CN114819152A - Graph embedding expert entity alignment method based on reinforcement learning enhancement - Google Patents

Abstract

The invention discloses a graph embedding entity alignment method based on reinforcement learning enhancement. The present invention adopts the method of constructing a heterogeneous subgraph, and only performs message aggregation on the n-hop neighbors of the pair of entities to be aligned, thereby directly reducing computing resource requirements. The graph embedding learning algorithm based on feature linear modulation is used, and the idea of super network is introduced to complete the message passing mechanism and node update mechanism of high computational complexity with a small number of parameters, so as to make better use of the interaction information between nodes. In addition, the present invention proposes a node selector enhanced by reinforcement learning, proposes and applies a reliability measurement method based on self-supervised signals in the node selector, samples a certain number of reliable edges, and limits the size of the heterogeneous subgraph. At the same time, the problem edges are filtered to ensure the reliability of the edges participating in the node update. The invention also realizes a node sampling number update strategy based on reinforcement learning, dynamically optimizes the number of sampling nodes, and strengthens the node selector.

Description

An Expert Entity Alignment Method for Graph Embedding Enhanced Based on Reinforcement Learning

technical field

The invention relates to the technical field of knowledge graphs in natural language processing, in particular to a graph embedding entity alignment method based on reinforcement learning enhancement.

Background technique

"The competition in the 21st century is the competition of talents", and the talent element occupies the core position of the industrial element. The talent knowledge base can be used as the upstream data knowledge support for downstream decision-making tasks such as factor scheduling, talent recommendation, and project precision investment. At present, massive amounts of talent-related data are widely distributed on Internet platforms such as academic institution websites and academic search engines. In order to build a large-scale, high-quality talent knowledge base, integrating the knowledge of other talent knowledge bases is one of the necessary means. As a hub linking different knowledge bases, the expert entity is very important for integrating various talent knowledge bases. The process of identifying expert entities representing the same individual in the real world in different talent knowledge bases is called expert entity alignment.

Entity alignment usually compares some features of the entity pair to be aligned, such as entity name, entity attribute and attribute value, and uses some machine learning methods or methods based on representation learning to calculate and score the similarity. However, for the talent knowledge base, the existing data characteristics and the application requirements of real-world scenarios put forward some requirements for the existing entity alignment methods: First, the available information is reduced. The relationships and attributes in the talent knowledge base are enumerable and do not need to be aligned. This feature makes some existing entity alignment methods unable to utilize the alignment information of relation predicates and attribute predicates, resulting in model performance degradation. Second, the computing resources are limited and the running results are unstable. The scale of entities in the talent knowledge base is very large, and a large number of paper entities or expert entities will be added every day, and the number of results published by different experts is also quite different, which may cause a certain degree of computational instability in practical application scenarios. Third, there are problem edges in the knowledge base. In various existing knowledge bases, it is relatively common to have wrong triple data, and the existence of these wrong problem edges will undoubtedly have a negative impact on whether the model determines whether the entity is the same real entity.

SUMMARY OF THE INVENTION

The purpose of the present invention is to overcome the deficiencies of the prior art, and to provide a graph embedding expert entity alignment method enhanced by reinforcement learning. The technical scheme of the present invention is as follows:

The present invention provides a graph embedding expert entity alignment method based on reinforcement learning enhancement, which includes the following steps:

Step 1: Obtain the data G ₁ =(E ₁ , T ₁ , R) and G ₂ =(E ₂ , T ₂ , R) of two talent knowledge bases, where E represents the entity set, and the entity types include expert entities, papers Entity; R stands for relation set; T stands for triple set, which is a subset of E×R×E;

Step 2: For each expert entity e in a certain talent knowledge base, according to the expert name, through the candidate entity pair generation module, based on the method of regular matching template generator, generate a candidate expert set C from another talent knowledge base ;

H₁、H₂分别代表两个人才知识库的实体初始向量集合；初始的节点向量表示h⁽⁰⁾：Step 3: For each candidate expert c∈C, construct a 2-hop heterogeneous subgraph HG=(V,H,T,R) about entity pair <e,c>, where

H ₁ and H ₂ respectively represent the entity initial vector sets of the two talent knowledge bases; the initial node vector represents h ⁽⁰⁾ :

h _struct =LINE(G)

