A multi-dimensional feature-based incremental knowledge graph entity alignment method and device

Abstract

The application provides a kind of based on the incremental knowledge graph entity alignment method and device of multi-dimensional feature, the steps of the method include: for the entity in first knowledge graph and second knowledge graph, attribute vector is constructed;First similarity matrix is calculated based on attribute vector;For the neighbor entity of each entity of first knowledge graph and second knowledge graph, the number of entity that the entity of first knowledge graph and the neighbor entity of the entity of second knowledge graph are mutually aligned is calculated, and the matrix of aligned neighbor entity number is constructed;Second similarity matrix is calculated based on first similarity matrix and aligned neighbor entity number matrix;Information quantity parameter is calculated based on the entity information of entity pair, and the pre-screening comparison value of entity pair is calculated based on second similarity and information quantity parameter, to obtain entity pair screening set;For each entity pair in entity pair screening set, the third similarity of two entities is calculated, and the alignment result of the two entities of the entity pair is determined based on the third similarity.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph technology, and in particular to an incremental knowledge graph entity alignment method and apparatus based on multi-dimensional features. Background Technology

[0002] Knowledge graphs, based on core elements such as entities, attributes, and relationships, achieve efficient representation of real-world knowledge and have been widely applied in artificial intelligence fields such as intelligent question answering and semantic search. However, single knowledge graphs are often limited by domain scope and data sources, resulting in problems such as sparse knowledge coverage and limited information dimensions, making it difficult to meet the needs of complex tasks. Therefore, fusing multi-source knowledge graphs has become crucial for improving the knowledge coverage of graphs. Entity alignment technology identifies entities in different graphs that point to the same object in the real world, thereby linking and fusing multi-source knowledge graphs to form a knowledge graph with broader coverage and richer semantics, providing more comprehensive and reliable knowledge support for upper-layer intelligent applications.

[0003] Existing entity alignment methods mainly utilize the attribute and relation information of entities in knowledge graphs, map entities to a low-dimensional embedding space based on graph neural network models, and use pre-aligned seed entities as supervision signals to train the model, so that corresponding entities are close to each other in the embedding vector space, thereby performing entity alignment based on vector relevance.

[0004] However, existing entity alignment methods do not make full use of entities with high information content as heuristic information for incremental alignment during the alignment process, and they do not make sufficient use of multidimensional correlation features of cross-graph entities, resulting in low accuracy of the final entity alignment and poor robustness when graph information is missing. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide an incremental knowledge graph entity alignment method based on multidimensional features to eliminate or improve one or more defects existing in the prior art.

[0006] One aspect of the present invention provides an incremental knowledge graph entity alignment method based on multi-dimensional features, the method comprising the following steps: Obtain the first and second knowledge graphs to be aligned, and construct attribute vectors for entities in the first and second knowledge graphs; Based on the attribute vectors, calculate the first similarity between each entity in the first knowledge graph and each entity in the second knowledge graph, and construct a first similarity matrix; Based on the neighboring entities of each entity in the first knowledge graph and the second knowledge graph, determine the number of aligned entities between the neighboring entities of each entity in the first knowledge graph and the neighboring entities of each entity in the second knowledge graph, and construct a matrix of aligned neighboring entities. A second similarity matrix is ​​calculated based on the first similarity matrix and the aligned neighbor entity number matrix. Each position in the second similarity matrix corresponds to an entity pair constructed from entities in the first knowledge graph and entities in the second knowledge graph. The information content parameter is calculated based on the entity information contained in each entity pair. The pre-screening comparison value of each entity pair is calculated based on the second similarity and information content parameter at each position in the second similarity matrix. The entity pairs are then screened based on the pre-screening comparison value to obtain the entity pair screening set. For each entity pair in the entity pair filtering set, calculate the multiple similarity between the two entities, perform a weighted calculation on the multiple similarity to obtain a third similarity, and determine the entity alignment of the two entities in the entity pair based on the third similarity.

