Heterogeneous graph embedding method based on information completion

通过在异构图中筛选和加权平均邻居节点的特征,学习相似度和生成目标矩阵,进行自适应权重训练和特征嵌入,解决了异构图中属性缺失和长尾问题,提升了模型性能和 The effect of downstream tasks.

CN118981554AActive Publication Date: 2024-11-19JIANGNAN UNIV

Patent Information

Application Number
CN202411021558.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-29
Publication Date
2024-11-19
Estimated Expiration
2044-07-29

AI Technical Summary

Technical Problem

When the existing technology deals with missing nodes in attributes in heterogeneous graphs, simple filling operations are prone to introduce noise, resulting in loss of model performance. The method of using the adjacency matrix for relationship completion cannot effectively solve the long-tail problem, resulting in a degradation of model performance. Influence the performance of downstream tasks.

Method used

Using the heterogeneous graph embedding method based on information completion, the target node set is determined, the initial feature vector is constructed, the first k neighbor nodes are filtered, the weighted average is performed, the similarity is learned, the target adjacency matrix is ​​generated, and the adaptive weight matrix is ​​trained. Feature embedding and neighborhood transfer are performed, and node representations are finally obtained through attention mechanism aggregation.

Benefits of technology

Effectively solve the feature completion problem of missing attribute nodes, improve the performance of the model in heterogeneous graph, especially for nodes with fewer relationship connections, learn more information, improve the performance of downstream tasks, and achieve more complete node feature representation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118981554A_ABST
    Figure CN118981554A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of graph data mining, in particular to a heterogeneous graph embedding method based on information completion, which comprises the following steps of: carrying out feature aggregation on target nodes to obtain preliminary completion feature vectors of the target nodes, and learning similarity between the nodes based on a relationship between the nodes in a heterogeneous graph to obtain a final completion feature vector of the target nodes; obtaining a target adjacent matrix of the target node in each relation; circularly updating the node features to obtain target feature vectors of the nodes; and performing neighborhood transfer on the target node based on the target adjacency matrix of the target node in each relation and the target feature vector of the node to obtain a complete and complemented feature vector of the target node, and performing aggregation by adopting an attention mechanism to obtain a final node representation of the target node. According to the method, the missing attribute features of the target node are complemented, the neighborhood information of the target node is perfected, and the target node with less relation connection can learn more information, so that more complete node feature representation is obtained, and the downstream task performance is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of graph data mining, and in particular to a heterogeneous graph embedding method based on information completion. Background Art

[0002] As a type of graph data structure, heterogeneous graphs have diverse node and edge types and carry complex and rich information, which poses a great challenge to data analysis. It also makes the analysis and mining of heterogeneous graph information a hot topic in academia. Among the many graph mining methods, graph embedding methods stand out. They can project data in high-dimensional networks into low-dimensional continuous space while capturing the intrinsic information of the network. Compared with other graph mining methods, their performance has certain advantages.

[0003] When using graph embedding methods to capture the neighborhood information of a network, the rich semantic information in node attributes is used to significantly enhance the representation ability of nodes, thereby achieving excellent performance in various graph tasks. However, in real-world scenarios, nodes in heterogeneous graphs can be divided into two categories: one is nodes with complete attributes, which have a complete set of attributes; the other is nodes with missing attributes. Since the model based on attribute feature aggregation is highly sensitive to node attributes, if the graph embedding method is directly applied to a heterogeneous graph with nodes with missing attributes, it will seriously affect the performance of the model based on attribute feature aggregation in the graph embedding method, resulting in the inability of the model based on attribute feature aggregation to accurately capture the features of the nodes. To solve this problem, the existing technology simply uses one-hot encoding or calculates and fills in with the help of other node information when dealing with such nodes with missing attributes. The node features obtained in this way often cannot accurately reflect the true characteristics of the nodes, and are prone to introduce noise information, affecting the performance of the model.

[0004] At present, there have been mature studies on the long-tail problem in homogeneous graphs, but the solution to the long-tail problem in homogeneous graphs cannot be directly applied to heterogeneous graphs. In the prior art, attention is paid to the imbalance of the number of relational connections between different nodes in heterogeneous graphs, so there are few studies considering that the node representation learned by the model is insufficient for nodes with fewer relational connections in heterogeneous graphs. In heterogeneous graphs, for nodes with more relational connections (usually called head nodes), their neighborhood information is relatively rich, and more features can be effectively aggregated; while for nodes with fewer relational connections (usually called tail nodes), their neighborhood information is relatively scarce, resulting in the model being unable to fully learn node representations, thereby reducing the performance of downstream tasks. In order to solve the above problems, the prior art uses adjacency matrices for relationship completion, which to a certain extent alleviates the long-tail problem in heterogeneous graphs, but simply increasing the connection of the tail node may introduce small neighborhoods with bias or insufficient representation, which will still lead to a decline in model performance, making it impossible for the model to learn the feature representation of the node well, thereby affecting the performance of downstream tasks, for example, inaccurate classification in node classification, poor node information aggregation in node clustering tasks, and inaccurate generation of node representations in link prediction. Summary of the invention

[0005] To this end, the technical problem to be solved by the present invention is to overcome the problem that in the prior art, only a simple filling operation is adopted for nodes with missing attributes, which easily introduces noise and causes loss of model performance; the method of using the adjacency matrix for relationship completion cannot solve the long tail problem existing in the heterogeneous graph well, and still causes the model performance to decline, thereby reducing the performance of downstream tasks.

[0006] In order to solve the above technical problems, the present invention provides a heterogeneous graph embedding method based on information completion, comprising:

[0007] A heterogeneous graph embedding method based on information completion, characterized by comprising:

[0008] Based on the heterogeneous graph, the target node set is determined and constructed; the initial feature vectors of all nodes in the heterogeneous graph are mapped to a unified shared semantic space to obtain the projected feature vectors of all nodes;

[0009] According to the order of the similarity of the attribute features between the current target node and its neighbor nodes, the first k neighbor nodes of the current target node are selected; the projected feature vectors of the current target node and its first k neighbor nodes are weighted averaged to obtain the preliminary completed feature vector of the current target node;

[0010] Based on the preliminary completed feature vectors of each target node and the projected feature vectors of each neighbor node of each target node, the similarity between nodes in the heterogeneous graph is learned, and the relational adjacency graph of each relationship in the heterogeneous graph is obtained; the initial adjacency matrix of each relationship is aggregated with the relational adjacency graph to obtain the target adjacency matrix of each relationship, thereby obtaining the target adjacency matrix of the current target node in the current relationship;

[0011] Based on all nodes in the heterogeneous graph, the adaptive weight matrix corresponding to each node in the initial round and the number of training times are set, and the adaptive feature vector of the current node in the current round is obtained based on the adaptive feature vector of the current node in the previous round and the adaptive weight matrix corresponding to the current node in the current round; until the preset number of training times is reached, the target feature vector of each node is obtained; wherein, when the training round is 1, the adaptive feature vector of the current target node in the previous round is its preliminary completion feature vector, and the adaptive feature vector of each neighbor node of the current target node in the previous round is its projected feature vector;

[0012] Based on the target adjacency matrix of the current target node in the current relationship, the target feature vectors of each neighbor node of the current target node in the current relationship are normalized to obtain the feature embedding vectors of each neighbor node of the current target node in the current relationship;

[0013] The target feature vector of the current target node and the feature embedding vectors of each neighbor node of the current target node in the current relationship are weighted averaged to obtain the neighborhood embedding vector of the current target node in the current relationship as the actual neighborhood of the current target node in the current relationship; the neighborhood transfer vector of the current target node in the current relationship is set, the ideal neighborhood of the current target node in the current relationship is defined, and the difference between the ideal neighborhood and the actual neighborhood of the current target node in the current relationship is determined; the difference between the ideal neighborhood and the actual neighborhood of the current target node in the current relationship is minimized by scaling and transfer transformation to obtain the neighborhood transfer vector of the current target node in the current relationship, and the neighborhood transfer vector of the current target node in each relationship is obtained in turn; based on the target feature vectors of the current target node and its each neighbor node and the neighborhood transfer vector of the current target node in each relationship, a complete completion feature vector of the current target node is obtained;

[0014] Based on the complete completed feature vector of each target node and the target feature vector of each neighboring node of each target node, the relationship strength between nodes in the heterogeneous graph is calculated, and the attention mechanism is used for aggregation to obtain the contextual representation of each target node, and residual connection is performed to obtain the final node representation of each target node.

[0015] Preferably, based on all nodes in the heterogeneous graph, the adaptive weight matrix corresponding to each node in the initial round and the number of training times are set, and based on the adaptive feature vector of the current node in the previous round and the adaptive weight matrix corresponding to the current node in the current round, the adaptive feature vector of the current node in the current round is obtained; until the preset number of training times is reached, the target feature vector of each node is obtained; wherein, when the training round is 1, the adaptive feature vector of the current target node in the previous round is its preliminary completed feature vector, and the adaptive feature vector of each neighbor node of the current target node in the previous round is its projected feature vector, including:

[0016] Based on the adaptive feature vector of the target node v in the (p-1) round and the adaptive weight matrix corresponding to the target node v in the p round The adaptive feature vector of the target node v in round p is obtained, and its expression is:

[0017]

[0018] in, Represents the adaptive feature vector of the target node v in round p; represents the adaptive feature vector of the target node v in round (p-1). In the initial round, i.e. when p=1, Represents the adaptive weight matrix corresponding to the target node v in round p;

[0019] Based on the adaptive feature vector of any neighbor node u of the target node v in round (p-1) and the adaptive weight matrix corresponding to the neighbor node u of the target node v in round p The adaptive feature vector of the neighbor node u of the target node v in round p is obtained, and its expression is:

[0020]

[0021] in, Represents the adaptive feature vector of neighbor node u of target node v in round p; represents the adaptive feature vector of the neighbor node u of the target node v in round (p-1). In the initial round, i.e. when p=1, Represents the adaptive weight matrix of neighbor node u of target node v in round p;

[0022] Until the preset number of training times is reached, the target feature vector h′ of the target node v is obtained v , the target feature vector h′ of the neighbor node u of the target node v u .

