A cross-language entity alignment method and system based on relationship semantic enhancement

Abstract

The application relates to the field of natural language processing, and discloses a cross-language entity alignment method and system based on relationship semantic enhancement, which comprises the following steps: constructing entity initial embedding and relationship embedding representation of a cross-language knowledge graph; a key neighbor screening stage; screening key neighbor entities with structural matching and semantic matching based on an entity similarity matrix and a relationship similarity matrix and in combination with a preset semantic correlation threshold; neighbor-enhanced entity embedding updating; adopting a graph attention network to fuse semantic features of target entities and key neighbors thereof, and generating context-aware entity embedding; semantic consistency alignment propagation: first-order entity feature alignment and second-order relationship structure alignment are jointly optimized through a shared permutation matrix; dynamic seed expansion iteration: high-confidence unaligned entity pairs are dynamically identified according to Manhattan distances of entity embedding, entity pairs with distances lower than a threshold are added to a seed set, and the foregoing steps are iteratively executed until convergence.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of natural language processing, in particular to a cross-language entity alignment method and system based on relationship semantic enhancement. BACKGROUND

[0002] Traditional entity alignment methods, especially vector space mapping-based techniques, often ignore the relationship information between entities, resulting in inaccurate alignment results. To this end, this paper proposes an entity alignment method based on entity and relationship interaction, which effectively models the semantic association between entities and relationships through knowledge graph embedding technology, thereby improving the accuracy and consistency of entity alignment between different data sources.

[0003] In current cross-language knowledge graph (KG) alignment research, the accuracy of entity embedding directly affects the alignment performance.

[0004] However, the same entity across languages has structural heterogeneity, and its neighbor structure often differs significantly due to language or cultural differences, thereby introducing noise to the embedding and reducing the alignment effect. Existing methods mainly rely on the nested similarity between entities to select "important neighbors", but ignore the role of relationship semantics between entities and their neighbors, which cannot fully capture the real semantic relationship between entities, resulting in the selected neighbors being unrepresentative and ultimately affecting the alignment performance. SUMMARY

[0005] The purpose of the present application is to provide a cross-language entity alignment method and system based on relationship semantic enhancement, which solves the above technical problems.

[0006] The purpose of the present application can be achieved by the following technical solutions:

[0007] A cross-language entity alignment method based on relationship semantic enhancement, comprising the following steps:

[0008] S1, constructing entity initial embedding and relationship embedding representation of cross-language knowledge graph;

[0009] S2, key neighbor screening stage;

[0010] Based on the entity similarity matrix and the relationship similarity matrix, and combined with the preset semantic correlation threshold, the key neighbor entities with structural matching and semantic matching are screened;

[0011] S3, neighbor-enhanced entity embedding update;

[0012] The semantic features of the target entity and its key neighbors are fused using a graph attention network to generate context-aware entity embedding;

[0013] S4, semantic consistency alignment propagation:

[0014] The first-order entity feature alignment and the second-order relationship structure alignment are jointly optimized by sharing the permutation matrix.

[0015] S5, dynamic seed expansion iteration:

[0016] According to the Manhattan distance of the entity embedding, high-confidence unaligned entity pairs are dynamically identified, and entity pairs with a distance below a threshold are added to the seed set. Steps S2 to S4 are iteratively executed until convergence.

[0017] As a further technical solution, the key neighbor screening in step S2 specifically includes:

[0018] The entity similarity matrix is calculated by the L1 distance of the entity embedding vector;

[0019] The relationship similarity matrix is calculated by the semantic correlation of the relationship embedding vector;

[0020] Only neighbor entities with entity similarity and relationship similarity exceeding a preset threshold are retained.

[0021] As a further technical solution, the second-order relationship structure alignment in step S4 fuses spatial structure information and semantic association information of the connection relationship, and the specific implementation is:

[0022] By minimizing the relationship structure matrix difference Optimize the shared permutation matrix P; wherein , The association matrix that fuses entity spatial structure and relationship semantics;

[0023] The shared permutation matrix P is constrained to satisfy row normalization, column normalization, and element binary value {0, 1}.

[0024] As a further technical solution, the relationship structure matrix construction method includes:

[0025] If there is a direct relationship connection between two entities, the relationship embedding is fused into the entity spatial structure matrix in a dot product manner;

[0026] If there is no direct relationship, the association matrix is calculated based only on the entity spatial structure information.