[0007] The above scheme first calculates the first similarity based on the entity's attributes and constructs a first similarity matrix. Then, based on the alignment status of each entity's neighboring nodes, it constructs a matrix of aligned neighboring entities. Combining the above two types of information, it constructs a second similarity matrix. Next, it calculates the information content parameter based on the information contained in each entity in the entity pair. Combining the information content parameter and the second similarity matrix, it calculates the pre-screening comparison value. In the process of calculating the pre-screening comparison value, it fully utilizes various information of the entities and calculates the combination information of the two entities. By filtering entity pairs through the pre-screening comparison value, it can make full use of entities with higher information content as heuristic information for incremental alignment. In subsequent steps, a third similarity is calculated on the selected entity pairs to determine whether to align the two entities, ensuring the accuracy of entity alignment.

[0008] In some embodiments of the present invention, in the step of determining the number of aligned entities of each entity in the first knowledge graph and each entity in the second knowledge graph based on the neighbor entities of each entity in the first knowledge graph and the second knowledge graph, and constructing an aligned neighbor entity number matrix, the number of entities whose neighbor entities of the first knowledge graph and the second knowledge graph are aligned with each other is calculated as the matrix value in the aligned neighbor entity number matrix.

[0009] In some embodiments of the present invention, in the step of calculating the information content parameter based on the entity information contained in each entity pair, for the two entities of the entity pair, the number of non-empty attributes of the entity, the number of neighboring entities, and the number of aligned entities among the neighboring entities are counted respectively. The sum of the number of non-empty attributes, the number of neighboring entities, and the number of aligned entities among the neighboring entities is calculated as the entity information content. The sum of the entity information content of the two entities of the entity pair is calculated as the information content parameter.

[0010] In some embodiments of the present invention, in the step of calculating the pre-screening comparison value of each entity pair based on the second similarity and information content parameters at each position in the second similarity matrix, the pre-screening comparison value is calculated using the following formula: in, Represents entities in the first knowledge graph Entities in the second knowledge graph The pre-screened comparison values ​​of the constructed entity pairs. Represents entity pairs The second similarity in the second similarity matrix, Represents entity pairs Information content parameters and These represent the weight values ​​corresponding to the second similarity and information content parameters, respectively.

[0011] In some embodiments of the present invention, the method further includes the following steps: The entity information of entity pairs in the entity pair filtering set is used to construct a first prompt word. The first prompt word is then input into a preset large language model to update the first knowledge graph and the second knowledge graph.

[0012] In some embodiments of the present invention, multiple similarity includes large model inference similarity. In the step of calculating multiple similarity between two entities for each entity pair in the entity pair filtering set, the entity information of the entity pair in the entity pair filtering set is used to construct a second prompt word, and the second prompt word is input into a preset large language model to obtain the large model inference similarity.

[0013] In some embodiments of the present invention, multiple similarity includes embedding vector similarity. In the step of calculating multiple similarity between two entities for each entity pair in the entity pair screening set, the first knowledge graph and the second knowledge graph are input into a graph neural network model. The graph neural network model outputs embedding vectors corresponding to the first knowledge graph and the second knowledge graph. For each entity pair in the entity pair screening set, the embedding vector similarity between the two entities is calculated based on the corresponding embedding vector.

[0014] In some embodiments of the present invention, multiple similarity includes local structural similarity. In the step of calculating multiple similarity between two entities for each entity pair in the entity pair screening set, for each entity pair in the entity pair screening set, the number of mutually aligned neighboring entities of the entity in the first knowledge graph of the entity pair and the neighboring entities of the entity in the second knowledge graph of the entity pair is calculated as local structural similarity.

[0015] In some embodiments of the present invention, multiple similarity includes global structural similarity. In the step of calculating the multiple similarity between two entities for each entity pair in the entity pair filtering set, the shortest path length and the total number of paths between the two entities in the entity pair are calculated, and the global structural similarity is calculated based on the shortest path length and the total number of paths. in, Represents entities in the first knowledge graph Entities in the second knowledge graph The global structural similarity of the constructed entity pairs. Represents entity pairs The shortest path length, Represents entity pairs The total number of paths.

[0016] A second aspect of the present invention also provides an incremental knowledge graph entity alignment device based on multidimensional features. The device includes a computer device, which includes a processor and a memory. The memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the device implements the steps of the method described above.

[0017] A third aspect of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned incremental knowledge graph entity alignment method based on multidimensional features.

[0018] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the text, or may be learned by practice of the invention. The objects and other advantages of the invention will become apparent from the description and the accompanying drawings.

[0019] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description. Attached Figure Description

[0020] The accompanying drawings, which are provided to further illustrate the invention and form part of this application, are not intended to limit the scope of the invention.