[0023] Preferably, the normalizing operation is performed on the target feature vectors of each neighbor node of the current target node in the current relationship based on the target adjacency matrix of the current target node in the current relationship to obtain the feature embedding vectors of each neighbor node of the current target node in the current relationship, including:

[0024] Based on the target adjacency matrix A′ of the target node v in the relation r v,r , the target feature vector h′ of the neighbor node u of the target node v in the relation r u Perform normalization to obtain the feature embedding vector of the neighbor node u of the target node v in the relationship r Its expression is:

[0025]

[0026] Among them, h′ u Represents the target feature vector of the neighbor node u of the target node v; Represents the normalized coefficient of the target node v in the relation r, and its expression is: D r A′ v,r The degree matrix of

[0027] Preferably, the target feature vector of the current target node and the feature embedding vectors of each neighbor node of the current target node in the current relationship are weighted averaged to obtain the neighborhood embedding vector of the current target node in the current relationship as the actual neighborhood of the current target node in the current relationship; the neighborhood transfer vector of the current target node in the current relationship is set, the ideal neighborhood of the current target node in the current relationship is defined, and the difference between the ideal neighborhood and the actual neighborhood of the current target node in the current relationship is determined, including:

[0028] The target feature vector h′ for the target node v v , the feature embedding vectors of each neighboring node of the target node v in the relation r are weighted averaged to obtain the neighborhood embedding vector of the target node v in the relation r, which is expressed as:

[0029]

[0030] in, Represents the neighborhood embedding vector of the target node v in the relation r and serves as the actual neighborhood of the target node v in the relation r; N v,r Represents the set of neighbor nodes of the target node v in the relation r; |N v,r | indicates N v,r The number of neighbor nodes in ; Represents the feature embedding vector of the neighbor node u of the target node v in the relation r;

[0031] Set the neighborhood transfer vector of the target node v in the relation r and define the ideal neighborhood of the target node v in the relation r Its expression is:

[0032]

[0033] Among them, z v,r represents the neighborhood transfer vector of the target node v in the relation r; h′ v represents the target feature vector of the target node v;

[0034] The ideal neighborhood of the target node v in the relation r The actual neighborhood The difference C v,r , whose expression is:

[0035]

[0036] Among them, C v,r Represents the difference between the ideal neighborhood and the actual neighborhood of the target node v in the relation r.

[0037] Preferably, the scaling and transfer transformation is used to minimize the difference between the ideal neighborhood and the actual neighborhood of the current target node in the current relationship to obtain the neighborhood transfer vector of the current target node in the current relationship, and the neighborhood transfer vector of the current target node in each relationship is obtained in turn, including:

[0038] In the l-th layer network, use the scaling and transfer transformation to obtain the neighborhood transfer vector of the target node v in the l-th layer network in the relationship r Its expression is:

[0039]

[0040] in, Represents the scaling vector of the target node v in the relation r of the l-th layer network; Represents the transfer vector of the target node v in the l-th layer network in the relationship r; Set The dimensions and The dimensions are the same; τ represents a unit vector; Represents the sum of neighborhood transfer vectors of all target nodes in the l-th layer network in relation r; Represents the learnable weights of the l-th layer network;

[0041] According to the neighborhood transfer vector of the target node v in each layer of the network in the relationship r, minimize the ideal neighborhood of the target node v in the relationship r The actual neighborhood The difference C v,r, get the neighborhood transfer vector z of the target node v in the relationship r v,r .

[0042] Preferably, the expression of the complete completion feature vector F[v] of the target node v is:

[0043]

[0044] Among them, h′ v represents the target feature vector of the target node v; z v,r represents the neighborhood transfer vector of the target node v in the relation r; M v represents the set of relationship types corresponding to the target node; hu represents the target feature vector of the neighbor node u of the target node v in the relationship r; N v,r Represents the set of neighbor nodes of the target node v in the relation r; Represents the sum of target feature vectors of each neighbor node in the neighbor node set of the target node v in the relation r; Represents the difference between the neighborhood transfer vector of the target node v in the relation r and the sum of the target feature vectors of each neighbor node in the neighbor node set.

[0045] Preferably, the screening of the top k neighbor nodes of the current target node by sorting the similarities of the attribute features between the current target node and each of its neighbor nodes comprises:

[0046] In the neighborhood of the target node v, determine the neighbor node set N of the target node v v , based on the cosine similarity calculation formula, determine the attribute feature similarity Γ between the target node v and any of its neighbor nodes u u,v , whose expression is:

[0047]

[0048] Among them, f u represents the projected feature vector of neighbor node u; f v represents the projected feature vector of the target node v; e represents the edge between the neighbor node u and the target node v;

[0049] Obtain the attribute feature similarity of the target node v and each neighbor node in its neighbor node set in turn; sort the attribute feature similarity of the target node v and each neighbor node in its neighbor node set from large to small, select the neighbor nodes corresponding to the first k attribute feature similarities of the target node v as the first k neighbor nodes of the target node v, and construct the first k neighbor node set N of the target node v v,k = {u1, u2, ..., u k}.

[0050] Preferably, performing weighted averaging on the projected feature vectors of the current target node and its first k neighbor nodes to obtain a preliminary completed feature vector of the current target node comprises:

[0051] For each neighbor node in the set of the first k neighbor nodes of the target node v, set the corresponding weight of each neighbor node; perform weighted average on the projected feature vectors of the first k neighbor nodes of the target node v and their corresponding weights to obtain the aggregated feature embedding value of the current target node Its expression is:

[0052]

[0053] Among them, u a Represents the first k neighbor node set N of the target node v v,k The ath neighbor node in ; Represents the first k neighbor node set N of the target node v v,k The weight corresponding to the ath neighbor node in ; Represents the first k neighbor node set N of the target node v v,k The projection feature vector of the ath neighbor node in; k represents the first k neighbor node set N of the target node v v,k The number of neighbor nodes in ;

[0054] Aggregate feature embedding value for target node v With its projected eigenvector f v Perform weighted averaging to obtain the initial completed feature vector of the target node v Its expression is:

[0055]

[0056] in, represents the aggregated feature embedding value of the target node v; f v Represents the projected feature vector of the target node v.

[0057] Preferably, the similarity between nodes in the heterogeneous graph is learned based on the preliminary completed feature vectors of each target node and the projected feature vectors of each neighbor node of each target node, and a relational adjacency graph of each relationship in the heterogeneous graph is obtained; the initial adjacency matrix of each relationship is aggregated with the relational adjacency graph to obtain a target adjacency matrix of each relationship, thereby obtaining a target adjacency matrix of the current target node in the current relationship.

[0058] In S31-S35, the tail node t m and t n is the target node; the initial completion feature vector of each target node All use f* Represents, that is, the initial completion feature vector of the target node v Use f v express;

[0059] S31: For a relation r in a heterogeneous graph, the first-order feature similarity between node features is obtained through metric learning, and its expression is:

[0060]

[0061] in, Indicates that the node under relationship r and the first-order feature similarity between node η; Y represents the number of heads; It represents the corresponding weight when the number of heads is y under the relation r; Representation Node Projected feature vector / preliminary completed feature vector; f η Represents the projected feature vector / preliminary completed feature vector of node η; if node is the target node, then Representation Node The initial completed feature vector of ; if node ζ is not the target node, then Representation Node The projected eigenvector of ; ⊙ represents the Hadamard product;

[0062] According to the first-order feature similarity Get the similarity graph Its expression is:

[0063]

[0064] Among them, ε F ∈[0, 1] represents the control similarity graph The sparsity parameter of

[0065] S32: Based on the relationship r, calculate the feature similarity graph between different types of nodes, that is, the source node s j and the tail node t m Similarity graph between Its expression is:

[0066]

[0067] in, Represents the source node s j and the tail node t m The first-order feature similarity between FB Represents the control similarity graph The sparsity parameter of

[0068] Based on the relationship r, the feature similarity graph between nodes of the same type is calculated, and its expression is:

[0069]

[0070] in, Represents the source node s i and j Feature similarity graph between ; Represents the tail node t m and t n Feature similarity graph between ; Represents the source node s i and j The first-order feature similarity between ; Represents the tail node t m and t n The first-order feature similarity between i 、f j Represents the source node s i 、s j The projected eigenvector of m 、f n Respectively represent the tail node t m ,t n The initial completion feature vector of FS , ε FT Represent the control similarity graph The sparsity parameter of

[0071] S33: Topological structure propagation through heterogeneous graphs, based on source node s i and j The feature similarity graph between the tail node t m and t n The feature similarity graph between them generates a potential propagation graph, which is expressed as:

[0072]

[0073] in, Respectively The corresponding propagation diagram; A r Represents the initial adjacency matrix of relation r;

[0074] S34: Using channel attention to feature similarity graph Fusion is performed to generate a relational adjacency graph of relation r Its expression is:

[0075]

[0076] in, Represents the weight corresponding to the relationship r;

[0077] S35: Connect the relation adjacency graph of relation r The initial adjacency matrix A of the relationship r r Aggregate and get the target adjacency matrix of relation r, which is expressed as:

[0078]

[0079] Among them, softmax(·) means that w ψ,r The weights are normalized; thus, the target adjacency matrix of the target node v in the relationship r can be obtained as A′ v,r .