[0027] As a further technical solution, the relationship embedding generation method in step S1 includes:

[0028] The relationship vector is constructed by concatenating the head entity average embedding and the tail entity average embedding of the relationship ; wherein represents a connection operation, , represents the relationship an average embedding of all different head entities and tail entities of the triples;

[0029] Introducing TransE-based triple regularization loss Optimizing the relation embeddings.

[0030] As a further technical solution, the dynamic seed expansion of step S5 specifically includes:

[0031] In each iteration, traverse the unaligned entities of the source knowledge graph, and find the candidate entity with the smallest Manhattan distance in the target graph;

[0032] If the distance is less than a threshold , add the entity pair to the seed alignment set and update the embedding representation.

[0033] As a further technical solution, the generation method of the entity embedding in step S1 is:

[0034] Update the entity representation using the multi-layer propagation rule of the graph convolutional neural network:

[0035] ;

[0036] wherein, is the adjacency matrix of the graph plus the identity matrix , is the entity feature vector of the first layer, is the weight matrix of the first layer, denotes the activation function, is the degree matrix of ;

[0037] Optimize the embedding based on the boundary loss function:

[0038] ;

[0039] wherein, denotes the loss function of the entity embedding, denotes a boundary parameter greater than 0, denotes the initial seed alignment set of the entity, denotes a set of negative alignment sets generated by nearest neighbor sampling, denotes the Manhattan distance of the entity embedding, denotes the Manhattan distance of the negative sample entity pair.

[0040] As a further technical solution, a semi-supervised iterative strategy is used to train the model:

[0041] The joint entity alignment loss and the relationship regularization loss are balanced by a weight coefficient.

[0042] The entity distance matrix and the relationship distance matrix are updated synchronously in the iteration process.

[0043] A cross-language entity alignment system based on relationship semantic enhancement, which is used to perform the cross-language entity alignment method based on relationship semantic enhancement.

[0044] The beneficial effects of the present application are:

[0045] The first-order entity feature alignment is used to capture the neighbor information of the entity, and the second-order relationship structure alignment is used to mine the deep semantic relationship between entities, so that equivalent entities in different knowledge graphs can be accurately matched. BRIEF DESCRIPTION OF DRAWINGS

[0046] The present application will be further described below with reference to the accompanying drawings.

[0047] Figure 1 A schematic diagram for the neighbor heterogeneity problem in cross-language entity alignment;

[0048] Figure 2 A schematic diagram of the framework of the present application;

[0049] Figure 3 A process of relationship structure alignment. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0051] Please refer to Figures 1-3 The present application is a cross-language entity alignment method based on relationship semantic enhancement, characterized by the following steps:

[0052] S1, constructing the initial entity embedding and relationship embedding representation of the cross-language knowledge graph;

[0053] S2, key neighbor screening stage;

[0054] Based on the entity similarity matrix and the relationship similarity matrix, and in combination with a preset semantic correlation threshold, key neighbor entities are screened for structure matching and semantic matching;

[0055] S3, neighbor-enhanced entity embedding updating;

[0056] The semantic features of the target entity and its key neighbors are fused by using a graph attention network to generate context-aware entity embedding.

[0057] S4, semantic consistency alignment propagation:

[0058] First-order entity feature alignment and second-order relationship structure alignment are jointly optimized through a shared permutation matrix.

[0059] S5, dynamic seed expansion iteration:

[0060] High-confidence misaligned entity pairs are dynamically identified according to the Manhattan distance of entity embedding, and entity pairs with a distance below a threshold are added to the seed set. Steps S2 to S4 are iteratively executed until convergence.

[0061] In this embodiment, first, the entities and relationships in the two knowledge graphs are embedded using GCNs with a TransE regularizer to obtain entity and relationship embedding vectors, respectively. Second, the entity embedding vectors and the relationship embedding vectors are associated to construct a graph embedding representation matrix. Third, a first-order entity feature alignment module is constructed based on the neighbor information of the entities to capture the context information of the entities, and a second-order relationship structure alignment module is further constructed using the spatial structure and semantic association information of the connection relationships, with the purpose of mining deep semantic relationships between entities. Next, a shared permutation matrix P is learned by jointly optimizing the first-order entity feature alignment and the second-order relationship structure alignment to effectively promote the positive interaction between entities and relationships, thereby improving the alignment performance. Finally, the positive interaction between the entity alignment task and the relationship alignment task is jointly learned in a semi-supervised iterative manner. In each iteration, the embedding representation of the entities and relationships is updated using the shared permutation matrix, and the similarity of the embedding vectors is used to determine the alignment of the new entities and relationships.