[0021] Figure 1 This is a schematic diagram illustrating one implementation of the incremental knowledge graph entity alignment method based on multi-dimensional features in this scheme. Figure 2 This is a schematic diagram of the overall processing architecture of this solution. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0023] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0024] Existing entity alignment techniques have the following main drawbacks: (1) The entities with high information content were not fully utilized as heuristic information for incremental alignment during the alignment process. (2) Only existing attributes and relational features in the graph can be used for entity alignment, lacking the supplementation of external knowledge and common sense. (3) The multidimensional features cannot be effectively used to judge entity similarity, resulting in poor robustness of the existing technology when graph information is missing.

[0025] like Figure 1 As shown, this invention proposes an incremental knowledge graph entity alignment method based on multi-dimensional features. The steps of the method include: Step S100: Obtain the first knowledge graph and the second knowledge graph to be aligned, and construct attribute vectors for the entities in the first knowledge graph and the second knowledge graph. In the specific implementation process, a pre-trained model is used to vectorize all entity attributes in the two graphs to obtain the representation vector of each entity in the two graphs. Specifically, the pre-trained model can be a BERT model.

[0026] Step S200: Calculate the first similarity between each entity in the first knowledge graph and each entity in the second knowledge graph based on the attribute vector, and construct a first similarity matrix. In practical implementation, the first similarity can be calculated using cosine similarity, and the first similarity can be expressed as: in, Represents entities in the first knowledge graph Entities in the second knowledge graph The first similarity between them.

[0027] Step S300: Based on the neighbor entities of each entity in the first knowledge graph and the second knowledge graph, determine the number of aligned entities between the neighbor entities of each entity in the first knowledge graph and the neighbor entities of each entity in the second knowledge graph, and construct an aligned neighbor entity number matrix. In practice, the aligned entities can be entities aligned based on their attribute information, such as the correspondence between account numbers and ID cards.

[0028] In the specific implementation process, the number of common neighbors of cross-graph entity pairs is calculated, and the number of aligned nodes contained in the common neighbors is further counted. The matrix of aligned neighbor entities of cross-graph entity pairs can be represented as follows: in, Represents entities in the first knowledge graph Entities in the second knowledge graph The number of aligned neighbor entities between them.

[0029] Specifically, in this scheme, all neighboring entities are one-hop distance neighboring entities. If the entity Including neighboring entities a1, a2, a3, and a4, the entities Including neighboring entities b1, b2, b3, and b4, where a1 is aligned with b3 and a4 is aligned with b2, then there exist two pairs of aligned entities. The value is 2.

[0030] Step S400: Calculate a second similarity matrix based on the first similarity matrix and the aligned neighbor entity number matrix. Each position in the second similarity matrix corresponds to an entity pair constructed from entities in the first knowledge graph and entities in the second knowledge graph. In the specific implementation process, in the step of calculating the second similarity matrix based on the first similarity matrix and the aligned neighbor entity number matrix, the first similarity matrix and the aligned neighbor entity number matrix are multiplied by a dot to obtain the second similarity matrix used for candidate entity screening. : Step S500: Calculate the information content parameter based on the entity information contained in each entity pair; calculate the pre-screening comparison value of each entity pair based on the second similarity and information content parameter at each position in the second similarity matrix; and filter the entity pairs based on the pre-screening comparison value to obtain the entity pair screening set. Step S600: For each entity pair in the entity pair filtering set, calculate the multiple similarity between the two entities, perform a weighted calculation on the multiple similarity to obtain a third similarity, and determine the entity alignment of the two entities in the entity pair based on the third similarity.

[0031] The above scheme first calculates the first similarity based on the entity's attributes and constructs a first similarity matrix. Then, based on the alignment status of each entity's neighboring nodes, it constructs a matrix of aligned neighboring entities. Combining the above two types of information, it constructs a second similarity matrix. Next, it calculates the information content parameter based on the information contained in each entity in the entity pair. Combining the information content parameter and the second similarity matrix, it calculates the pre-screening comparison value. In the process of calculating the pre-screening comparison value, it fully utilizes various information of the entities and calculates the combination information of the two entities. By filtering entity pairs through the pre-screening comparison value, it can make full use of entities with higher information content as heuristic information for incremental alignment. In subsequent steps, a third similarity is calculated on the selected entity pairs to determine whether to align the two entities, ensuring the accuracy of entity alignment.