表示向量拼接操作，h_attr为实体各属性特征通过skip-gram模型获得的词向量的平均向量，h_struct则是通过LINE模型对知识库中每个实体的结构信息进行编码得到的结构向量；in

Represents the vector splicing operation, h _attr is the average vector of word vectors obtained by each attribute feature of the entity through the skip-gram model, and h _struct is the structure vector obtained by encoding the structural information of each entity in the knowledge base through the LINE model;

Step 4: In the node vector update module, in each layer of the graph embedding layer, use the reliability measurement method based on the self-supervised signal to calculate the reliability of each edge;

Step 5: In each layer of graph embedding layer, for the heterogeneous subgraph HG, use the top-p sampling strategy, according to the reliability calculated in step 4, sample p _r reliable edges for each relationship from large to small, And use the node sampling number update strategy based on reinforcement learning to update the node sampling number;

Step 6: In each layer of graph embedding layer, after obtaining the sampled heterogeneous subgraph, use the graph embedding learning algorithm based on feature linear modulation to update the node vector;

和

通过多层感知机计算匹配概率

Step 7: After passing through the L-layer graph embedding layer, take out the updated node vector of the entity pair <e, c> to be aligned

and

Calculation of Matching Probabilities by Multilayer Perceptron

Step 8: According to the matching probability of all candidate entities, the candidate expert with the highest probability is selected as the matching expert.

Compared with the existing methods, the advantages of the method of the present invention are:

(1) Aiming at the problem of reducing the available information, a graph embedding learning algorithm based on feature linear modulation is used, and the idea of super network is introduced. The relationship weight matrix can be dynamically generated according to the target node vector, and the message with high computational complexity can be completed with a small number of parameters. Transfer mechanism and node update mechanism, so as to better utilize the interaction information between nodes;

(2) Aiming at the problem edges existing in the knowledge base, the method of the present invention proposes a node selector enhanced by reinforcement learning, which applies a reliability measurement method based on self-supervised signals, constructs self-supervised signals based on existing edges, This makes it possible to sample relational edges according to reliability, so as to filter the problem edges and ensure the reliability of the edges participating in node update. Since the size of the sampled heterogeneous subgraph is controlled, the problems of limited computing resources and unstable running results are also solved.

(3) In order to avoid a large number of manual adjustment of the number of node sampling, the node selector also implements a node sampling number update strategy based on reinforcement learning to dynamically optimize the number of sampling nodes.

Description of drawings

Figure 1 is the overall framework diagram of the expert entity alignment method for graph embedding based on reinforcement learning enhancement.

Figure 2 is the overall flow chart of the candidate entity regular matching template generator.

Figure 3 is the overall flow chart of the node vector update module, in which the dashed box is only implemented during the training process.

Detailed ways

The present invention will be further elaborated and described below in conjunction with specific embodiments. The described embodiments are merely exemplary of the present disclosure and do not delineate the scope of limitation. The technical features of the various embodiments of the present invention can be combined correspondingly on the premise that there is no conflict with each other.

表示用于训练的已对齐实体对集合。专家实体对齐任务旨在利用已知的实体对信息训练得到一个模型，并以此预测潜在的专家对齐结果

其中等号代表两个专家实体指向真实世界中同一个体。The talent knowledge base usually contains a variety of entity types and relationship type information. Given two talent knowledge bases, G ₁ =(E ₁ , T ₁ , R) and G ₂ =(E ₂ , T ₂ , R), where E Represents entity set, entity types include expert entity and paper entity; R represents relation set; T represents triple set, which is a subset of E×R×E. set of seed entity pairs

Represents the set of aligned entity pairs used for training. The expert entity alignment task aims to use the known entity pair information to train a model to predict potential expert alignment results.

The equal sign indicates that two expert entities refer to the same individual in the real world.

Given an expert entity in a talent knowledge base, the process of finding its corresponding expert entity in another talent knowledge base can be regarded as a matching problem, that is, in a certain feature space, calculating the relationship between a given expert entity and another The matching probability of all candidate expert entities in the talent knowledge base, and the entity with the highest matching score is regarded as the alignment result.