[0062] Entity embedding:

[0063] The entity embedding uses a given pair of knowledge graphs and a set of seed alignments of entities to embed all entities in the two knowledge graphs into the same latent space using GCNs, thereby taking into account the structural information of the two knowledge graphs.

[0064] Specifically, first, the entity representations are initialized by using pre-trained word embeddings to provide necessary semantic information for entities; the above initialization method not only ensures the similarity of all entities on a common basis, but also lays the foundation for the subsequent refinement of GCN layers; second, the context information of the two KGs is captured by GCNs, which effectively embeds all entities into the shared latent space. The core of GCNs lies in its multi-layer propagation rule, which constantly updates the representation of each entity by aggregating the features of neighboring entities; in the above process, the representation of each entity is adjusted and optimized according to the information of its neighbors, so as to achieve a more accurate and comprehensive entity representation. In GCN, the update rule of a single layer can be expressed as:

[0065] Update the entity representation using the multi-layer propagation rule of the graph convolutional neural network:

[0066] ;

[0067] wherein, is the adjacency matrix of the graph plus the unit matrix , is the entity feature vector of the th layer, is the weight matrix of the th layer, denotes the activation function, is the degree matrix of ;

[0068] In the initialization stage, is the entity embedding vector obtained from the pre-trained word embedding; the activation function is the ReLU function, i.e. ; the neighbor aggregation strategy in GCNs can effectively capture the topological structure and entity feature information of the graph;

[0069] The final entity representation is defined as , wherein denotes the dimension of the entity embedding, denotes the number of entities;

[0070] In order to embed the entities in the two into the same latent space, seed alignment is used as the training sample. At the same time, a boundary-based loss function is designed to encourage clear separation between different classes by maximizing the interval between classes;

[0071] Optimize the embedding based on the boundary loss function:

[0072] ;

[0073] wherein, a loss function representing entity embeddings, a boundary parameter representing a value greater than 0, a set of initial seed alignments representing entities, a set of negative alignments generated by nearest-neighbor sampling, a Manhattan distance representing entity embeddings, a Manhattan distance representing pairs of negative sample entities.

[0074] is an L1 norm representing a vector, The lower the value of the loss function means that the entities and are more strongly aligned; the principle of this loss function is that the distance between aligned pairs of entities should be close to zero; conversely, the distance between negative samples should be maximized as much as possible.

[0075] Relation embeddings:

[0076] In , facts are encoded as triples, i.e., ( , , ), where the meaning of the relation is associated with the two entities connected to it; in order to exploit the semantic association information of the connected entities, the embeddings of the head and tail entities learned from GCNs are used to represent the relations in the knowledge graph:

[0077] ;

[0078] , representing the relation , denotes a connection operation, , represent the average embedding of all different head and tail entities of the relation , respectively;

[0079] Introducing a TransE-based triple regularization loss to optimize the relation embeddings. The purpose is to promote the sparsity of the model through L1 regularization, which helps the model to focus on the most important features, thereby reducing the risk of overfitting.

[0080] ;

[0081] wherein and represent the triple sets of two given knowledge graphs and , respectively; denotes a conversion matrix from the latent space of the relation to the latent space of the entity, which is a model parameter to be learned.

[0082] To jointly learn the embeddings of entities and relations, the objective function is designed as:

[0083] ; is the weight coefficient, which is used to balance the entity alignment loss and the regularization loss while considering the influence of the relation embedding; the main goal is to minimize the objective function after the pre-training of the entity embedding is completed.