[0032] In some embodiments of the present invention, in the step of determining the number of aligned entities of each entity in the first knowledge graph and each entity in the second knowledge graph based on the neighbor entities of each entity in the first knowledge graph and the second knowledge graph, and constructing an aligned neighbor entity number matrix, the number of entities whose neighbor entities of the first knowledge graph and the second knowledge graph are aligned with each other is calculated as the matrix value in the aligned neighbor entity number matrix.

[0033] In some embodiments of the present invention, in the step of calculating the information content parameter based on the entity information contained in each entity pair, for the two entities of the entity pair, the number of non-empty attributes of the entity, the number of neighboring entities, and the number of aligned entities among the neighboring entities are counted respectively. The sum of the number of non-empty attributes, the number of neighboring entities, and the number of aligned entities among the neighboring entities is calculated as the entity information content. The sum of the entity information content of the two entities of the entity pair is calculated as the information content parameter.

[0034] In the specific implementation process, for each entity in the entity pair, the number of non-empty attributes of each entity is counted. And count the number of its neighboring entities. and the number of aligned entities among neighboring entities. The sum of these amounts gives the entity information. The information content of an entity pair is the sum of the information content of the two entities. .

[0035] In the specific implementation process, if the entity exist Number of neighboring entities Among the number of neighboring entities, there are If an entity is aligned with other entities, then the number of neighboring entities is . The number of aligned entities among the neighboring entities is .

[0036] In some embodiments of the present invention, in the step of calculating the pre-screening comparison value of each entity pair based on the second similarity and information content parameters at each position in the second similarity matrix, the pre-screening comparison value is calculated using the following formula: in, Represents entities in the first knowledge graph Entities in the second knowledge graph The pre-screened comparison values ​​of the constructed entity pairs. Represents entity pairs The second similarity in the second similarity matrix, Represents entity pairs Information content parameters and These represent the weight values ​​corresponding to the second similarity and information content parameters, respectively.

[0037] In the specific implementation process, in the step of calculating the pre-screening comparison value of each entity pair based on the second similarity and information content parameters at each position in the second similarity matrix, and filtering entity pairs based on the pre-screening comparison values ​​to obtain the entity pair screening set, the entity pairs with the largest preset number of pre-screening comparison values ​​are selected as the entity pair screening set. For the entity pairs in the entity pair screening set, the next step of graph reasoning and completion is performed.

[0038] In some embodiments of the present invention, the method further includes the following steps: The entity information of entity pairs in the entity pair filtering set is used to construct a first prompt word. The first prompt word is then input into a preset large language model to update the first knowledge graph and the second knowledge graph.

[0039] In some embodiments of the present invention, if the entity information of entity pairs in the entity pair filtering set is used to construct a first prompt word, and the first prompt word is input into a preset large language model to update the first knowledge graph and the second knowledge graph, then in step S600, the third similarity is calculated by updating the first knowledge graph and the second knowledge graph.

[0040] In the specific implementation process, in the step of constructing the first prompt word from the entity information of the entity pairs in the entity pair screening set, the candidate entity pair information in the knowledge graph and the enhanced text retrieved from the outside are comprehensively utilized, and the large model is used to reason about the missing attributes of the candidate entities and their one-hop neighbors, the intra-graph relations and inter-graph relations.

[0041] The design for the first prompt word in the large model is as follows: Knowledge Graph 1 contains entities. Its attributes include [att1, att2, ..., attm], and its neighboring entities include [e11, e12, ..., e1n]. Knowledge Graph 2 contains entities. Its attributes include [at1, at2, ..., atp], and its one-hop neighbor entities include [e21, e22, ..., e2n]. Entities already aligned among the one-hop neighbors include [ea1, ea2, ..., eaw]. Entities are inferred using enhanced contextual information. and entity Missing attributes and relationships, and entities and its one-hop neighbors and entities It also includes other relationships between the neighbor and its one-hop neighbor, and provides the confidence of the completed attributes and relationships.

[0042] In some embodiments of the present invention, multiple similarity includes large model inference similarity. In the step of calculating multiple similarity between two entities for each entity pair in the entity pair filtering set, the entity information of the entity pair in the entity pair filtering set is used to construct a second prompt word, and the second prompt word is input into a preset large language model to obtain the large model inference similarity.

