, the potential mapping rate of the sub-graph can be improved, so that the extendibility, accuracy and robustness of knowledge graph fusion can be improved, and powerful support is provided for sharing mechanisms in the fields of search engines, question-answering systems, recommendation systems, personalized services and the like.

Description

Technical Field

[0001] The present invention belongs to the field of graph partitioning based on large-scale entity alignment, and mainly relates to a graph partitioning method based on node segmentation and a multi-partition edge node enhancement method. Background Art

[0002] Knowledge graphs utilize entity and relationship concepts to provide a structured representation of the real world, a characteristic that underpins their importance in knowledge-driven applications. Due to the incompleteness of knowledge graphs, a single knowledge base often cannot meet knowledge needs. To address the language gap in multilingual knowledge bases, aligning entities across different language knowledge bases has become a critical task, a process known as entity alignment. Traditional methods often rely on global features, but as graphs scale, computing global features becomes increasingly difficult.

[0003] In order to solve the above problems, some scholars have redesigned the representation model by optimizing the representation learning model, thereby improving the accuracy of entity alignment on large-scale multilingual knowledge graphs. However, these models cannot be combined with semantic representation due to the limitation of computer performance, and therefore only represent the structural embedding of nodes. Other scholars have introduced local graphs to solve the problem of high computational complexity. However, the models that previously introduced graph partitioning methods will lead to sparsity problems in the edge node field, and some models have high computational complexity and cannot achieve alignment across subgraphs, resulting in a limited range of node alignment. Summary of the Invention

[0004] In order to address the shortcomings of the above-mentioned existing technologies, the present invention proposes a large-scale multilingual knowledge graph alignment method based on node cutting and multiple partitioning, in order to improve the potential mapping rate of subgraphs, thereby improving the scalability, accuracy and robustness of knowledge graph fusion, and providing strong support for sharing mechanisms in fields such as search engines and question-answering systems, recommendation systems and personalized services.

[0005] In order to solve the technical problem, the present invention adopts the following technical solution:

[0006] The large-scale multilingual knowledge graph alignment method based on node cutting and multiple partitioning of the present invention is characterized in that it is performed according to the following steps:

[0007] Step 1: Obtain a large-scale Chinese knowledge graph ,in, yes The Chinese entity set, yes The Chinese relation set of yes The set of Chinese triples, Any Chinese triple in ,in, For any Chinese head entity, is any Chinese suffix entity, Is any Chinese relationship;

[0008] Obtaining a large-scale English knowledge graph ,in, yes The English entity set of yes The English relation set of yes The set of English triples, Any English triple in ,in, is any English head entity, is any English suffix entity, Is any English relationship;

[0009] make Any query Chinese entity in is denoted as ,make Any English entity in ;

[0010] Step 2: Construct a large-scale multilingual knowledge graph alignment network, including: semantic vector representation module, structural similarity subgraph module, structural similarity fusion, and and Processing to obtain semantically similar and fusion structure similarity ;in, express The embedding vector of express Embedding vector of

[0011] Step 3: Calculate using formula (5) and The final similarity between , thus obtaining and The final similarities of all English entities in , and the English entity corresponding to the maximum final similarity is selected as the target English entity;

[0012] (5)

[0013] In formula (5), is the fusion weight;

[0014] Step 4: The target English entities corresponding to all query Chinese entities in the query are merged to obtain the aligned multilingual knowledge graph.

[0015] The large-scale multilingual knowledge graph alignment method based on node cutting and adaptive similarity fusion described in the present invention is also characterized in that step 2 includes:

[0016] Step 2.1: The semantic vector representation module uses formula (1) to and Process and obtain Embedding vector of , Embedding vector of , and then use formula (2) to calculate the semantic similarity ;

[0017] (1)

[0018] (2)

[0019] In formula (1) and formula (2), represents the pre-trained language model, Symbols that represent the semantics of entities; represents cosine similarity;

[0020] Step 2.2: The structural similarity subgraph module and Process and obtain and The cosine similarity between them is used to construct the subgraph structure loss function , thus Embedding vector of as well as Embedding vector of Learn and output the learned Structural embedding as well as Structural embedding ;

[0021] Step 2.3: The structural similarity fusion is based on as well as , and get the fusion structure similarity .