[0080] Preferably, the method of calculating the relationship strength between nodes in the heterogeneous graph based on the complete completion feature vector of each target node and the target feature vector of each neighboring node of each target node, and aggregating using the attention mechanism to obtain the context representation of each target node, and performing residual connection to obtain the final node representation of each target node includes:

[0081] In S51-S54, the target feature vector h′ of each neighboring node of each target node is * They are all represented by F[*], that is, the target feature vector h′ of the neighbor node u of the target node v u Denoted by F[u];

[0082] S51: For relation triples The target node is the tail node t, and the neighbor node of the tail node t is the source node s. The relationship strength between the source node s and the tail node t is calculated to obtain the attention value ATT-head(s, e, t) of the relationship, which is expressed as:

[0083] K(s)=K-Linearφ(s)(F[s])

[0084] Q(t)=Q-Linearφ(t)(F[t])

[0085]

[0086] Among them, φ(s) represents the type of source node s; φ(t) represents the type of tail node t, and tail node t is the target node; represents the type of connection relationship e between the source node s and the tail node t; K(s) represents the strength of the relationship between the source node s; Q(t) represents the strength of the relationship between the tail node t; F[s] represents the target feature vector of the source node s; F[t] represents the complete complement feature vector of the tail node t; Linear φ (*) indicates linear projection, which transforms the dimension of the source node s or the tail node t from R dbecome d represents the vector dimension, δ represents the number of attention heads, and d / δ represents the vector dimension of each head; Represents the parameter for adjusting the attention value; Indicates the weight corresponding to ATT-head(s, e, t);

[0087] S52: Connect the δ attention heads together to obtain the attention vector of each node pair, which is expressed as:

[0088]

[0089] Among them, the Softmax function is used to normalize the attention distribution of the source node s adjacent to the tail node t; s∈N(t) indicates that the source node s belongs to the set of neighbor nodes of the tail node t;

[0090] S53: Combine the attention vector of each node pair with the message vector from the adjacent nodes to obtain the context representation of each target node, which is expressed as:

[0091]

[0092] Among them, the tail node t is the target node; represents the context representation of the tail node t in the ξ-th layer relation stack; represents the message vector from the neighboring nodes; Indicates MSG-head γ The weight corresponding to (s, e, t);

[0093] S54: Map the context representation of each target node to the distribution of the node type corresponding to each target node, and perform residual connection to obtain the final node representation of each target node, which is expressed as:

[0094]

[0095] Among them, the tail node t is the target node; H (ξ) [t] represents the final node representation of the tail node t in the ξ-th layer relation stack; H (ξ-1) [t] represents the final node representation of the tail node t in the (ξ-1)th layer of the relationship stack; ξ represents the number of layers of the relationship stack; A-Linear φ(t) Represents a linear projection.

[0096] The above technical solution of the present invention has the following beneficial effects compared with the prior art:

[0097] (1) The heterogeneous graph embedding method based on information completion described in the present invention selects the neighbor nodes corresponding to the first k attribute similarities of each target node according to the attribute similarity between each target node and each neighbor node in the neighborhood of each target node, and then completes the features of each target node with missing attributes by aggregating the attribute information of the first k neighbor nodes of each target node, so as to obtain the preliminary completed feature vector of each target node; since in a heterogeneous graph, nodes usually carry rich feature information and can be used to learn the missing attributes of other nodes, the nodes most similar to each target node with missing attributes can be used to improve the quality of each node data, thereby preliminarily solving the problem of missing node attributes and laying a foundation for the subsequent model to better learn the feature representation of nodes;

[0098] (2) The heterogeneous graph embedding method based on information completion described in the present invention, based on an adaptive weight mechanism, multiplies the preliminary completed feature vector of each target node and the projected feature vector of each neighbor node of each target node with their corresponding adaptive weights to obtain the adaptive feature vector of each target node and the adaptive feature vector of each neighbor node of each target node, which can enable the model to better learn from the node data; at the same time, based on the target adjacency matrix of each type of relationship corresponding to each target node, the adaptive feature vector of each neighbor node of each target node is normalized to obtain the feature embedding vector of each neighbor node of the target node in each type of relationship; the target adjacency matrix is ​​obtained by analyzing the connection information between nodes, mining feature similarity, and applying channel attention The mechanism helps the model to understand the relationship between nodes more comprehensively and improve the accuracy of subsequent acquisition of neighborhood information of each target node. In addition, the method also takes neighborhood transfer into account, and improves the neighborhood information of the target node by minimizing the difference between the ideal neighborhood and the actual neighborhood of the target node. Since the neighborhood information of the target head node is relatively rich and is not easily affected by information loss and information error, the target head node can be used as the learning object of the target tail node. Therefore, by learning the neighborhood transfer vector of the target head node, the ideal neighborhood information of the target tail node can be obtained, the complete completion operation of the target node is realized, and the long-tail problem is solved, so that the target nodes with fewer relationship connections can learn more information, so as to obtain a more complete node feature representation, thereby improving the performance of downstream tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:

[0100] Figure 1 It is a flow chart of a heterogeneous graph embedding method based on information completion provided by the present invention;

[0101] Figure 2 It is the schematic diagram of the relationship diagram generator;

[0102] Figure 3 It is a schematic diagram of the relational level aggregator;

[0103] Figure 4 It is the HGNN-I C model diagram;

[0104] Figure 5 is the link prediction result graph;

[0105] Figure 6 is the visualization result diagram; Figure 6 (a) is the result graph of selecting 1500 nodes on the ACM dataset. Figure 6 (b) is the result graph of selecting 500 nodes on the IMDB dataset. Figure 6 (c) is the result graph of selecting 1500 nodes on the DBLP dataset;

[0106] Figure 7 is the result diagram of the ablation experiment; Figure 7 (a) is the comparison result of Macro-F1 value on three datasets: ACM, IMDB, and DBLP. Figure 7 (b) is the comparison result of Micro-F1 values ​​on three datasets: ACM, IMDB, and DBLP. DETAILED DESCRIPTION

[0107] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0108] Reference Figure 1 As shown, Figure 1 A flowchart of a heterogeneous graph embedding method based on information completion provided by the present invention; specifically comprising:

[0109] S1: Based on the heterogeneous graph, determine and construct the target node set; map the initial feature vectors of all nodes in the heterogeneous graph to a unified shared semantic space to obtain the projected feature vectors of all nodes; in this process, unify the node feature vectors through the node feature projection module to ensure that nodes of various types have consistent representations;

[0110] Among them, the initial feature vector of node θ is mapped to a unified shared semantic space to obtain the projected feature vector of node u, which is expressed as:

[0111]

[0112] in, represents the projected feature vector of node θ; represents the initial feature vector of node θ; φ(θ) represents the type of node θ; σ(·) represents the nonlinear activation function; represents the projection matrix corresponding to the φ(θ) type node; b φ (θ)∈R 1×d Indicates the bias parameter corresponding to the φ(θ) type node;

[0113] S2: Sort the attribute feature similarities between the current target node and its neighbor nodes, and select the first k neighbor nodes of the current target node; perform weighted average on the projected feature vectors of the current target node and its first k neighbor nodes to obtain the preliminary completed feature vector of the current target node, including:

[0114] S21: Based on each neighbor node in the neighborhood of the current target node, combined with the cosine similarity calculation formula, the attribute feature similarity between the current target node and its neighbor nodes is obtained; the attribute feature similarities corresponding to the current target node are sorted from large to small, the first k attribute feature similarities of the current target node are selected, and the first k neighbor nodes of the current target node are obtained, that is:

[0115] In the neighborhood of the target node v, determine the neighbor node set N of the target node v v , based on the cosine similarity calculation formula, determine the attribute feature similarity Γ between the target node v and any of its neighbor nodes u u,v , whose expression is:

[0116]

[0117] Among them, f u represents the projected feature vector of neighbor node u; f v represents the projected feature vector of the target node v; e represents the edge between the neighbor node u and the target node v;

[0118] Obtain the attribute feature similarity of the target node v and each neighbor node in its neighbor node set in turn; sort the attribute feature similarity of the target node v and each neighbor node in its neighbor node set from large to small, select the neighbor nodes corresponding to the first k attribute feature similarities of the target node v as the first k neighbor nodes of the target node v, and construct the first k neighbor node set N of the target node v v,k = {u1, u2, ..., u k};

[0119] S22: Perform weighted averaging on the projected feature vectors of the first k neighbor nodes of the current target node and their corresponding weights to obtain the aggregated feature embedding value of the current target node, and then perform weighted averaging on the projected feature vector of the current target node to obtain the preliminary completed feature vector of the current target node, that is:

[0120] For each neighbor node in the set of the first k neighbor nodes of the target node v, set the corresponding weight of each neighbor node; perform weighted average on the projected feature vectors of the first k neighbor nodes of the target node v and their corresponding weights to obtain the aggregated feature embedding value of the current target node Its expression is:

[0121]

[0122] Among them, u a Represents the first k neighbor node set N of the target node v v,k The ath neighbor node in ; Represents the first k neighbor node set N of the target node v v,k The weight corresponding to the ath neighbor node in ; Represents the first k neighbor node set N of the target node v v,k The projection feature vector of the ath neighbor node in; k represents the first k neighbor node set N of the target node v v,k The number of neighbor nodes in ;

[0123] Aggregate feature embedding value for target node v With its projected eigenvector f v Perform weighted averaging to obtain the initial completed feature vector of the target node v Its expression is:

[0124]

[0125] in, represents the aggregated feature embedding value of the target node v; f v Represents the projected feature vector of the target node v;

[0126] In a heterogeneous graph, nodes usually carry rich feature information, which can be used to learn the missing attributes of other nodes; however, a heterogeneous graph usually contains a large number of nodes, and directly using all nodes to predict the missing attributes of the target node is likely to lead to excessively high computational costs, resulting in excessive consumption of memory and time; in order to avoid these problems, the present invention implements a preliminary completion operation of the attribute features of the target node with missing attributes in S2 by weighted aggregation of the projected node features by k nearest neighbors (KNN method), and this process can effectively select the node most similar to the target node, thereby generating a feature embedding of the target node; the feature aggregation operation effectively aggregates the attribute information of similar neighbor nodes of the target node, thereby implementing a preliminary completion of the attribute features of the target node v;

[0127] S3: Based on the preliminary completed feature vectors of each target node and the projected feature vectors of each neighboring node of each target node, learn the similarity between each node and obtain the relational adjacency graph of various relations; aggregate the initial adjacency matrix of various relations with the relational adjacency graph to obtain the target adjacency matrix of various relations, thereby obtaining the target adjacency matrix of the current target node corresponding to the current relation, including:

[0128] In a heterogeneous graph, the more similar the features of two nodes are in the feature vector space, the closer the nodes are to each other in the embedding space, the higher the probability that they belong to the same type, and the higher the probability that there is an edge between the two similar nodes. Based on the topological structure of the graph, the present invention introduces a relationship graph generator, which uses the relationship between nodes to learn the similarity between node features. The principle is as follows Figure 2 As shown; Figure 2 Contains only one relationship r represents the type of relationship between a node, which covers the following three relationship triplets: i , e1, t m >, j , e2, t m >, j ,e3,t n >

[0129] In S31-S35, the tail node t m and t n is the target node; the initial completion feature vector of each target node All use f * Represents, that is, the initial completion feature vector of the target node v Use f v express;

[0130] S31: For a relation r in a heterogeneous graph, the first-order feature similarity between node features is obtained through metric learning, and its expression is: ​​​

[0131]

[0132] in, Indicates that the node under relationship r and the first-order feature similarity between node η; Y represents the number of heads; It represents the corresponding weight when the number of heads is y under the relation r; Representation Node Projected feature vector / preliminary completed feature vector; f η Represents the projected feature vector / preliminary completed feature vector of node η; if node is the target node, then Representation Node The initial completed feature vector of ; if node ζ is not the target node, then Representation Node The projected eigenvector of ; ⊙ represents the Hadamard product;

[0133] According to the first-order feature similarity Get the similarity graph Its expression is:

[0134]

[0135] Among them, ε F ∈[0, 1] represents the control similarity graph The sparsity parameter of F The closer the value of is to 1, the sparser the heterogeneous graph is, which means that very similar nodes are retained; if ε F is smaller, the heterogeneous graph will contain more non-zero elements, reflecting more similar nodes;

[0136] S32: Figure 2 As shown, according to the above steps, based on the relationship r, the node Node η is replaced by source node s j , tail node t m , we can get the source node s j and the tail node t m Similarity graph between

[0137] For the same type of relationship Calculate the feature similarity graph between different types of nodes, that is, the source node s j and the tail node t m Similarity graph between Its expression is:

[0138]

[0139] in, Represents the source node s j and the tail node t m The first-order feature similarity between FB Represents the control similarity graph The sparsity parameter of

[0140] Two nodes with similar characteristics are likely to have similar neighbors, so for the same type of relationship It is necessary to study its high-order similarity, that is, to find the neighbors of nodes with similar characteristics through the structure of heterogeneous graphs; where the same type of nodes can be source nodes or tail nodes;

[0141] Based on the relationship r, the feature similarity graph between nodes of the same type is calculated, and its expression is:

[0142]

[0143] in, The calculation method of similar; Represents the source node s i and j Feature similarity graph between ; Represents the tail node t m and t n Feature similarity graph between ; Represents the source node s i and j The first-order feature similarity between ; Represents the tail node t m and t n The first-order feature similarity between i 、f j Represents the source node s i 、s j The projected eigenvector of m 、f n Respectively represent the tail node t m ,t n The initial completion feature vector of FS , ε FT Represent the control similarity graph The sparsity parameter of

[0144] S33: Topological structure propagation through heterogeneous graphs, based on source node s i and j The feature similarity graph between the tail node t m and t n The feature similarity graph between them generates a potential propagation graph, which is expressed as:

[0145]

[0146] in, Respectively The corresponding propagation diagram; A r Represents the initial adjacency matrix of relation r;

[0147] S34: Using channel attention to feature similarity graph Fusion is performed to generate a relational adjacency graph of relation r Its expression is:

[0148]

[0149] in, Represents the weight corresponding to the relationship r;

[0150] S35: In order to more comprehensively consider the impact of relations on the graph and further improve the model performance, the relation adjacency graph of relation r is The initial adjacency matrix A of the relationship r r Aggregate and get the target adjacency matrix of relation r, which is expressed as:

[0151]

[0152] Among them, softmax(·) means that w ψ,r The weights are normalized;

[0153] In summary, the target adjacency matrix of the target node v corresponding to the relationship r is A′ v,r ;

[0154] In S3, by analyzing the connection information between nodes, mining feature similarity, and applying the channel attention mechanism, a new relational adjacency matrix is ​​learned; the complete graph structure information helps to more fully understand the relationship between nodes, including complex information between different node types;

[0155] S4: Based on the new relationship graph learned in S3, perform neighborhood transfer to complete the attribute features of the target node, including:

[0156] S41: Based on all nodes in the heterogeneous graph, the adaptive weight matrix corresponding to each node in the initial round and the number of training times are set, and based on the adaptive feature vector of the current node in the previous round and the adaptive weight matrix corresponding to the current node in the current round, the adaptive feature vector of the current node in the current round is obtained; until the preset number of training times is reached, the target feature vector of each node is obtained; wherein, when the training round is 1, the adaptive feature vector of the current target node in the previous round is its preliminary completion feature vector, and the adaptive feature vector of each neighbor node of the current target node in the previous round is its projected feature vector, including:

[0157] Based on the adaptive feature vector of the target node v in the (p-1) round and the adaptive weight matrix corresponding to the target node v in the p round The adaptive feature vector of the target node v in round p is obtained, and its expression is:

[0158]

[0159] in, Represents the adaptive feature vector of the target node v in round p; represents the adaptive feature vector of the target node v in round (p-1). In the initial round, i.e. when p=1, Represents the adaptive weight matrix corresponding to the target node v in round p;

[0160] Based on the adaptive feature vector of any neighbor node u of the target node v in round (p-1) and the adaptive weight matrix corresponding to the neighbor node u of the target node v in round p The adaptive feature vector of the neighbor node u of the target node v in round p is obtained, and its expression is:

[0161]

[0162] in, Represents the adaptive feature vector of neighbor node u of target node v in round p; represents the adaptive feature vector of the neighbor node u of the target node v in round (p-1). In the initial round, i.e. when p=1, Represents the adaptive weight matrix of neighbor node u of target node v in round p;

[0163] Until the preset number of training times is reached, the target feature vector h′ of the target node v is obtained v , the target feature vector h′ of the neighbor node u of the target node v u ;

[0164] S42: Based on the target adjacency matrix of the current target node in the current relationship, a target feature vector of each neighbor node of the current target node in the current relationship is normalized to obtain a feature embedding vector of each neighbor node of the current target node in the current relationship, including:

[0165] Based on the target adjacency matrix A′ of the target node v in the relation r v,r , normalize the target feature vector hu of the neighbor node u of the target node v in the relation r, and obtain the feature embedding vector of the neighbor node u of the target node v in the relation r Its expression is:

[0166]

[0167] Among them, h′ u Represents the target feature vector of the neighbor node u of the target node v; Represents the normalized coefficient of the target node v in the relation r, and its expression is: D r A′ v,r The degree matrix of

[0168] Among them, the use of adaptive weight allocation mechanism can enable the model to learn better from node data;

[0169] S43: performing weighted averaging on the target feature vector of the current target node and the feature embedding vectors of each neighboring node of the current target node in the current relationship, obtaining the neighborhood embedding vector of the current target node in the current relationship as the actual neighborhood of the current target node in the current relationship; setting the neighborhood transfer vector of the current target node in the current relationship, defining the ideal neighborhood of the current target node in the current relationship, and determining the difference between the ideal neighborhood and the actual neighborhood of the current target node in the current relationship, including:

[0170] The target feature vector hj of the target node v and the feature embedding vectors of each neighboring node of the target node v in the relationship r are weighted averaged to obtain the neighborhood embedding vector of the target node v in the relationship r, which is expressed as:

[0171]