[0043] In some embodiments of the present invention, the step of constructing a second prompt word from the entity information of entity pairs in the entity pair filtering set is exemplified as follows: The first knowledge graph contains entities. Its attributes include One-hop neighbor entities include The second knowledge graph contains entities. Its attributes include One-hop neighbor entities include The entities that are already aligned in the one-hop neighbor list include... Determine the entity and entity Probability of being the same entity , The range of values ​​is .

[0044] The above scheme employs an incremental alignment approach. First, it aligns similar entity pairs with abundant feature information. Then, using the aligned entity information, it incrementally aligns subsequent entity pairs. Large-scale model retrieval enhancement is used to complete missing attributes and relationships in the graph. Entity alignment is achieved by comprehensively utilizing large-scale model inference information features, node attribute vector features, local structural features, and global structural features.

[0045] In some embodiments of the present invention, multiple similarity includes embedding vector similarity. In the step of calculating multiple similarity between two entities for each entity pair in the entity pair screening set, the first knowledge graph and the second knowledge graph are input into a graph neural network model. The graph neural network model outputs embedding vectors corresponding to the first knowledge graph and the second knowledge graph. For each entity pair in the entity pair screening set, the embedding vector similarity between the two entities is calculated based on the corresponding embedding vector.

[0046] In some embodiments of the present invention, in the step of inputting the first knowledge graph and the second knowledge graph into the graph neural network model, an initial embedding vector is constructed for each entity in the first knowledge graph and the second knowledge graph using the BERT model.

[0047] Using aligned entities and graph structure as supervision information, the embedding vectors of entity pairs are learned and updated based on the graph neural network model. The loss function of the graph neural network model is defined as follows: in, The loss for hierarchical reconstruction of the first knowledge graph. The graph neural network model represents the first... The degree matrix is ​​constructed from the output of the first knowledge graph corresponding to each layer. The graph neural network model represents the first... The adjacency matrix is ​​constructed from the output of the first knowledge graph corresponding to each layer. The graph neural network model represents the first... The adjacency matrix is ​​constructed from the output of the second knowledge graph corresponding to the layer. The graph neural network model represents the first... The degree matrix is ​​constructed from the output of the second knowledge graph corresponding to the layer. The graph neural network model represents the first... The corresponding output of the first knowledge graph for each layer. express transpose, The graph neural network model represents the first... The output of the corresponding second knowledge graph of the layer, express transpose, The loss for hierarchical reconstruction of the second knowledge graph. Let be the embedding vector matrix of the nodes in the second knowledge graph. This represents a subset constructed from aligned entities. Represents entities in the first knowledge graph within a subset. Entities in the second knowledge graph The constructed entity pairs Representing entities respectively and entity The corresponding embedding vector is the final output of the graph neural network model.

[0048] In some embodiments of the present invention, multiple similarity includes local structural similarity. In the step of calculating multiple similarity between two entities for each entity pair in the entity pair screening set, for each entity pair in the entity pair screening set, the number of mutually aligned neighboring entities of the entity in the first knowledge graph of the entity pair and the neighboring entities of the entity in the second knowledge graph of the entity pair is calculated as local structural similarity.

[0049] In some embodiments of the present invention, multiple similarity includes global structural similarity. In the step of calculating the multiple similarity between two entities for each entity pair in the entity pair filtering set, the shortest path length and the total number of paths between the two entities in the entity pair are calculated, and the global structural similarity is calculated based on the shortest path length and the total number of paths. in, Represents entities in the first knowledge graph Entities in the second knowledge graph The global structural similarity of the constructed entity pairs. Represents entity pairs The shortest path length, Represents entity pairs The total number of paths.

[0050] In some embodiments of the present invention, if the multiple similarity includes large model inference similarity, embedding vector similarity, local structural similarity, and global structural similarity, then in the step of calculating the multiple similarity between two entities for each entity pair in the entity pair selection set, the similarity value of each dimension is normalized and weighted summed using the following formula: + + in, , , and These represent the large model inference similarity, embedding vector similarity, local structural similarity, and global structural similarity, respectively. , , and These represent the weight values ​​for large model inference similarity, embedding vector similarity, local structural similarity, and global structural similarity, respectively. Indicates the third similarity. + + + =1.

[0051] In the specific implementation process, a similarity threshold is set. ,when If the entity pairs correspond to the same entity, then entity alignment is performed.