[0022] Furthermore, step 2.2 includes:

[0023] Step 2.2.1: Use node segmentation to and Divide and construct the Chinese subgraph set accordingly

[0024] and English subgraph collection ;in, Indicates the Chinese subgraphs, Indicates the English subgraphs; is the number of subgraph pairs;

[0025] Step 2.2.2: Get and The semantically aligned positive English entities are denoted as ;exist Get and Any negative English entity with semantic non-alignment is denoted as ,exist Search The Chinese subgraph , and randomly initialize The embedding vector of 、 The embedding vector of 、 The embedding vector of ; thus calculating and The cosine similarity between as well as and The cosine similarity between ;

[0026] Step 2.2.3: Use formula (2) to construct the subgraph structure loss function :

[0027] (2)

[0028] In formula (2), and are two hyperparameters, Exponential function, To minimize operations;

[0029] Step 2.2.4: Use gradient descent to 、 and Learn and calculate To update the module parameters until Until convergence, get the learning Structural embedding as well as Structural embedding .

[0030] Furthermore, it is characterized in that step 2.3 includes:

[0031] Step 2.3.1: Use formula (3) to get exist Above and exist Structural similarity on :

[0032] (3)

[0033] In formula (3), express and The cosine similarity between

[0034] Step 2.3.2: Use formula (4) to get and Fusion structural similarity on all subgraphs :

[0035] (4)

[0036] In formula (4), represents the indicator function, if exist Zhongqie exist In, then, is 1, otherwise, is 0.

[0037] Furthermore, the step 2.2.1 is performed as follows:

[0038] Step 2.2.1.1, as well as Add to processing queue In the example, the maximum number of nodes in the graph after sub-graph partitioning is set to ,exist Get and Any semantically aligned positive English entity is denoted as , thus forming a key-value pair ;

[0039] Step 2.2.1.2, from Take the top element of the stack and assign it to the Chinese picture and English pictures ; and extract The entity node collection in and The entity node collection in ; Re-extraction Query Chinese entity collection in as well as Zhongyu Semantically aligned English positive entity set ;

[0040] calculate exist The neighbor set on ;in, express The set of neighbors of

[0041] Will and Merge into an expanded node set ; thus constructing a joint graph ,in, represents the set of joint triples, and , Indicates that, express any Chinese triple; Indicates Any head node in Indicates Any tail node in express and relationship;

[0042] Step 2.2.1.3, set the number of subgraphs to k, and use the Metis algorithm to Divide and obtain the subgraph set after division ;in, represents the jth subgraph;

[0043] Step 2.2.1.4, Remap to Chinese map and English pictures , get the divided Chinese subgraph set and English subgraphs ,in, express Middle Chinese subgraphs, express Middle English subgraphs, and , , u represents a Chinese entity in any key-value pair, v represents an English entity in any key-value pair, Represents any key-value pair, L represents Get and A key-value pair set consisting of all the positive English entities that are semantically aligned with the query Chinese entities; ;

[0044] Step 2.2.1.5, and Perform iterative partitioning to obtain The set of nodes finally assigned to the jth subgraph and English pictures The set of nodes finally assigned to the jth subgraph ;

[0045] Step a: define the current number of iterations as m, initialize m=1, and define the maximum number of iterations as T;

[0046] Define and initialize the Chinese graph at the m-1th iteration The set of nodes assigned to the jth Chinese subgraph in ;

[0047] Define and initialize the English graph at the m-1th iteration The set of nodes assigned to the jth subgraph in ;

[0048] Define and initialize the Chinese graph at the m-1th iteration The set of unassigned nodes in ;

[0049] Define and initialize the English graph at the m-1th iteration The set of unassigned nodes in ;\ indicates removal operation;

[0050] Step b: Calculate the Chinese image at the m-1th iteration Diffusion boundary in the jth subgraph ;

[0051] Calculate the Chinese graph at the mth iteration The set of nodes assigned to the jth Chinese subgraph in ;

[0052] English graph when calculating the m-1th iteration Diffusion boundary in the jth subgraph ;