[0172] in, Represents the feature embedding vector of the target node v in the relation r and serves as the actual neighborhood of the target node v in the relation r; N v,r Represents the set of neighbor nodes of the target node v in the relation r; |N v,r | indicates N v,r The number of neighbor nodes in ; Represents the feature embedding vector of the neighbor node u of the target node v in the relation r;

[0173] Based on the relational transformation model, the neighborhood transfer operation is used to model the relationship between the target node v and its neighborhood, which is expressed as:

[0174]

[0175] Among them, z v,r represents the neighborhood transfer vector of the target node v in the relation r, which is used to model the missing information in the neighborhood; h′ v represents the target feature vector of the target node v;

[0176] Set the neighborhood transfer vector of the target node v in the relation r and define the ideal neighborhood of the target node v in the relation r Its expression is:

[0177]

[0178] in, represents the neighborhood transfer vector of the target node v in the relation r; h′ v represents the target feature vector of the target node v;

[0179] The ideal neighborhood of the target node v in the relation r The actual neighborhood The difference C v,r , whose expression is:

[0180]

[0181] Among them, C v,r Represents the difference between the ideal neighborhood and the actual neighborhood of the target node v in the relation r;

[0182] If the target node is the head node, since their neighborhood information in the heterogeneous graph is relatively complete, the difference C v,r should be close to 0; and if the target node is the tail node, their difference is:

[0183] The neighborhood information of the head node is relatively rich and is not easily affected by information loss and information error. Although their neighborhood structure is not the optimal neighborhood, it can be used as the object of learning for the tail node. Therefore, by learning the neighborhood transfer vector of the head node, the ideal neighborhood information of the tail node can be obtained.

[0184] S44: Minimize the difference between the ideal neighborhood and the actual neighborhood of the current target node in the current relationship by using scaling and transfer transformation, obtain the neighborhood transfer vector of the current target node in the current relationship, and obtain the neighborhood transfer vector of the current target node in each relationship in turn, including:

[0185] In the l-th layer network, use the scaling and transfer transformation to obtain the neighborhood transfer vector of the target node v in the l-th layer network in the relationship r Its expression is:

[0186]

[0187] in, Represents the scaling vector of the target node v in the relation r of the l-th layer network; Represents the transfer vector of the target node v in the l-th layer network in the relationship r; Set The dimensions and The dimensions of are the same, in order to facilitate scaling and transfer in an element-wise manner; τ represents a unit vector; Represents the sum of neighborhood transfer vectors of all target nodes in the l-th layer network in relation r; Represents the learnable weights of the l-th layer network;

[0188] According to the neighborhood transfer vector of the target node v in each layer of the network in the relationship r, minimize the ideal neighborhood of the target node v in the relationship r The actual neighborhood The difference C v,r , get the neighborhood transfer vector z of the target node v in the relationship r v,r ;

[0189] S45: The expression of the complete completion feature vector F[v] of the target node v is:

[0190]

[0191] Among them, h′ v represents the target feature vector of the target node v; z v,r represents the neighborhood transfer vector of the target node v in the relation r; M v represents the set of relationship types corresponding to the target node v; hu represents the target feature vector of the neighbor node u of the target node v in the relationship r; N v,r Represents the set of neighbor nodes of the target node v in the relation r; Represents the sum of target feature vectors of each neighbor node in the neighbor node set of the target node v in the relation r; Represents the difference between the neighborhood transfer vector of the target node v in the relation r and the sum of the target feature vectors of each neighbor node in the neighbor node set;

[0192] Among them, in order to improve the efficiency of the model, a neighborhood transfer vector is set, which can represent a simple interactive relationship between nodes; in a heterogeneous graph, different types of nodes have different relationship structures, and the neighborhood transfer vector needs to reflect the different contextual information of the nodes, that is, each node itself and the unique neighbor information; through calculation, the neighborhood transfer vector is obtained, the neighborhood information of the tail node is obtained, and the information gap between the head node and the tail node is balanced, so as to obtain the feature representation of the tail node, and then the feature representation of each node can be obtained;

[0193] S5: Based on the complete completion feature vector of each target node and the target feature vector of each neighboring node of each target node, the relationship strength between nodes in the heterogeneous graph is calculated, and the attention mechanism is used for aggregation to obtain the context representation of each target node, and residual connections are performed to obtain the final node representation of each target node, including:

[0194] For the target node, information from different types of neighbor objects may have different effects; therefore, in order to effectively utilize different relationships, such as Figure 3 As shown in Figure 1, a new node feature representation F is obtained by using neighborhood transfer, the importance weights of different types of neighbor objects are learned, and the attention mechanism is used to aggregate the information of the source node to obtain the contextual representation of the target node. Figure 3 There are two different relationships in And the relationship triple<s,e1,t> ,<h,e2,t> Corresponding to them respectively;

[0195] In S51-S54, the target feature vector h′ of each neighboring node of each target node is * They are all represented by F[*], that is, the target feature vector h′ of the neighbor node u of the target node v u Denoted by F[u];

[0196] S51: For relation triples The target node is the tail node t, and the neighbor node of the tail node t is the source node s. The relationship strength between the source node s and the tail node t is calculated to obtain the attention value ATT-head(s, e, t) of the relationship, which is expressed as:

[0197] K(s)=K-Linear φ(s) (F[s])

[0198] Q(t)=Q-Linear φ(t) (F[t])

[0199]

[0200] Among them, φ(s) represents the type of source node s; φ(t) represents the type of tail node t, and tail node t is the target node; represents the type of connection relationship e between the source node s and the tail node t; K(s) represents the strength of the relationship between the source node s; Q(t) represents the strength of the relationship between the tail node t; F[s] represents the target feature vector of the source node s; F[t] represents the complete complement feature vector of the tail node t; Linear φ(*) Represents a linear projection, which changes the dimension of the source node s or the tail node t from R d become d represents the vector dimension, δ represents the number of attention heads, and d / δ represents the vector dimension of each head; Represents the parameters for adjusting the attention value to adapt to different relationships; represents the weight corresponding to ATT-head(s, e, t); since there are multiple relationships between node pairs in the heterogeneous graph, we use Can adapt to different sides;

[0201] S52: In order to more comprehensively capture the information of different types of neighbor nodes, the δ attention heads are connected together to obtain the attention vector of each node pair, which is expressed as:

[0202]

[0203] Among them, the Softmax function is used to normalize the attention distribution of the source node s adjacent to the tail node t; s∈N(t) represents the set of neighbor nodes of the source node s belonging to the tail node t; for each target node, the Softmax processing is performed on the attention values ​​of all its neighbor nodes to ensure that the attention distribution of the neighbor nodes satisfies the normalization, so as to effectively perform weighted averaging on different neighbor nodes;

[0204] S53: Combine the attention vector of each node pair with the message vector from the adjacent nodes to obtain the context representation of each target node, which is expressed as:

[0205]

[0206] Among them, the tail node t is the target node; represents the context representation of the tail node t in the ξ-th layer relation stack; represents the message vector from the neighboring nodes; Indicates MSG-head γ The attention vector weights the information of different types of neighbor nodes, ensuring that more attention is paid to the neighbor nodes that have the greatest impact on the target node.

[0207] S54: Map the context representation of each target node to the distribution of the node type corresponding to each target node, and perform residual connection to obtain the final node representation of each target node, which is expressed as:

[0208]

[0209] Among them, the tail node t is the target node; H (ξ) [t] represents the final node representation of the tail node t in the ξ-th layer relation stack; H (ξ-1) [t] represents the final node representation of the tail node t in the (ξ-1)th layer of the relationship stack; ξ represents the number of layers of the relationship stack; A-Linear φ(t) Represents a linear projection.

[0210] Based on the above steps S1-S5, in the process of completing the attributes of the heterogeneous graph, the target node is the node in the heterogeneous graph whose attribute data is incomplete and needs to be completed; after completing the attribute completion of the target node in the heterogeneous graph, the node representation of the entire heterogeneous graph can be obtained through multi-layer stacking relationship-level aggregation modules, so that it can be used for downstream tasks. In the downstream tasks, the target node is the node that needs to be classified and clustered. Experiments;

[0211] In order to solve the problem that the existing technology only takes simple filling operations for nodes with missing attributes, which easily introduces noise and causes model performance loss; the method of using adjacency matrix for relationship completion cannot solve the long tail problem in heterogeneous graphs well, and still leads to model performance degradation, thereby reducing the performance of downstream tasks, feature aggregation and neighborhood transfer modules are introduced. Based on steps S1-S5, a heterogeneous graph embedding model (HGNN-IC model) based on information completion is constructed. Specifically, Figure 4 As shown in the figure; the HGNN-IC model takes the original heterogeneous graph as input. First, by transforming and projecting the node features and weightedly aggregating the k nearest neighbors, the preliminary completion operation of the target node attribute features with missing attributes is achieved; secondly, the graph is updated using the relationship graph generator and the topological information between the nodes; thirdly, based on the learned new graph, the neighborhood transfer is performed to achieve the complete completion operation of the target node attribute features; finally, the relationship-level aggregation module aggregates the previous features to obtain the node representation of the entire graph; in order to demonstrate the excellent performance of the HGNN-IC model provided by the present invention, extensive experiments were conducted on three typical heterogeneous information network datasets, and detailed comparisons were made with multiple baseline models. The HGNN-IC model is significantly superior to the most advanced models in downstream tasks such as node classification and clustering.