[0052] like Figure 2 As shown, in the specific implementation process, if no new alignment entity pairs are added in the current round, the alignment process ends. If an alignment entity pair exists in the current round, the next iteration begins, and each iteration performs steps S100 to S600.

[0053] In summary, this scheme first selects candidate entity pairs with high similarity and abundant information for alignment, using the high-quality aligned entity pairs as heuristic information for subsequent calculations. Then, it utilizes a large-scale model retrieval enhancement method to infer and complete the missing factual information in the knowledge graph. Finally, it comprehensively determines the similarity of entities in two knowledge graphs by integrating the features of large-scale model inference information, node attribute vector features, local structural features, and global structural features. This scheme effectively improves the accuracy of entity alignment and exhibits strong robustness even in the presence of noise and missing information in the graph data.

[0054] The beneficial effects of this plan include: 1. This scheme prioritizes the alignment of candidate entity pairs with high similarity and abundant information, and then uses the aligned entity information as heuristic information for the next round of incremental alignment, which can effectively improve the accuracy and efficiency of alignment.

[0055] 2. This solution utilizes the reasoning capabilities of large models and external knowledge bases to supplement missing attributes and relationships in the knowledge graph, effectively improving its completeness.

[0056] 3. This scheme comprehensively utilizes multi-dimensional features such as large model inference similarity, embedding vector similarity, local structural similarity, and global structural similarity to determine the correlation between two entities. This can avoid inference errors caused by the large model illusion problem, effectively improve the accuracy of entity alignment, and has strong robustness in the presence of noise and missing data in the graph data.

[0057] 4. This scheme uses entity similarity and information content to filter candidate entity pairs; then it uses a large model to complete the attributes of candidate entity pairs and the relationships within and across the graph; then it comprehensively uses the large model's reasoning similarity, entity embedding vector similarity, local structural similarity and global structural similarity to comprehensively determine the relevance of candidate entity pairs, and determines those with a relevance greater than a certain threshold as the same entity and adds them to the aligned entity set.

[0058] 5. This scheme comprehensively utilizes the information content of entities and the similarity of entity pairs to filter out entities that are prioritized for alignment.

[0059] 6. This solution is based on a large model to design prompts and search enhancement methods to complete entity attributes, intra-graph relationships, and cross-graph relationships.

[0060] 7. This scheme comprehensively utilizes large model inference similarity, embedding vector similarity, local structural similarity, and global network structural similarity to determine the similarity between entities to be aligned.

[0061] This invention also provides an incremental knowledge graph entity alignment device based on multi-dimensional features. The device includes a computer device, which includes a processor and a memory. The memory stores computer instructions, and the processor executes the computer instructions stored in the memory. When the computer instructions are executed by the processor, the device implements the steps of the method described above.

[0062] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the aforementioned incremental knowledge graph entity alignment method based on multi-dimensional features. The computer-readable storage medium can be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disks, hard disks, removable storage disks, CD-ROMs, or any other form of storage medium known in the art.

[0063] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.

[0064] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0065] In this invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations of the embodiments of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An incremental knowledge graph entity alignment method based on multi-dimensional features, characterized in that, The steps of the method include: Obtain the first and second knowledge graphs to be aligned, and construct attribute vectors for entities in the first and second knowledge graphs; Based on the attribute vectors, calculate the first similarity between each entity in the first knowledge graph and each entity in the second knowledge graph, and construct a first similarity matrix; Based on the neighboring entities of each entity in the first knowledge graph and the second knowledge graph, calculate the number of entities that are mutually aligned with the neighboring entities of each entity in the first knowledge graph and the neighboring entities of each entity in the second knowledge graph, and construct a matrix of aligned neighboring entities. A second similarity matrix is ​​calculated based on the first similarity matrix and the aligned neighbor entity number matrix. Each position in the second similarity matrix corresponds to an entity pair constructed from entities in the first knowledge graph and entities in the second knowledge graph. The information content parameter is calculated based on the entity information contained in each entity pair. The pre-screening comparison value of each entity pair is calculated based on the second similarity and information content parameter at each position in the second similarity matrix. The entity pairs are then screened based on the pre-screening comparison value to obtain the entity pair screening set. For each entity pair in the entity pair filtering set, calculate the multiple similarity between the two entities, perform a weighted calculation on the multiple similarity to obtain a third similarity, and determine the entity alignment of the two entities in the entity pair based on the third similarity.