[0053] English diagram when calculating the mth iteration The set of nodes assigned to the jth subgraph in ;Thus calculating the Chinese graph at the mth iteration The set of unassigned nodes in and the English graph at the mth iteration The set of unassigned nodes in ;

[0054] Step c, after assigning m+1 to m, if m>T, then we get The set of nodes finally assigned to the jth subgraph and English pictures The set of nodes finally assigned to the jth subgraph Otherwise, return to step b and execute sequentially;

[0055] Step 2.2.1.6, if , then As the top element of the stack, return to step 2.2.1.2 and execute sequentially. Otherwise, Add to In Add to Thus, the final Chinese subgraph set is obtained and English subgraph collection ,in, Indicates the threshold value.

[0056] The electronic device of the present invention includes a memory and a processor, and is characterized in that the memory is used to store a program that supports the processor to execute the large-scale multilingual knowledge graph alignment method, and the processor is configured to execute the program stored in the memory.

[0057] The present invention provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, which is characterized in that when the computer program is run by a processor, it executes the steps of the large-scale multilingual knowledge graph alignment method.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] 1. To address the problem that traditional large-scale knowledge graph representation methods cannot effectively generate vector representations of semantic features, this paper uses the pre-trained BERT model to improve it. Based on the Transformer architecture and attention mechanism, this model learns rich language representation capabilities through unsupervised learning tasks on massive amounts of text data. Compared to traditional methods, BERT can more accurately capture the potential semantic information between entities in cross-lingual knowledge graphs, thereby significantly improving the accuracy of entity alignment.

[0060] 2. To address the limitations of traditional subgraph partitioning methods (such as edge partitioning), which cannot effectively achieve cross-subgraph entity alignment, this paper proposes a subgraph partitioning method based on node partitioning. Unlike edge partitioning, node partitioning allows edge nodes to appear in multiple subgraphs simultaneously, thereby expanding the matching range of edge nodes and significantly improving the effectiveness of large-scale cross-lingual entity alignment.

[0061] 3. By learning from a limited set of alignment seeds, this method can automatically align more entities in large-scale multilingual knowledge graphs, significantly improving the automation of alignment. This method is particularly suitable for handling sparse entity matching across subgraphs and domains, successfully overcoming the limitations of traditional methods. Compared to existing technologies, this method not only significantly improves alignment speed but also reduces reliance on manual annotation, providing an efficient and reliable solution for automated entity alignment in large-scale knowledge graphs. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 Flowchart of the method of the present invention. DETAILED DESCRIPTION

[0063] In this embodiment, a large-scale multilingual knowledge graph alignment method based on node cutting and multiple partitioning is adopted. In addition to considering the challenges of large-scale processing capabilities posed by knowledge graph scale and device computing power limitations, the model also considers the impact of node semantics on alignment capabilities, utilizes the principle of locality, and improves the potential alignment mapping rate by expanding the matching range of edge nodes. Specifically, Figure 1 As shown, the method is carried out in the following steps:

[0064] Step 1: Obtain a large-scale Chinese knowledge graph ,in, yes The Chinese entity set, yes The Chinese relation set of yes The set of Chinese triples, Any Chinese triple in ,in, For any Chinese head entity, is any Chinese suffix entity, Is any Chinese relationship;

[0065] Obtaining a large-scale English knowledge graph ,in, yes The English entity set of yes The English relation set of yes The set of English triples, Any English triple in ,in, is any English head entity, is any English suffix entity, Is any English relationship;

[0066] make Any query Chinese entity in is denoted as ,make Any English entity in .

[0067] Step 2: Construct a large-scale multilingual knowledge graph alignment network, including: semantic vector representation module, structural similarity subgraph module, structural similarity fusion, and and Processing to obtain semantically similar and fusion structure similarity ;

[0068] Step 2.1: The semantic vector representation module uses formula (1) to and Process and obtain Embedding vector of , Embedding vector of , and then use formula (2) to calculate the semantic similarity .

[0069] (1)

[0070] (2)

[0071] In formula (1) and formula (2), represents the pre-trained language model, Symbols that represent the semantics of entities; Represents cosine similarity.