As shown in Fig. 1, the present invention designs an overall framework diagram of a graph embedding expert entity alignment method enhanced by reinforcement learning: for two talent knowledge bases, first generate candidate entity pairs according to names; then for each pair of entity pairs, construct a 2-hop Heterogeneous subgraph; then in the node vector update module, in each layer of the graph embedding layer, the node selector based on the self-supervised signal is used to sample the reliable edges of the heterogeneous subgraph, and at the same time, the reinforcement learning module is used to perform reliable edge sampling. The number of samples is dynamically updated, and the heterogeneous subgraph after filtering the problem edge completes the node vector update of the current layer through the graph embedding based on feature linear modulation, with a total of L layers of graph embedding layers; finally, the final node vector representation of the expert entity pair to be aligned is taken out. , after the vector point product, the multi-layer perceptron is used to complete the calculation of the matching probability.

Each module (step) in the present invention will be described in detail below.

S1 candidate entity pair generation: In order to deal with the situation of Chinese and English names due to abbreviations, polyphonic characters, and complex surnames, a method based on regular matching template generator is implemented. Figure 2 details the process of how the regular matching template generator generates the regular matching template.

First, the Chinese and English names are processed separately, and the name parser is used to generate a parsing dictionary. For English names, first parse the surname and first name, determine whether the order of the surname and first name can be determined according to certain rules, and then generate a parsing dictionary. For Chinese names, the Chinese names are converted into pinyin form, that is, into English names, and polyphonic characters and compound surnames are considered at the same time, and the intermediate analysis results are generated and processed as English names. Table 1 shows the parsing rules for converting Chinese names into English names, Table 2 shows the rules for determining the order of surnames and first names for English names/transformed Chinese Pinyin names, and Table 3 shows the method of generating a parsing dictionary.

Table 1 Parsing rules for converting Chinese names into English

Table 2 Rules for determining the order of last names and first names

Table 3 Parsing dictionary generation method

The parsing dictionary generated by the name parser is used as the input of the template generator, and the corresponding regular template is output. First, according to whether the order of the surname and the first name can be determined, then determine whether the first name is an abbreviation, and finally generate 2 or 4 groups of regular templates according to the two orders of "surname first" and "surname last". Table 4 shows the abbreviation judgment rules, and Table 5 shows the rules for generating regular templates from the parsing dictionary.

Table 4 Abbreviation Judgment Rules

Table 5 Rules for generating regular templates from parsing dictionaries

S2 constructs an n-hop heterogeneous subgraph: taking n=2 as an example, for the expert entity pair <a ₁ , a' ₁ > to be aligned, obtain its first-order neighbor nodes from G ₁ and G ₂ respectively, including the published Paper nodes, conference/journal nodes, expert nodes of co-authorship relationships, and paper nodes and conference/journal nodes of first-order neighbor expert nodes, thus constructing a subgraph with a ₁ or a' ₁ as the core. Under the result of known partial entity alignment (such as whether the conference/journal name is the same, whether the paper title is the same, and the expert entity has been aligned), the two originally separated subgraphs can be merged to obtain a pair of expert entities to be aligned < a ₁ , a′ ₁ > as the core heterogeneous subgraph.

S3 node vector update module: This step completes the update of the node vector representation in the heterogeneous subgraph through the L-layer graph embedding layer, and the initial node vector representation h ⁽⁰⁾ needs to be obtained. The specific process of updating the node vector of each graph embedding layer is as follows As shown in Figure 3, the reliability of each edge is first calculated by the reliability measurement method based on the self-supervision signal, and then the top-p sampling strategy is adopted. According to the calculated reliability, a certain number of each relationship is sampled from large to small. Reliable edge of p _r , after obtaining the sampled heterogeneous subgraph, use the graph embedding learning algorithm based on feature linear modulation to update the node vector, and finally take out the final node vector representation of the entity pair to be aligned, and calculate the matching probability through the multilayer perceptron . In the training stage, in order to obtain a better reliability measurement method, the non-existing relationship edge SPF is constructed by sampling the existing relationship edge SPT, and sampling nodes that do not have the relationship with the target node at a ratio of 1:1 to construct a self-supervised relationship. signal, computes the average edge sampling loss to train the weight matrix in the reliability measure. In addition, in order to dynamically update the number of node samples _pr for each relation, and calculate the average unreliability of sampled edges, the number of samples is updated using a reinforcement learning-based node sample number update strategy.