2. The incremental knowledge graph entity alignment method based on multi-dimensional features according to claim 1, characterized in that, In the step of determining the number of aligned entities between each entity in the first knowledge graph and each entity in the second knowledge graph based on the neighboring entities of each entity in the first knowledge graph and the second knowledge graph, and constructing an aligned neighboring entity number matrix, the number of entities whose neighboring entities in the first knowledge graph and the second knowledge graph are aligned with each other is calculated as the matrix value in the aligned neighboring entity number matrix.

3. The incremental knowledge graph entity alignment method based on multi-dimensional features according to claim 1, characterized in that, In the step of calculating the information content parameter based on the entity information contained in each entity pair, for the two entities in the entity pair, the number of non-empty attributes of the entity, the number of neighboring entities, and the number of aligned entities among the neighboring entities are counted respectively. The sum of the number of non-empty attributes, the number of neighboring entities, and the number of aligned entities among the neighboring entities is calculated as the entity information content. The sum of the entity information content of the two entities in the entity pair is calculated as the information content parameter.

4. The incremental knowledge graph entity alignment method based on multi-dimensional features according to claim 1, characterized in that, In the step of calculating the pre-screening comparison value of each entity pair based on the second similarity and information content parameters at each position in the second similarity matrix, the pre-screening comparison value is calculated using the following formula: in, Represents entities in the first knowledge graph Entities in the second knowledge graph The pre-screened comparison values ​​of the constructed entity pairs. Represents entity pairs The second similarity in the second similarity matrix, Represents entity pairs Information content parameters and These represent the weight values ​​corresponding to the second similarity and information content parameters, respectively.

5. The incremental knowledge graph entity alignment method based on multi-dimensional features according to claim 1, characterized in that, The method further includes the following steps: The entity information of entity pairs in the entity pair filtering set is used to construct a first prompt word. The first prompt word is then input into a preset large language model to update the first knowledge graph and the second knowledge graph.

6. The incremental knowledge graph entity alignment method based on multi-dimensional features according to any one of claims 1 to 5, characterized in that, Multiple similarity includes large model inference similarity. In the step of calculating multiple similarity between two entities for each entity pair in the entity pair filtering set, the entity information of the entity pair in the entity pair filtering set is used to construct a second prompt word. The second prompt word is input into a preset large language model to obtain the large model inference similarity.

7. The incremental knowledge graph entity alignment method based on multi-dimensional features according to any one of claims 1 to 5, characterized in that, Multiple similarity includes embedding vector similarity. In the step of calculating multiple similarity between two entities for each entity pair in the entity pair selection set, the first knowledge graph and the second knowledge graph are input into the graph neural network model. The graph neural network model outputs embedding vectors corresponding to the first knowledge graph and the second knowledge graph. For each entity pair in the entity pair selection set, the embedding vector similarity between the two entities is calculated based on the corresponding embedding vector.

8. The incremental knowledge graph entity alignment method based on multi-dimensional features according to any one of claims 1 to 5, characterized in that, Multiple similarity includes local structural similarity. In the step of calculating multiple similarity between two entities for each entity pair in the entity pair screening set, for each entity pair in the entity pair screening set, the number of mutually aligned entities between the neighboring entities of the entity in the first knowledge graph and the neighboring entities of the entity in the second knowledge graph is calculated as the local structural similarity.

9. The incremental knowledge graph entity alignment method based on multi-dimensional features according to any one of claims 1 to 5, characterized in that, Multiple similarity includes global structural similarity. In the step of calculating the multiple similarity between two entities for each entity pair in the entity pair filtering set, the shortest path length and the total number of paths between the two entities in the entity pair are calculated, and the global structural similarity is calculated based on the shortest path length and the total number of paths. in, Represents entities in the first knowledge graph Entities in the second knowledge graph The global structural similarity of the constructed entity pairs. Represents entity pairs The shortest path length, Represents entity pairs The total number of paths.

10. An incremental knowledge graph entity alignment device based on multi-dimensional features, characterized in that, The device includes a computer device, which includes a processor and a memory, wherein computer instructions are stored in the memory, and the processor is configured to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the device implements the steps of the method as described in any one of claims 1 to 9.

Images