[0072] Step 2.2: Structural similarity subgraph module pair and Processing is performed to obtain cosine similarity, which is used to construct the subgraph structure loss function , thus Embedding vector of as well as Embedding vector of Learn and output the learned Structural embedding as well as Structural embedding .

[0073] Step 2.2.1.1, as well as Add to processing queue In the example, the maximum number of nodes in the graph after sub-graph partitioning is set to ,exist Get and Any semantically aligned positive English entity is denoted as , thus forming a key-value pair ;

[0074] Step 2.2.1.2, from Take the top element of the stack and assign it to the Chinese picture and English pictures ; and extract The entity node collection in and The entity node collection in ; Re-extraction Query Chinese entity collection in as well as Zhongyu Semantically aligned English positive entity set ;

[0075] calculate exist The neighbor set on ;in, express The set of neighbors of

[0076] Will and Merge into an expanded node set ; thus constructing a joint graph ,in, represents the set of joint triples, and , Indicates that, express any Chinese triple; Indicates Any head node in Indicates Any tail node in express and relationship;

[0077] Step 2.2.1.3, set the number of subgraphs to k, and use the Metis algorithm to Divide and obtain the subgraph set after division ;in, represents the jth subgraph;

[0078] Step 2.2.1.4, Remap to Chinese map and English pictures , get the divided Chinese subgraph set and English subgraphs ,in, express Middle Chinese subgraphs, express Middle English subgraphs, and , , u represents a Chinese entity in any key-value pair, v represents an English entity in any key-value pair, Represents any key-value pair, L represents Get and A key-value pair set consisting of all the positive English entities that are semantically aligned with the query Chinese entities; ;

[0079] Step 2.2.1.5, and Perform iterative partitioning to obtain The set of nodes finally assigned to the jth subgraph and English pictures The set of nodes finally assigned to the jth subgraph ;

[0080] Step a: define the current number of iterations as m, initialize m=1, and define the maximum number of iterations as T;

[0081] Define and initialize the Chinese graph at the m-1th iteration The set of nodes assigned to the jth Chinese subgraph in ;

[0082] Define and initialize the English graph at the m-1th iteration The set of nodes assigned to the jth subgraph in ;

[0083] Define and initialize the Chinese graph at the m-1th iteration The set of unassigned nodes in ;

[0084] Define and initialize the English graph at the m-1th iteration The set of unassigned nodes in ;\ indicates removal operation;

[0085] Step b: Calculate the Chinese image at the m-1th iteration Diffusion boundary in the jth subgraph ;

[0086] Calculate the Chinese graph at the mth iteration The set of nodes assigned to the jth Chinese subgraph in ;

[0087] English graph when calculating the m-1th iteration Diffusion boundary in the jth subgraph ;

[0088] English diagram when calculating the mth iteration The set of nodes assigned to the jth subgraph in ;Thus calculating the Chinese graph at the mth iteration The set of unassigned nodes in and the English graph at the mth iteration The set of unassigned nodes in ;

[0089] Step c, after assigning m+1 to m, if m>T, then we get The set of nodes finally assigned to the jth subgraph and English pictures The set of nodes finally assigned to the jth subgraph Otherwise, return to step b and execute sequentially;

[0090] Step 2.2.1.6, if , then As the top element of the stack, return to step 2.2.1.2 and execute sequentially. Otherwise, Add to In Add to Thus, the final Chinese subgraph set is obtained and English subgraph collection ,in, Indicates the threshold value.

[0091] Step 2.2.2: Get and The semantically aligned positive English entities are denoted as ;exist Get and Any negative English entity with semantic non-alignment is denoted as ,exist Search The Chinese subgraph , and randomly initialize The embedding vector of 、 The embedding vector of 、 The embedding vector of ; thus calculating and The cosine similarity between as well as and The cosine similarity between ;

[0092] Step 2.2.3: Use formula (2) to construct the subgraph structure loss function :

[0093] (2)

[0094] In formula (2), and are two hyperparameters, Exponential function, To minimize operations;

[0095] Step 2.2.4: Use gradient descent to 、 and Learn and calculate To update the module parameters until Until convergence, get the learning Structural embedding as well as Structural embedding .