[0084] Entity feature alignment:

[0085] By comparing the similarity of the first-order neighbor feature information of entities, the correspondence between entities in different knowledge graphs is maximized; the first-order feature information of each entity is obtained from the entity embedding module, which contains neighbor and context information related to the entity. In the context of aligning the first-order entity features, if the entity in and the entity in , respectively, show greater similarity than other entities, then the possibility of a strong correspondence between entity and is increased; in order to achieve alignment, an optimization objective is used to minimize the difference between entity features. The expression of the objective is shown in the formula:

[0086] ;

[0087] , , ;

[0088] wherein, represents the embedding representation of all entities in , represents the embedding representation of all entities in , and represent the number of entities in and , represents the Frobenius norm, which aims to measure the overall size of the matrix, which is mainly obtained by calculating the square root of the sum of the squares of the absolute values of each element in the matrix; in the above formula, represents a shared permutation matrix, which aims to incorporate permutation-based operations into the alignment task; is a model training parameter, which promotes the exploration of multiple alignment possibilities and captures various alignment patterns, which is useful for establishing different The correspondence between entities is crucial; specifically, the equation indicates that when the matrix is multiplied by a vector consisting entirely of 1s, the resulting vector also consists entirely of 1s, indicating that the sum of the elements in each row of the matrix is equal to 1. This constraint ensures that is row-normalized, i.e., the sum of each row is 1, thus preserving the integrity of the row distribution;

[0089] Similarly, the equation confirms that the sum of the elements in each column of the matrix is equal to 1, ensuring column normalization; the above formula guarantees that the matrix is simultaneously row and column normalized; at the same time, a constraint is imposed on the shared permutation matrix , which can only contain 0s and 1s, thus ensuring a one-to-one correspondence between entities in the two knowledge graphs, thereby improving alignment performance.

[0090] Relationship structure alignment:

[0091] On the basis of first-order entity feature alignment, second-order relationship structure alignment is introduced; the goal is to ensure that the connection between entities and their neighbors remains comparable, mainly considering the spatial structure of the connection and its semantic association information to improve alignment performance in different knowledge graphs.

[0092] First, find the one-to-one similarity in the spatial structure of the relationship by optimizing the shared permutation matrix to help align entities.

[0093] For example, as Figure 3 , there are two different knowledge graphs; if and are equivalent entities, and are also equivalent entities, then the spatial structure of the connection relationship between entities ( to , to ) should also be similar.

[0094] Therefore, when it is known that and are equivalent, by finding the spatial structure similarity information of the connection relationship ( to , to ) through optimization of the shared permutation matrix, it can be determined that to is similar.

[0095] Second, the semantic association information of the connection is combined to enhance alignment;

[0096] The inherent semantic associations between the relationships in two knowledge graphs can be better facilitated by one-to-one matching to promote entity alignment.

[0097] By combining the semantic association information of the relation embedding vector with the spatial structure information, a matrix containing the spatial structure and semantic association information of the connection relationship is formed.

[0098] For example, such as Figure 3 As shown, Existence Relationship connect and ,as well as Existence Relationship connect and ,and Equivalent to When found through the shared permutation matrix and When similar, it can be inferred and They are similar, thus improving entity alignment performance. Therefore, by minimizing the function shown in the formula, alignment of second-order relation structures can be achieved:

[0099] ;

[0100] , , ;

[0101] in, , , and They represent and A matrix containing spatial structure and semantic association information of connection relationships. and They are and The embedded representations of all relations obtained from the relation embedding module contain semantic association information of the relations. and They described respectively and The internal spatial structure of the two entity sets;

[0102] The specific proposed design method is as follows;

[0103] If in There exists and The connection relationship between them is then embedded in the vector. The spatial structure information between entities is integrated in the form of point multiplication to better preserve the spatial structure and semantic association information of the relationship. If There is no connection relationship in the form of triple between , the spatial structure information between entities is considered for alignment, such as Figure 3 In and , similarly, the same operation is also performed in to improve the alignment performance; by minimizing the formula, the convergence of the relationship structure alignment to high consistency is realized; the shared permutation matrix plays a crucial role in capturing the interaction between entity features and relationship structures, thereby enhancing the comprehensiveness of the alignment method.

[0104] Joint interaction learning optimization goal:

[0105] The optimization goal of joint interaction learning is to guide the finding of one-to-one corresponding similar entity neighbors and relationship spatial structures in different knowledge graph subgraphs by training the shared permutation matrix P through the combination of entity feature alignment and relationship structure alignment modules, so as to minimize the optimization goal of joint interaction learning; as shown in the formula:

[0106] ;

[0107] , , ;

[0108] Among them, is a balance hyperparameter, and the constraints of the optimization problem are and , that is, the sum of the elements of each row and each column of the permutation matrix is equal to 1; in addition, the value constraint of P is , that is, The elements of can only take 0 or 1; by optimizing the shared permutation matrix , the interaction between entity features and relationship structures can be effectively promoted; the above comprehensive alignment method considers the complexity between entities and relationships in the knowledge graph.