The initial node vector representation h ⁽⁰⁾ :

h _struct =LINE(G)

表示向量拼接操作，h_attr为实体各属性特征(包括论文标题、摘要、会议 /期刊名等)通过skip-gram模型获得的词向量的平均向量，h_struct则是通过LINE 模型对知识库中每个实体的结构信息进行编码。in

Represents the vector splicing operation, h _attr is the average vector of the word vectors obtained by the skip-gram model of each attribute feature of the entity (including paper title, abstract, conference/journal name, etc.), and h _struct is the LINE model. Encoding the structural information of each entity.

Calculate the reliability of each edge: For triples <u, r, v> (that is, there is a relationship r between node u and node v, and the direction is u points to v), the calculation of its reliability S(u, v) The process is as follows:

First compute the feature representation η of a node on each relation:

η _u , η _v =σ ₁ (W _τ(u) h _u ), σ ₁ (W _τ(v) h _v ))

where W _τ(u) and W _τ(v) respectively represent the weight matrix related to the corresponding node type, its dimension is R ^|R|×d , h _u and h _v are the nodes u and v in the graph embedding layer of this layer The node vector representation of , with dimension d.

Then calculate the probability α _{u, v} that the node pair <u, v> has a relationship r:

α _u,v =σ ₂ (W _TF ·(η _u ⊙η _v ))

The dimension of W _TF is R ^2×|R| , and ⊙ is the vector dot product operation.

Finally, the unreliability D(u, v) is calculated using the Manhattan distance, and the reliability S(u, v) is obtained:

D(u, v)=||α _{u, v} || ₁

S(u,v)=1-D(u,v)

top-p sampling strategy: For each node in the heterogeneous subgraph, for each relationship _r , according to the calculated reliability, sample pr edges from large to small.

更新为

h_v∈R^d：The graph embedding learning algorithm based on feature linear modulation updates the node vector: at the lth layer, for node v, its node vector is given by

update to

h _v ∈ R ^d :

分别是节点v在第l层图嵌入层中计算所得的每个邻居节点u的消息权重：where W _r ^(l) ∈ R ^d×d is the weight of the message conversion function, σ is the ReLU function,

are the message weights of each neighbor node u calculated by node v in the lth layer of graph embedding layer:

where g(h _v ; θ _{g, l, r} ) is a super-network implemented using a single linear layer, and θ _{g, l, r} ∈ R ^2d×d are the parameters of this hyper-network.

和

通过多层感知机计算匹配概率

具体计算过程为：S4 calculates the matching probability: After passing through the L-layer graph embedding layer, take out the updated node vector of the entity pair <e, c> to be aligned

and

Calculation of Matching Probabilities by Multilayer Perceptron

The specific calculation process is as follows:

where FC is a 3-layer linear layer, and each layer uses a ReLU nonlinear function.

S5 training phase to update the number of node samples and calculate the loss function: (1) Calculate the average unreliability of the sampled edges, and use the reinforcement learning strategy to dynamically optimize the number of node samples; (2) Build a self-supervised signal and calculate the average edge sampling loss, And sum with the alignment loss to get the final loss.

作为负例，构建基于已有关系边的自监督信号。Build a self-supervised signal: For each relation r, 60% of the existing partial relation edges in the sampling graph SPT={<u, r, v>|<u, r, v>∈T}, as a positive example, And sample nodes that do not have the relationship with the target node at a ratio of 1:1 to construct non-existing relationship edges

As a negative example, a self-supervised signal based on existing relational edges is constructed.

从而训练计算可靠性时使用的各项权重系数：Calculate the average edge sampling loss: Minimize the average edge sampling loss across all graph embedding layers using the cross-entropy loss as the loss function

Thus training the various weight coefficients used in calculating reliability:

Where ψ(u, r, v)=1 means that the edge <u, r, v> exists, otherwise, ψ(u, r, v)=0 means that the edge does not exist.

Calculate the mean unreliability of sampled edges: at each epoch e, for relation r, calculate the mean unreliability of sampled edges _{Er, sampled}

Update the number of node samples: According to the change trend between the average unreliability between two adjacent epochs, the number of node samples p _r to update the relationship r, the Action, Reward, and Termination in the reinforcement learning strategy used are:

Action={+∈,-∈}

where ∈ is a small fixed integer.

If the average unreliability of the current epoch is smaller than the previous epoch, the reward function is positive, and the instant reward method is used to greedily increase p _r , otherwise, decrease p _r .