[0096] Step 2.3: Structural similarity fusion based on as well as , and get the fusion structure similarity .

[0097] Step 2.3.1: Use formula (3) to get exist Above and exist Structural similarity on :

[0098] (3)

[0099] In formula (3), express and The cosine similarity between

[0100] Step 2.3.2: Use formula (4) to get and Fusion structural similarity on all subgraphs :

[0101] (4)

[0102] In formula (4), represents the indicator function. If is in and is in , then is 1; otherwise, is 0.

[0103] For example, for the entity (anhui) in the English graph, it exists in two sub - graphs (hfut - anhui) and (anhui - china). The present invention fuses the similarities of the entity (anhui) in different sub - graphs.

[0104] Step 3: Calculate the final similarity between and using formula (5), so as to obtain and the final similarities of all English entities in

[0105] (5)

[0106] In formula (5), is the fusion weight;

[0107] Step 4: Merge all the target English entities corresponding to the query Chinese entities in to obtain the aligned multilingual knowledge graph.

[0108] In this embodiment, an electronic device includes a memory and a processor. The memory is used to store a program that supports the processor to execute the above - mentioned method, and the processor is configured to execute the program stored in the memory.

[0109] In this embodiment, a computer - readable storage medium stores a computer program, and when the computer program is run by a processor, it executes the steps of the above - mentioned method.

Claims

1. A large-scale multilingual knowledge graph alignment method based on node cutting and multiple partitioning, characterized by: The steps are as follows: Step 1: Obtain a large-scale Chinese knowledge graph ,in, yes The Chinese entity set, yes The Chinese relation set of yes The set of Chinese triples, Any Chinese triple in ,in, For any Chinese head entity, is any Chinese suffix entity, Is any Chinese relationship; Obtaining a large-scale English knowledge graph ,in, yes The English entity set of yes The English relation set of yes The set of English triples, Any English triple in ,in, is any English head entity, is any English suffix entity, Is any English relationship; make Any query Chinese entity in is denoted as ,make Any English entity in ; Step 2: Construct a large-scale multilingual knowledge graph alignment network, including: semantic vector representation module, structural similarity subgraph module, structural similarity fusion, and and Processing to obtain semantically similar and fusion structure similarity ;in, express The embedding vector of express Embedding vector of Step 3: Calculate using formula (5) and The final similarity between , thus obtaining and The final similarities of all English entities in , and the English entity corresponding to the maximum final similarity is selected as the target English entity; (5) In formula (5), is the fusion weight; Step 4: The target English entities corresponding to all query Chinese entities in the query are merged to obtain the aligned multilingual knowledge graph.

2. A large-scale multilingual knowledge graph alignment method based on node cutting and adaptive similarity fusion according to claim 1, characterized in that: The step 2 includes: Step 2.1: The semantic vector representation module uses formula (1) to and Process and obtain Embedding vector of , Embedding vector of , and then use formula (2) to calculate the semantic similarity ; (1) (2) In formula (1) and formula (2), represents the pre-trained language model, Symbols that represent the semantics of entities; represents cosine similarity; Step 2.2: The structural similarity subgraph module and Process and obtain and The cosine similarity between them is used to construct the subgraph structure loss function , thus Embedding vector of as well as Embedding vector of Learn and output the learned Structural embedding as well as Structural embedding ; Step 2.3: The structural similarity fusion is based on as well as , and get the fusion structure similarity .

3. The large-scale multilingual knowledge graph alignment method based on node cutting and adaptive similarity fusion according to claim 2 is characterized in that: Step 2.2 includes: Step 2.2.1: Use node segmentation to and Divide and construct the Chinese subgraph set accordingly and English subgraph collection ;in, Indicates the Chinese subgraphs, Indicates the English subgraphs; is the number of subgraph pairs; Step 2.2.2: Get and The semantically aligned positive English entities are denoted as ;exist Get and Any negative English entity with semantic non-alignment is denoted as ,exist Search The Chinese subgraph , and randomly initialize The embedding vector of 、 The embedding vector of 、 The embedding vector of ; thus calculating and The cosine similarity between as well as and The cosine similarity between ; Step 2.2.3: Use formula (2) to construct the subgraph structure loss function : (2) In formula (2), and are two hyperparameters, Exponential function, To minimize operations; Step 2.2.4: Use gradient descent to 、 and Learn and calculate To update the module parameters until Until convergence, get the learning Structural embedding as well as Structural embedding .