[0109] Semi-supervised iterative strategy and implementation details:

[0110] In order to train an effective EA model, the objective function of embedding learning in the formula and the objective function of joint interaction learning are combined, and the final optimization goal is represented as:

[0111] ;

[0112] Among them, is the weight coefficient of the joint interactive learning objective function, by changing the value of , the influence of joint interactive learning on the whole optimization process can be controlled; during the training process, different objectives may guide the model to optimize in different directions, and the parameter can ensure that the above two objectives can promote each other;

[0113] The ultimate goal is to learn the interaction between entity features and relationship structures, so as to realize a comprehensive alignment method. For this purpose, a semi-supervised framework is proposed, which uses the positive interaction between entity and relationship alignment tasks to iteratively enhance alignment, thereby improving the robustness and accuracy of the model.

[0114] The distance matrix of entity pairs from and is represented as: The distance matrix of relationship pairs from to is represented as: .

[0115] Using the learned entity and relationship embeddings, and are defined as shown in the formula:

[0116] ;

[0117] ;

[0118] Within a specified number of iterations , the model refines the alignment process by using the corresponding embeddings, thereby realizing the alignment of entities and relationships between two different knowledge graphs; the above iteration process is guided by the distance thresholds and of entities and relationships, and in each iteration, the algorithm traverses each entity and relationship in , and finds the closest corresponding item in ; if the distance between the aligned entity and the aligned relationship is lower than the respective threshold, the algorithm identifies a potential alignment; then, the entity alignment set and the relationship alignment set are updated accordingly; subsequently, the model updates the distance matrix of entities and the distance matrix of relationships by optimizing the first-order entity feature alignment and the second-order relationship structure optimization The iteration of the alignment and the refinement of the distance matrix are continued until a maximum number of iterations is reached ; finally, the algorithm outputs the updated distance matrix and and the updated entity alignment set and the relation alignment set .

[0119] The present application aims to improve the accuracy and robustness of cross-language entity alignment. The core idea is to identify and utilize key neighbor information in the entity alignment process, integrate entity and relationship semantic information, and optimize entity embedding. It mainly includes the following stages:

[0120] Step 1, data preprocessing and initial entity representation construction;

[0121] Extracting multi-modal information of structure, attribute, and text description of entities in cross-language knowledge graph;

[0122] Using existing embedding methods (such as TransE, GCN) to generate initial representations for entities;

[0123] According to the existing aligned entities, construct a seed set to provide initial supervision signals for subsequent alignment training;

[0124] Step 2, entity-relation similarity matrix calculation and neighbor screening;

[0125] Calculate entity similarity matrix and relation similarity matrix respectively to measure the semantic similarity between cross-language entities and their relationships;

[0126] Set a threshold for the similarity matrix to ensure that only the best neighbor entity is selected for each row;

[0127] Eliminate low-quality and semantically misleading neighbors and only keep key neighbors that are highly matched in structure and semantics;

[0128] Step 3, neighbor-enhanced embedding update mechanism design;

[0129] Design an embedding update module based on graph attention network (GAT) and introduce high-quality neighbor information after screening;

[0130] Fuse the entity's own representation with its key neighbor representation, recalculate the entity embedding, and strengthen the semantic context;

[0131] Step 4, semantic alignment propagation mechanism;

[0132] Optimize the embedding space for the initial aligned entity pair through the triple embedding criterion ( );

[0133] Utilize the semantic consistency between the aligned entity pairs to guide the update of the unaligned entity embeddings;

[0134] Realize the linkage optimization of the entire embedding space, and improve the global alignment effect;

[0135] The fifth step is a dynamic seed expansion mechanism;

[0136] Take the Manhattan distance as the alignment judgment standard, if the distance between the unaligned entity pairs is less than the set threshold, the new aligned entity is determined;

[0137] Add the new aligned entity meeting the condition to the seed set, and cyclically optimize the embedding and the alignment boundary;

[0138] Repeat the training for several rounds until convergence, and finally complete the large-scale cross-language entity alignment.