该损失的目标是使得对齐的专家实体对的向量表示对<z_e，z_c>经过多层感知机后得到的概率越高，接近于1，而非对齐的向量表示对的分数接近于0：During the training process, the loss of the expert entity alignment task includes the alignment loss of all entity pairs to be aligned in addition to the average edge sampling loss of the reliability measure based on self-supervised signals

The goal of this loss is to make the vector representation pair <z _e , z _c > of aligned expert entity pairs get a higher probability after passing through the MLP, close to 1, while the score of the unaligned vector representation pair is close to 0 :

The final loss function of the model is the weighted sum of the alignment loss and the average edge sampling loss:

where ||Θ|| ₂ is the L2 regularization penalty term, λ ₃ is the corresponding penalty term coefficient, and λ ₁ and λ ₂ are the weight coefficients for alignment loss and average edge sampling loss, respectively.

In the experiment, this example uses the data set OAG, which involves two major academic networks, Microsoft Academic Network and AMiner. The size of the network is shown in Table 6, and the basic statistical information of the data set is shown in Table 7.

Table 6 The scale of network data in the OAG dataset

network Number of conferences/journals Number of papers Number of experts relationship type AMiner 69,397 172, 209, 563 113, 171, 945 co-author, write, publish_on MAG 52,678 208, 915, 369 253, 144, 301 co-author, write, publish_on

Table 7 Basic statistics of OAG dataset

Evaluation Metrics: Use Precision, Recall and F1-score as evaluation metrics. Precision is the proportion of correct predictions in the positive samples predicted by the classifier. The larger the value, the higher the accuracy of the model's prediction of positive samples. Recall is the ratio of the correct positive samples predicted by the evaluation classifier to all positive samples. The larger the value, the higher the probability that the model will predict the correct positive samples. F1-score comprehensively evaluates Precision and Recall. The higher the value, the better the overall prediction performance of the model.

Hyperparameter settings: In the experiment, epoch=100, the size of each batch is 64, the dimension of the vector representation of each node is d=300, the optimizer selects Adam, and the learning rate of the graph embedding learning algorithm based on feature linear modulation is 0.005 , and the learning rate of the reliability measurement method based on self-supervised signals is 0.001; the initial sampling number p _r of each relationship is 30, and each action update ∈ in the reinforcement learning update strategy is 2; the graph in the node vector update module The number of embedding layers is L=3, and the loss coefficients λ ₁ , λ ₂ , and λ ₃ in the loss function are 2, 1, and 0.001, respectively. The dimensions of the weight coefficients of the three linear layers in the multilayer perceptron are R ^3d×d , R ^dX3d , and R ^d×2 in turn.

The method of this embodiment is compared with 5 methods: (1) Exact Name Match, if the entity to be aligned matches the name completely, it is considered that the two entities to be aligned are matched; (2) SVM, using the name of the expert, the affiliation , character-level 4-gram similarity of attributes such as frequently-occurring conference/journal names, paper title keywords, and co-author names; (3) COSNET, which uses the attributes of two entities as local factors, the relationship between entity pairs As a correlation factor, two aligned entities are considered to have the same label correspondingly, and a factor graph model is used to propagate this label information; (4) MEgo2Vec, mining potential matching entity pairs, as the nodes of the matching ego network, using a The multi-view node embedding method models the literal semantic features of different attributes, and uses the attention mechanism to distinguish the influence of different neighbor nodes, and adds the topological structure of the graph to obtain the regularized structural embedding; (5) LinKG, also uses the potential matching Entity pair information, however, the network nodes constructed by it are still entities, not entity pairs. It uses a multi-head attention mechanism based on node type to calculate the attention weight and complete the node vector update based on the relationship graph attention network. At the same time, two ablation experiments are designed: (1) this method - no reliability, only uses the graph embedding learning algorithm based on feature linear modulation; (2) this method - no reinforcement learning, on the basis of this method, the reinforcement learning dynamic is removed Update strategy; respectively verify the contribution of the graph embedding learning algorithm based on feature linear modulation, the reliability measurement method based on self-supervised signal and the node sampling number update strategy based on reinforcement learning to the model. Table 8 shows the performance comparison between the method of this implementation example and other comparative methods on the OAG dataset.