4. The large-scale multilingual knowledge graph alignment method based on node cutting and adaptive similarity fusion according to claim 3 is characterized in that: Step 2.3 includes: Step 2.3.1: Use formula (3) to get exist Above and exist Structural similarity on : (3) In formula (3), express and The cosine similarity between Step 2.3.2: Use formula (4) to get and Fusion structural similarity on all subgraphs : (4) In formula (4), represents the indicator function, if exist Zhongqie exist In, then, is 1, otherwise, is 0.

5. The large-scale multilingual knowledge graph alignment method based on node cutting and adaptive similarity fusion according to claim 1 is characterized in that: Step 2.2.1 is performed as follows: Step 2.2.1.1, as well as Add to processing queue In the example, the maximum number of nodes in the graph after sub-graph partitioning is set to ,exist Get and Any semantically aligned positive English entity is denoted as , thus forming a key-value pair ; Step 2.2.1.2, from Take the top element of the stack and assign it to the Chinese picture and English pictures ; and extract The entity node collection in and The entity node collection in ; Re-extraction Query Chinese entity collection in as well as Zhongyu Semantically aligned English positive entity set ; calculate exist The neighbor set on ;in, express The set of neighbors of Will and Merge into an expanded node set ; thus constructing a joint graph ,in, represents the set of joint triples, and , Indicates that, express any Chinese triple; Indicates Any head node in Indicates Any tail node in express and relationship; Step 2.2.1.3, set the number of subgraphs to k, and use the Metis algorithm to Divide and obtain the subgraph set after division ;in, represents the jth subgraph; Step 2.2.1.4, Remap to Chinese map and English pictures , get the divided Chinese subgraph set and English subgraphs ,in, express Middle Chinese subgraphs, express Middle English subgraphs, and , , u represents a Chinese entity in any key-value pair, v represents an English entity in any key-value pair, Represents any key-value pair, L represents Get and A key-value pair set consisting of all the positive English entities that are semantically aligned with the query Chinese entities; ; Step 2.2.1.5, and Perform iterative partitioning to obtain The set of nodes finally assigned to the jth subgraph and English pictures The set of nodes finally assigned to the jth subgraph ; Step a: define the current number of iterations as m, initialize m=1, and define the maximum number of iterations as T; Define and initialize the Chinese graph at the m-1th iteration The set of nodes assigned to the jth Chinese subgraph in ; Define and initialize the English graph at the m-1th iteration The set of nodes assigned to the jth subgraph in ; Define and initialize the Chinese graph at the m-1th iteration The set of unassigned nodes in ; Define and initialize the English graph at the m-1th iteration The set of unassigned nodes in ;\ indicates removal operation; Step b: Calculate the Chinese image at the m-1th iteration Diffusion boundary in the jth subgraph ; Calculate the Chinese graph at the mth iteration The set of nodes assigned to the jth Chinese subgraph in ; English graph when calculating the m-1th iteration Diffusion boundary in the jth subgraph ; English diagram when calculating the mth iteration The set of nodes assigned to the jth subgraph in ;Thus calculating the Chinese graph at the mth iteration The set of unassigned nodes in and the English graph at the mth iteration The set of unassigned nodes in ; Step c, after assigning m+1 to m, if m>T, then we get The set of nodes finally assigned to the jth subgraph and English pictures The set of nodes finally assigned to the jth subgraph Otherwise, return to step b and execute sequentially; Step 2.2.1.6, if , then As the top element of the stack, return to step 2.2.1.2 and execute it sequentially. Otherwise, Add to In Add to Thus, the final Chinese subgraph set is obtained and English subgraph collection ,in, Indicates the threshold value.

6. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the large-scale multilingual knowledge graph alignment method described in any one of claims 1 to 5, and the processor is configured to execute the program stored in the memory.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the large-scale multilingual knowledge graph alignment method according to any one of claims 1 to 5 are performed.

Images