[0139] The above has carried out the detailed description to one embodiment of the application, but the content described is only the preferred embodiment of the application, cannot be considered for limiting the implementation range of the application. All equivalent changes and improvements made according to the scope of the application are still attributed to the patent coverage range of the application.

Claims

1. A cross-language entity alignment method based on relationship semantic enhancement, characterized in that, Comprising the following steps: S1, constructing an initial embedding of entities and a relation embedding of a cross-language knowledge graph; S2, key neighbor screening stage; Based on the entity similarity matrix and the relation similarity matrix, combined with the preset semantic correlation threshold, the key neighbor entities of structural matching and semantic matching are screened; S3, neighbor enhanced entity embedding update; Adopting a graph attention network to fuse the semantic features of the target entity and its key neighbors to generate context-aware entity embedding; S4, semantic consistency alignment propagation: Optimizing the first-order entity feature alignment and the second-order relation structure alignment jointly through a shared permutation matrix; S5, dynamic seed expansion iteration: According to the Manhattan distance of the entity embedding, dynamically identify high-confidence unaligned entity pairs, add entity pairs with a distance below the threshold to the seed set, and iteratively execute steps S2 to S4 until convergence; The second-order relation structure alignment in step S4 fuses the spatial structure information and semantic association information of the connection relation, and the specific implementation is: By minimizing the difference of the relational structure matrix Optimizing the shared permutation matrix P; wherein , The association matrix fuses the entity space structure and the relationship semantics. The shared permutation matrix P satisfies row normalization, column normalization, and element is binary value {0, 1}; The relation embedding generation method in step S1 includes: The relation vector is constructed by concatenating the head entity average embedding and the tail entity average embedding ; wherein denotes a concatenation operation, , denote the average embeddings of all different head entities and tail entities of the relation , respectively. Introducing TransE-based triple regularization loss Optimizing relation embeddings; A transformation matrix representing the mapping from the latent space of relations to the latent space of entities is a model parameter to be learned; The generation method of the entity embedding in step S1 is: Update the entity representation using the multi-layer propagation rule of the graph convolutional neural network: ； wherein, is the adjacency matrix of the graph plus the identity matrix , is the entity feature vector of the layer, is the weight matrix of the layer, denotes the activation function, is the degree matrix of ; Optimize the embedding based on the boundary loss function: ； wherein, represents a loss function for entity embeddings, represents a boundary parameter greater than 0, represents an initial seed alignment set of entities, represents a set of negative alignment sets generated by nearest neighbor sampling, represents a Manhattan distance for entity embeddings, represents a Manhattan distance for negative sample entity pairs.

2. The method of claim 1, wherein the relationship semantics enhanced cross-lingual entity alignment method is characterized by, The key neighbor screening in step S2 specifically includes: The entity similarity matrix is calculated by the L1 distance of the entity embedding vector; The relation similarity matrix is calculated by the semantic association degree of the relation embedding vector; Only the neighbor entities whose entity similarity and relation similarity both exceed the preset threshold are retained.

3. The method of claim 1, wherein the relationship semantics are enhanced based on, The construction method of the relation structure matrix includes: If there is a direct relation connection between two entities, the relation embedding is fused into the entity spatial structure matrix in a dot product manner; If there is no direct relation, only the association matrix is calculated based on the entity spatial structure information.

4. The method of claim 1, wherein the relationship semantics are enhanced based on, The dynamic seed expansion of step S5 specifically includes: In each iteration, traverse the unaligned entities of the source knowledge graph, and find the candidate entity with the smallest Manhattan distance in the target graph; If the distance is less than a threshold then the entity pair is added to the seed alignment set and the embedding representation is updated.

5. The method of claim 1, wherein the relationship semantics enhanced cross-lingual entity alignment method is characterized by, Adopting a semi-supervised iterative strategy to train the model: Joint entity alignment loss and relation regularization loss, balance the influence of the two through a weight coefficient; Update the entity distance matrix and the relation distance matrix synchronously during the iteration process.

6. A cross-language entity alignment system based on relation semantics enhancement, characterized in that, The system is used to perform the cross-language entity alignment method based on relation semantic enhancement according to any one of claims 1-4.

Images