Table 8 Comparative experimental results of each method on the OAG dataset

Model Precision Recall F1-score Exact Name Match 44.48 80.63 57.33 SVM 84.70 92.22 88.30 COSNET 91.73 85.33 88.42 MEgo2Vec 91.03 90.82 90.92 LinKG 95.37 93.48 94.42 This method - unreliable 96.74 96.86 96.80 This method - no reinforcement learning 97.18 96.82 97.00 this method 96.58 97.84 97.20

The experimental results show that this method is significantly better than the five comparison methods in the three evaluation indicators, and the F1 value is improved by at least 2.78%. Through the experimental result of this method - no reliability, its F1 value is at least 2.38% higher than that of the five comparison methods, which proves the effectiveness of the graph embedding learning algorithm based on feature linear modulation. By comparing the experimental results of this method without reinforcement learning and this method without reliability, the former achieves a 0.20% improvement in the F1 value, which verifies the effectiveness of the reliability measurement method based on self-supervised signals. By comparing the experimental results of this method and this method without reinforcement learning, the former achieves a 0.20% improvement in the F1 value, which verifies the effectiveness of the node sampling number update strategy based on reinforcement learning.

The above-mentioned embodiments only represent several embodiments of the present invention, and the descriptions thereof are specific and detailed, but should not be construed as a limitation on the scope of the patent of the present invention. For those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can be made, which all belong to the protection scope of the present invention.

Claims (8)

1. A graph embedding expert entity alignment method based on reinforcement learning enhancement is characterized by comprising the following steps:

step 1: data G of two talent knowledge bases is obtained ₁ ＝(E ₁ ,T ₁ R) and G ₂ ＝(E ₂ ,T ₂ R), wherein E represents an entity set, and the entity types comprise expert entities and paper entities; r represents a relationship set; t represents a triple set, which is a subset of E multiplied by R multiplied by E;

step 2: for each expert entity e in a talent knowledge base, generating a candidate expert set C from another talent knowledge base through a candidate entity pair generation module based on a regular matching template generator method according to the expert names;

and step 3: for each candidate expert C ∈ C, construct about entity pairs<e,c>The 2-hop isomeric diagram HG ═ (V, H, T, R), where

H ₁ 、H ₂ Respectively representing entity initial vector sets of two talent knowledge bases; initial node vector representation h ⁽⁰⁾ ：

h _struct ＝LINE(G)

Wherein

Represents a vector splicing operation, h _attr Average vector of word vectors obtained for each attribute feature of an entity through a skip-gram model, h _struct The structure vector is obtained by coding the structure information of each entity in the knowledge base through a LINE model;

and 4, step 4: in a node vector updating module, in each layer of graph embedding layer, calculating the reliability of each edge by using a reliability measurement method based on an automatic supervision signal;

and 5: in each layer of graph embedding layer, for the heterogeneous subgraphs HG, a top-p sampling strategy is used, and according to the reliability calculated in the step 4, p is sampled for each relation from large to small _r The method comprises the steps of cutting reliable edges, and updating the node sampling number by using a node sampling number updating strategy based on reinforcement learning;

and 6: in each layer of graph embedding layer, after obtaining the sampled heterogeneous subgraph, updating the node vector by using a graph embedding learning algorithm based on characteristic linear modulation;

and 7: after the L-layer graph embedding layer, the updated entity pair to be aligned is taken out<e,c>Node vector of

And

computing match probabilities through a multi-tier perceptron

And 8: and according to the matching probabilities of all candidate entities, taking the candidate expert with the highest probability as the matching expert.

2. The graph embedding expert entity alignment method based on reinforcement learning enhancement as claimed in claim 1, wherein the method is trained by a training set before application, and is applied to expert entity alignment after training is completed;

in the course of the training process,the loss of the expert entity alignment task includes the alignment loss of all the entity pairs to be aligned, in addition to the average edge sampling loss of the reliability measurement method based on the self-supervision signal

The goal of this loss is to make the vector representation of the aligned pair of expert entities pair<z _e ,z _c >The higher the probability obtained after passing through the multi-layered perceptron, the closer to 1, while the non-aligned vectors represent a score of the pair close to 0.

3. The reinforcement learning enhancement based graph embedding expert entity alignment method according to claim 1, wherein the step 2 specifically comprises the following steps:

step 2-1: processing Chinese and English names respectively, and generating an analytic dictionary by using a name analyzer;

step 2-2: and taking the analysis dictionary generated by the name analyzer as the input of the template generator, and outputting a corresponding regular template.