[0212] The experiments were conducted using the ACM, IMDB, and DBLP datasets. Table 1 summarizes the specific data of the three datasets. To facilitate the study of this method, nodes are randomly selected in the same proportion and their attributes are all set to empty to construct nodes with missing attributes, so as to more accurately evaluate the performance of this method.

[0213] Table 1 Dataset

[0214]

[0215] The ACM dataset comes from citation network data and covers three types of objects: authors (A), papers (P), and topics (S). Papers (P) are divided into three research fields: databases, data mining, and wireless communications, and are used as labels for papers; the features of papers are represented by bags of words consisting of their keywords, and the feature vector of authors is the average of the features of their connected papers; the features of topics are encoded using one-hot encoding.

[0216] After preprocessing, IMDB contains three types of objects: movies (M), actors (A), and directors (D). Movies (M) include three types of tags: action movies, comedies, and dramas; the features of movies are represented by bags of words consisting of their plot keywords, while the features of actors and directors are the averages of the corresponding movie features.

[0217] The DBLP dataset contains four types of objects: papers (P), authors (A), conferences (C), and terms (T). Authors (A) are divided into four categories according to their research fields: database, data mining, artificial intelligence, and information retrieval; the paper feature is the bag-of-words representation of the paper title, the author feature is the bag-of-words representation of keywords extracted from their published papers, the term feature is represented by pre-trained word vectors, and the conference feature is represented by one-hot vectors.

[0218] Experiment 1: Node Classification

[0219] The present invention focuses on the paper node of the ACM dataset, the author node of the DBLP dataset, and the movie node of the IMDB dataset to evaluate the performance of the HGNN-IC model in the node classification task; in the experiment, the present invention adopts different training ratios, specifically 20%, 40%, 60%, and 80%, and divides the target nodes according to these ratios, while the remaining nodes are used as validation sets and test sets; the present invention adopts Macro-F1 and Micro-F1 as evaluation indicators to evaluate the performance of different models in the node classification task; the node classification experimental results are shown in Table 2.

[0220] Table 2 Node classification experimental results

[0221]

[0222] Overall, the F1 scores of the HGNN-IC model on the three datasets are higher than those of other models, and its performance has been significantly improved; the HGNN-AC model has been optimized for the attribute missing problem, and compared with the Metapath2vec model, its performance on the ACM, IMDB and DBLP datasets has improved by 22%, 26% and 8.8% respectively, showing excellent performance; although the Tai l-GNN model compensates for the difference between the head and tail nodes under the guidance of the ideal neighborhood of the target head node, compared with other heterogeneous graph embedding models, Tai l-GNN fails to utilize high-order neighbor node information, and the embedding effect is limited, especially on the ACM dataset, where its performance shows a downward trend compared with the HGNN-IC model, decreasing by 5.2% and 5.1% respectively; compared with HRMNN, the HGNN-IC model effectively improves the performance in graph embedding, especially on the IMDB dataset, with improvements of 1.7% and 1.6% respectively.

[0223] Experiment 2: Node Clustering

[0224] The present invention uses K-Means algorithm for clustering; for the ACM dataset, the paper nodes are divided into training set (808), validation set (401), and test set (2816); for the DBLP dataset, the author nodes are divided into training set (400), validation set (400), and test set (3257); for the IMDB dataset, the movie nodes are divided into training set (400), validation set (400), and test set (3478); the present invention uses two evaluation indicators, normalized mutual information (NMI) and adjusted Rand index (ARI), to evaluate the performance of different algorithms on the node clustering task; the results of node clustering are shown in Table 3.

[0225] Table 3 Results of node clustering

[0226]

[0227] It can be observed from Table 3 that the HGNN-IC model shows superior performance on multiple datasets. Specifically, on the ACM dataset, the NM I value of HGNN-IC reaches 0.68 and the ARI is 0.66. On the I MDB dataset, its NM I and ARI are 0.12 and 0.13, respectively. On the DBLP dataset, they are 0.61 and 0.63, respectively. The overall performance is better than other models. GCN and Node2vec, as homogeneous graph models, do not perform well on heterogeneous graphs, especially compared with the Metapath2vec model specially designed to process heterogeneous graphs. On the ACM dataset, the performance of GCN and Node2vec decreased by 14% and 8.9%, respectively. On the I On the MDB dataset, the performance dropped by 30% and 25.5%; although the Metapath2vec model designs meta-paths for heterogeneous graphs to guide random walks, the performance is still limited due to the defects in its meta-path definition and the dependence of random walk behavior on the node structure; in addition, the existence of tail nodes also affects the performance of Metapath2vec; other models, such as HGT and HAN, although they all aggregate node information and consider the information carried by neighboring nodes and connections, have improved their performance by 33.1% and 28% compared with Metapath2vec; but compared with the HGNN-IC model, the performance of HGT and HAN dropped by 7.63% and 10.49% respectively, and there is still a certain gap.

[0228] Experiment 3: Link Prediction

[0229] When conducting link prediction experiments, the present invention uses the edge between the author node and the paper node in the DBLP dataset to demonstrate the performance of the HGNN-IC model; the present invention regards the connected paper-author pairs as positive node pairs and all unconnected paper-author pairs as negative node pairs; the present invention includes 13,751 positive node pairs in the training set and adds the same number of randomly sampled negative node pairs; the number of positive node pairs and negative node pairs in the validation set and the test set are also consistent; in this experiment, two evaluation indicators are used, namely the area under the curve (AUC) and the average precision (AP), and the specific results are shown as follows: Figure 5 shown.

[0230] from Figure 5It can be found from the experimental results that HRMNN uses a relationship graph generator to complete the adjacency matrix, which shows obvious advantages in the link prediction experiment, and the performance of this model is also better than other baseline models; on this basis, HGNN-IC further pays attention to the missing neighborhood of the tail node, and uses the neighborhood information of the head node to generate a transfer vector for completion, achieving 2.9% and 4.7% improvement in performance respectively; although both the HGNN-IC model and the Tail-GNN model are committed to solving the problem of completing the connection information of the tail node, Tail-GNN only aggregates the information of neighboring nodes when generating neighborhood information, while HGNN-IC adopts a more complex strategy, which not only considers the information of neighboring nodes, but also uses the attention mechanism to judge the importance of node information, and transmits information through edges, so as to generate node representation more accurately; therefore, the HGNN-IC model has achieved 14.8% and 17.1% performance improvement compared with Tail-GNN.

[0231] Experiment 4: Visualization

[0232] In order to more intuitively demonstrate the performance of the HGNN-IC model, we use the t-SNE algorithm to project the generated embedding into a two-dimensional space to visualize the node representation; the experimental results are shown in Figure 6 As shown;

[0233] Overall, the HGNN-IC model shows good performance and can generate node representations with a certain degree of discrimination. On the DBLP dataset, the HGNN-IC model performs best. Figure 6 It can be clearly observed that there are clear boundaries between different types of nodes, and nodes of the same type are closely clustered together to form a clear cluster structure; in the ACM dataset, although some circular nodes and square nodes overlap, overall, the node distribution is relatively concentrated, and the distance between different groups is also relatively obvious; in the IMDB dataset, the performance of the HGNN-IC model is relatively poor. It can be seen from the figure that triangular nodes cannot be effectively distinguished, and the distance between nodes of different categories is closer than that of the ACM dataset; this shows that on the IMDB dataset, the node representation learned by the model still lacks sufficient discrimination.

[0234] Experiment 5: Ablation experiment

[0235] To verify the effectiveness of each component in the HGNN-IC model, we use its variants for ablation studies and the same parameters for node classification experiments. The results are shown in Figure 7 As shown; the information of the variant model is as follows:

[0236] The HGNN-I CA variant focuses on completing attribute information; it uses a feature aggregation module to capture the similarities between nodes and completes the missing attribute information based on this; subsequently, this complete node attribute information is input into the HRMNN model to form the final node representation.

[0237] The HGNN-I CT variant focuses on completing the missing information of the tail node; it is combined with the HRMNN model to perform a neighborhood transfer operation to obtain a node embedding value that contains the neighborhood information of the tail node.

[0238] The F1 score of the HGNN-I C model is the highest, indicating that its performance is better than the other two variants; this advantage mainly comes from the effective combination of the feature aggregation module and the neighborhood transfer module, which jointly improves the overall performance of the model; by comparing the performance of HGNN-I CT and HGNN-I C, we can find that the feature aggregation module in the HGNN-I C model considers the effectiveness of the similarity strategy between nodes, which can reduce the noise propagation caused by missing attributes; on the ACM dataset, the performance of HGNN-I CA is reduced by 3.1% and 3.5% compared with the HGNN-I CT model, respectively. This may be due to the fact that there are more tail nodes in the ACM dataset, and the completion of attribute information alone is not enough to fully capture the characteristics of the nodes. HGNN-I CT completes the neighborhood information of the tail node through the neighborhood transfer module, thereby achieving better performance on the ACM dataset, which further verifies the effectiveness of the neighborhood transfer module in dealing with the tail node problem.