4. The method for aligning the graph embedding expert entities based on reinforcement learning enhancement as claimed in claim 1, wherein in the step 2-1, for an English name, the analysis of the surname and the first name is performed first, whether the sequence of the surname and the first name is unique is judged, then an analysis dictionary is generated, for a Chinese name, the Chinese name is converted into a Pinyin form, namely, the English name, and meanwhile, the polyphone and the compound surname are considered, and the intermediate analysis result is generated and then processed with the English name.

5. The reinforcement learning enhancement-based graph embedding expert entity alignment method according to claim 1, wherein in step 2-2, first, whether the order of surnames and first names can be determined is determined, then, whether the first names are abbreviations is judged, and finally, 2 groups or 4 groups of regular templates are generated according to two orders of "first surname" and "last" respectively.

6. The method according to claim 1, wherein in the step 4, for the triplet < u, r, v >, that is, the relationship r exists between the node u and the node v, and the direction u points to v, the calculation process of the reliability S (u, v) specifically includes the following steps:

step 4-1: firstly, the feature expression eta of the nodes on each relation is calculated:

η _u ,η _v ＝σ ₁ (W _τ(u) h _u ),σ ₁ (W _τ(v) h _v ))

wherein W _τ(u) And W _τ(v) Respectively representing the weight matrix associated with the corresponding node type with dimension R ^|R|×d ,h _u And h _v Is the node vector representation of the nodes u and v at the embedding layer of the layer diagram, and the dimension is d, sigma ₁ (. represents an activation function;

step 4-2: computing node pairs<u,v>Probability a of existence of relation r _u,v ：

α _u,v ＝σ ₂ (W _TF ·(η _u ⊙η _v ))

Wherein W _TF Has a dimension of R ^2×|R| As a vector dot product operation, σ ₂ (. represents an activation function;

step 4-3: calculating the unreliability D (u, v) using the manhattan distance, and obtaining the reliability S (u, v):

D(u,v)＝||α _u,v || ₁

S(u,v)＝1-D(u,v)

step 4-4: in the training phase, in order to make the proposed reliability measurement method more effective in calculating the reliability of the edge, the reliability measurement method based on the self-supervision signal samples the partial relationship edge SPT (spherical segment) existing in the graph<u,r,v>|<u,r,v>E.g. T, as a positive example, sampling by a proportion of 1:1, and constructing a nonexistent relationship edge with a node of the target node, which does not have the relationship

As a negative example, the construction is based on the existing relationsEdge-tied self-supervision signal and using cross-entropy loss as a loss function to minimize the average edge sampling loss across all graph embedding layers

Thus, the various weight coefficients used in calculating the reliability are trained:

where ψ (u, r, v) ═ 1 represents that the side < u, r, v > is present, whereas ψ (u, r, v) ═ 0 represents that the side is absent.

7. The graph-embedded expert entity alignment method based on reinforcement learning enhancement as claimed in claim 1, wherein in the step 5), the reinforcement learning based node sampling number update strategy specifically comprises the following steps:

step 5-1: at each epoche, for the relation r, the sampled edge E is computed _r,sampled Mean all unreliability of

Wherein | E |) _r,sampled The number of edges with the relation of r among the sampled edges is referred to;

step 5-2: updating the node sampling number p of the relation r according to the variation trend between the average unreliability between two adjacent epochs _r The actions, Reward and Termination in the reinforcement learning strategy are respectively as follows:

Action＝{+∈,-∈}

wherein epsilon is a smaller fixed integer; if the average unreliability of the current epoch is less than the previous epoch, the reward function is positive, and the method of real-time reward is adopted to greedy increase p _r Otherwise, decrease p _r 。

8. The reinforcement learning enhancement based graph embedding expert entity alignment method according to claim 1, wherein the step 6 is specifically:

at level l, for node v, its node vector is composed of

Is updated to

And is

h _v ∈R ^d ：

Wherein W _r ^(l) ∈R ^d×d Is the message conversion function weight, sigma is the ReLU function,

the message weight of each neighbor node u calculated by the node v in the l-th layer graph embedding layer is respectively:

wherein g (h) _v ；θ _g,l,r ) Is a super-network, implemented using a single linear layer, theta _g,l,r ∈R ^2d×d Is a parameter of this super network.

Images