[0239] A specific embodiment of the present invention further provides a heterogeneous graph embedding system based on information completion, including:

[0240] Projection module: Determine and construct the target node set based on the heterogeneous graph; map the initial feature vectors of all nodes in the heterogeneous graph into a unified shared semantic space to obtain the projected feature vectors of all nodes;

[0241] Feature aggregation module: sort the attribute feature similarities between the current target node and its neighbor nodes, and select the first k neighbor nodes of the current target node; perform weighted average on the projected feature vectors of the current target node and its first k neighbor nodes to obtain the preliminary completed feature vector of the current target node;

[0242] Relationship graph generator: Based on the preliminary completed feature vectors of each target node and the projected feature vectors of each neighbor node of each target node, learn the similarity between nodes in the heterogeneous graph and obtain the relationship adjacency graph of each relationship in the heterogeneous graph; aggregate the initial adjacency matrix of each relationship with the relationship adjacency graph to obtain the target adjacency matrix of each relationship, thereby obtaining the target adjacency matrix of the current target node in the current relationship;

[0243] Neighborhood transfer module: Based on all nodes in the heterogeneous graph, the adaptive weight matrix and the number of training times corresponding to each node in the initial round are set, and the adaptive feature vector of the current node in the current round is obtained based on the adaptive feature vector of the current node in the previous round and the adaptive weight matrix corresponding to the current node in the current round; until the preset number of training times is reached, the target feature vector of each node is obtained; wherein, when the training round is 1, the adaptive feature vector of the current target node in the previous round is its preliminary completion feature vector, and the adaptive feature vector of each neighbor node of the current target node in the previous round is its projected feature vector;

[0244] Based on the target adjacency matrix of the current target node in the current relationship, the target feature vectors of each neighbor node of the current target node in the current relationship are normalized to obtain the feature embedding vectors of each neighbor node of the current target node in the current relationship;

[0245] The target feature vector of the current target node and the feature embedding vectors of each neighbor node of the current target node in the current relationship are weighted averaged to obtain the neighborhood embedding vector of the current target node in the current relationship as the actual neighborhood of the current target node in the current relationship; the neighborhood transfer vector of the current target node in the current relationship is set, the ideal neighborhood of the current target node in the current relationship is defined, and the difference between the ideal neighborhood and the actual neighborhood of the current target node in the current relationship is determined; the difference between the ideal neighborhood and the actual neighborhood of the current target node in the current relationship is minimized by scaling and transfer transformation to obtain the neighborhood transfer vector of the current target node in the current relationship, and the neighborhood transfer vector of the current target node in each relationship is obtained in turn; based on the target feature vectors of the current target node and its each neighbor node and the neighborhood transfer vector of the current target node in each relationship, a complete completion feature vector of the current target node is obtained;

[0246] Relationship-level aggregator: Based on the complete completed feature vector of each target node and the target feature vector of each neighboring node of each target node, the relationship strength between nodes in the heterogeneous graph is calculated, and the attention mechanism is used for aggregation to obtain the contextual representation of each target node, and residual connections are performed to obtain the final node representation of each target node.

[0247] Obviously, the above embodiments are merely examples for clear explanation and are not intended to limit the implementation methods. For ordinary technicians in the relevant field, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from them are still within the protection scope of the invention.

Claims

1. A heterogeneous graph embedding method based on information completion, characterized in that: include: Based on the heterogeneous graph, determine and construct the target node set; Map the initial feature vectors of all nodes in the heterogeneous graph into a unified shared semantic space to obtain the projected feature vectors of all nodes; According to the order of the similarity of the attribute features between the current target node and its neighbor nodes, the first k neighbor nodes of the current target node are selected; the projected feature vectors of the current target node and its first k neighbor nodes are weighted averaged to obtain the preliminary completed feature vector of the current target node; Based on the preliminary completed feature vectors of each target node and the projected feature vectors of each neighbor node of each target node, the similarity between nodes in the heterogeneous graph is learned, and the relational adjacency graph of each relationship in the heterogeneous graph is obtained; the initial adjacency matrix of each relationship is aggregated with the relational adjacency graph to obtain the target adjacency matrix of each relationship, thereby obtaining the target adjacency matrix of the current target node in the current relationship; Based on all nodes in the heterogeneous graph, the adaptive weight matrix corresponding to each node in the initial round and the number of training times are set, and the adaptive feature vector of the current node in the current round is obtained based on the adaptive feature vector of the current node in the previous round and the adaptive weight matrix corresponding to the current node in the current round; until the preset number of training times is reached, the target feature vector of each node is obtained; wherein, when the training round is 1, the adaptive feature vector of the current target node in the previous round is its preliminary completion feature vector, and the adaptive feature vector of each neighbor node of the current target node in the previous round is its projected feature vector; Based on the target adjacency matrix of the current target node in the current relationship, the target feature vectors of each neighbor node of the current target node in the current relationship are normalized to obtain the feature embedding vectors of each neighbor node of the current target node in the current relationship; The target feature vector of the current target node and the feature embedding vectors of each neighbor node of the current target node in the current relationship are weighted averaged to obtain the neighborhood embedding vector of the current target node in the current relationship as the actual neighborhood of the current target node in the current relationship; the neighborhood transfer vector of the current target node in the current relationship is set, the ideal neighborhood of the current target node in the current relationship is defined, and the difference between the ideal neighborhood and the actual neighborhood of the current target node in the current relationship is determined; the difference between the ideal neighborhood and the actual neighborhood of the current target node in the current relationship is minimized by scaling and transfer transformation to obtain the neighborhood transfer vector of the current target node in the current relationship, and the neighborhood transfer vector of the current target node in each relationship is obtained in turn; based on the target feature vectors of the current target node and its each neighbor node and the neighborhood transfer vector of the current target node in each relationship, a complete completion feature vector of the current target node is obtained; Based on the complete completed feature vector of each target node and the target feature vector of each neighboring node of each target node, the relationship strength between nodes in the heterogeneous graph is calculated, and the attention mechanism is used for aggregation to obtain the contextual representation of each target node, and residual connection is performed to obtain the final node representation of each target node.

2. According to the method for embedding heterogeneous graphs based on information completion in claim 1, it is characterized in that: Based on all nodes in the heterogeneous graph, the adaptive weight matrix corresponding to each node in the initial round and the number of training times are set, and based on the adaptive feature vector of the current node in the previous round and the adaptive weight matrix corresponding to the current node in the current round, the adaptive feature vector of the current node in the current round is obtained; until the preset number of training times is reached, the target feature vector of each node is obtained; wherein, when the training round is 1, the adaptive feature vector of the current target node in the previous round is its preliminary completion feature vector, and the adaptive feature vector of each neighbor node of the current target node in the previous round is its projected feature vector, including: Based on the adaptive feature vector of the target node v in the (p-1) round and the adaptive weight matrix corresponding to the target node v in the p round The adaptive feature vector of the target node v in round p is obtained, and its expression is: in, Represents the adaptive feature vector of the target node v in round p; represents the adaptive feature vector of the target node v in the (p-1) round. In the initial round, that is, when p=1, Represents the adaptive weight matrix corresponding to the target node v in round p; Based on the adaptive feature vector of any neighbor node u of the target node v in round (p-1) and the adaptive weight matrix corresponding to the neighbor node u of the target node v in round p The adaptive feature vector of the neighbor node u of the target node v in round p is obtained, and its expression is: in, Represents the adaptive feature vector of neighbor node u of target node v in round p; represents the adaptive feature vector of the neighbor node u of the target node v in round (p-1). In the initial round, i.e. when p=1, Represents the adaptive weight matrix of neighbor node u of target node v in round p; Until the preset number of training times is reached, the target feature vector h' of the target node v is obtained v , the target feature vector h' of the neighbor node u of the target node v u .

3. The heterogeneous graph embedding method based on information completion according to claim 1, characterized in that: The target feature vectors of each neighbor node of the current target node in the current relationship are normalized based on the target adjacency matrix of the current target node in the current relationship to obtain the feature embedding vectors of each neighbor node of the current target node in the current relationship, including: Based on the target adjacency matrix A' of the target node v in relation r v,r , the target feature vector h' of the neighbor node u of the target node v in the relationship r u Perform normalization to obtain the feature embedding vector of the neighbor node u of the target node v in the relationship r Its expression is: Among them, h' u Represents the target feature vector of the neighbor node u of the target node v; Represents the normalized coefficient of the target node v in the relation r, and its expression is: D r A' v,r The degree matrix of 4. The heterogeneous graph embedding method based on information completion according to claim 1, characterized in that: The target feature vector of the current target node and the feature embedding vectors of each neighbor node of the current target node in the current relationship are weighted averaged to obtain the neighborhood embedding vector of the current target node in the current relationship as the actual neighborhood of the current target node in the current relationship; Setting the neighborhood transfer vector of the current target node in the current relationship, defining the ideal neighborhood of the current target node in the current relationship, and determining the difference between the ideal neighborhood and the actual neighborhood of the current target node in the current relationship include: The target feature vector h' for the target node v v , the feature embedding vectors of each neighboring node of the target node v in the relation r are weighted averaged to obtain the neighborhood embedding vector of the target node v in the relation r, which is expressed as: in, Represents the neighborhood embedding vector of the target node v in the relation r and serves as the actual neighborhood of the target node v in the relation r; N v,r Represents the set of neighbor nodes of the target node v in the relation r; |N v,r | indicates N v,r The number of neighbor nodes in ; Represents the feature embedding vector of the neighbor node u of the target node v in the relation r; Set the neighborhood transfer vector of the target node v in the relation r and define the ideal neighborhood of the target node v in the relation r Its expression is: Among them, z v,r represents the neighborhood transfer vector of the target node v in the relation r; h' v represents the target feature vector of the target node v; The ideal neighborhood of the target node v in the relation r The actual neighborhood The difference C v,r , whose expression is: Among them, C v,r Represents the difference between the ideal neighborhood and the actual neighborhood of the target node v in the relation r.

5. The heterogeneous graph embedding method based on information completion according to claim 1, characterized in that: The scaling and transfer transformation is used to minimize the difference between the ideal neighborhood and the actual neighborhood of the current target node in the current relationship, and the neighborhood transfer vector of the current target node in the current relationship is obtained. The neighborhood transfer vector of the current target node in each relationship is obtained in turn, including: In the l-th layer network, use the scaling and transfer transformation to obtain the neighborhood transfer vector of the target node v in the l-th layer network in the relationship r Its expression is: in, Represents the scaling vector of the target node v in the l-th layer network in the relation r; Represents the transfer vector of the target node v in the l-th layer network in the relationship r; Set The dimensions and The dimensions are the same; τ represents a unit vector; Represents the sum of neighborhood transfer vectors of all target nodes in the l-th layer network in relation r; Represents the learnable weights of the l-th layer network; According to the neighborhood transfer vector of the target node v in each layer of the network in the relationship r, minimize the ideal neighborhood of the target node v in the relationship r The actual neighborhood The difference C v,r , get the neighborhood transfer vector z of the target node v in the relationship r v,r .

6. The heterogeneous graph embedding method based on information completion according to claim 1, characterized in that: The expression of the complete completion feature vector F[v] of the target node v is: Among them, h' v represents the target feature vector of the target node v; z v,r represents the neighborhood transfer vector of the target node v in the relation r; M v Indicates the set of relationship types corresponding to the target node; h' u represents the target feature vector of the neighbor node u of the target node v in the relation r; N v,r Represents the set of neighbor nodes of the target node v in the relation r; Represents the sum of target feature vectors of each neighbor node in the neighbor node set of the target node v in the relation r; Represents the difference between the neighborhood transfer vector of the target node v in the relation r and the sum of the target feature vectors of each neighbor node in the neighbor node set.

7. The heterogeneous graph embedding method based on information completion according to claim 1, characterized in that: The method of sorting the top k neighbor nodes of the current target node based on the similarity of the attribute features between the current target node and each of its neighbor nodes includes: In the neighborhood of the target node v, determine the neighbor node set N of the target node v v , based on the cosine similarity calculation formula, determine the attribute feature similarity Γ between the target node v and any of its neighbor nodes u u,v , whose expression is: Among them, f u represents the projected feature vector of neighbor node u; f v represents the projected feature vector of the target node v; e represents the edge between the neighbor node u and the target node v; Obtain the attribute feature similarity of the target node v and each neighbor node in its neighbor node set in turn; sort the attribute feature similarity of the target node v and each neighbor node in its neighbor node set from large to small, select the neighbor nodes corresponding to the first k attribute feature similarities of the target node v as the first k neighbor nodes of the target node v, and construct the first k neighbor node set N of the target node v v,k ={u1,u2,...,u k }.

8. The heterogeneous graph embedding method based on information completion according to claim 1, characterized in that: The weighted average of the projected feature vectors of the current target node and its first k neighbor nodes to obtain the preliminary completed feature vector of the current target node includes: For each neighbor node in the set of the first k neighbor nodes of the target node v, set the corresponding weight of each neighbor node; perform weighted average on the projected feature vectors of the first k neighbor nodes of the target node v and their corresponding weights to obtain the aggregated feature embedding value of the current target node Its expression is: Among them, u a Represents the first k neighbor node set N of the target node v v,k The ath neighbor node in ; Represents the first k neighbor node set N of the target node v v,k The weight corresponding to the ath neighbor node in ; Represents the first k neighbor node set N of the target node v v,k The projection feature vector of the ath neighbor node in; k represents the first k neighbor node set N of the target node v v,k The number of neighbor nodes in ; Aggregate feature embedding value for target node v With its projected eigenvector f v Perform weighted averaging to obtain the initial completed feature vector of the target node v Its expression is: in, represents the aggregate feature embedding value of the target node v; f v Represents the projected feature vector of the target node v.

9. The heterogeneous graph embedding method based on information completion according to claim 1, characterized in that: The method of learning the similarity between nodes in the heterogeneous graph based on the preliminary completed feature vectors of each target node and the projected feature vectors of each neighbor node of each target node, and obtaining a relational adjacency graph of each relation in the heterogeneous graph; aggregating the initial adjacency matrix of each relation with the relational adjacency graph to obtain a target adjacency matrix of each relation, thereby obtaining a target adjacency matrix of the current target node in the current relation includes: In S31-S35, the tail node t m and t n is the target node; the initial completion feature vector of each target node All use f * Represents, that is, the initial completion feature vector of the target node v Use f v express; S31: For a relation r in a heterogeneous graph, the first-order feature similarity between node features is obtained through metric learning, and its expression is: in, represents the first-order feature similarity between node ζ and node η under relationship r; Y represents the number of heads; represents the weight corresponding to the number of heads y under relation r; f ζ represents the projected feature vector / preliminary completed feature vector of node ζ; f η represents the projected feature vector / preliminary completed feature vector of node η; if node ζ is the target node, then f ζ represents the initial completed feature vector of node ζ; if node ζ is not the target node, then f ζ represents the projected eigenvector of node ζ; ⊙ represents the Hadamard product; According to the first-order feature similarity Get the similarity graph Its expression is: Among them, ε F ∈[0,1] represents the control similarity graph The sparsity parameter of S32: Based on the relationship r, calculate the feature similarity graph between different types of nodes, that is, the source node s j and the tail node t m Similarity graph between Its expression is: in, Represents the source node s j and the tail node t m The first-order feature similarity between FB Represents the control similarity graph The sparsity parameter of Based on the relationship r, the feature similarity graph between nodes of the same type is calculated, and its expression is: in, Represents the source node s i and j Feature similarity graph between ; Represents the tail node t m and t n Feature similarity graph between ; Represents the source node s i and j The first-order feature similarity between ; Represents the tail node t m and t n The first-order feature similarity between i 、f j Represents the source node s i 、s j The projected eigenvector of m 、f n Respectively represent the tail node t m ,t n The initial completion feature vector of FS , ε FT Represent the control similarity graph The sparsity parameter of S33: Topological structure propagation through heterogeneous graphs, based on source node s i and j The feature similarity graph between the tail node t m and t n The feature similarity graph between them generates a potential propagation graph, which is expressed as: in, Respectively The corresponding propagation diagram; A r Represents the initial adjacency matrix of relation r; S34: Using channel attention to feature similarity map Fusion is performed to generate a relational adjacency graph of relation r Its expression is: in, Represents the weight corresponding to the relationship r; S35: Connect the relation adjacency graph of relation r The initial adjacency matrix A of the relationship r r Aggregate and get the target adjacency matrix of relation r, which is expressed as: Among them, softmax(·) means that w ψ,r The weights are normalized; thus, the target adjacency matrix of the target node v in the relationship r can be obtained as A' v,r .

10. The heterogeneous graph embedding method based on information completion according to claim 1, characterized in that: The method calculates the relationship strength between nodes in the heterogeneous graph based on the complete completion feature vector of each target node and the target feature vector of each neighboring node of each target node, and adopts the attention mechanism to aggregate, obtains the context representation of each target node, and performs residual connection to obtain the final node representation of each target node, including: In S51-S54, the target feature vector h' of each neighboring node of each target node is * They are all represented by F[*], that is, the target feature vector h' of the neighbor node u of the target node v u Denoted by F[u]; S51: For relation triples The target node is the tail node t, and the neighbor node of the tail node t is the source node s. The relationship strength between the source node s and the tail node t is calculated to obtain the attention value ATT-head(s,e,t) of the relationship, which is expressed as: K(s)=K-Linear φ(s) (F[s]) Q(t)=Q-Linear φ(t) (F[t]) Among them, φ(s) represents the type of source node s; φ(t) represents the type of tail node t, and tail node t is the target node; represents the type of connection relationship e between the source node s and the tail node t; K(s) represents the strength of the relationship between the source node s; Q(t) represents the strength of the relationship between the tail node t; F[s] represents the target feature vector of the source node s; F[t] represents the complete complement feature vector of the tail node t; Linear φ(*) Represents a linear projection, which changes the dimension of the source node s or the tail node t from R d become d represents the vector dimension, δ represents the number of attention heads, and d / δ represents the vector dimension of each head; Represents the parameter for adjusting the attention value; Indicates the weight corresponding to ATT-head(s,e,t); S52: Connect the δ attention heads together to obtain the attention vector of each node pair, which is expressed as: Among them, the Softmax function is used to normalize the attention distribution of the source node s adjacent to the tail node t; s∈N(t) indicates that the source node s belongs to the set of neighbor nodes of the tail node t; S53: Combine the attention vector of each node pair with the message vector from the adjacent nodes to obtain the context representation of each target node, which is expressed as: Among them, the tail node t is the target node; represents the context representation of the tail node t in the ξ-th layer relation stack; represents the message vector from the neighboring nodes; Indicates MSG-head γ The weight corresponding to (s, e, t); S54: Map the context representation of each target node to the distribution of the node type corresponding to each target node, and perform residual connection to obtain the final node representation of each target node, which is expressed as: Among them, the tail node t is the target node; H (ξ) [t] represents the final node representation of the tail node t in the ξ-th layer relation stack; H (ξ-1) [t] represents the final node representation of the tail node t in the (ξ-1)th layer of the relationship stack; ξ represents the number of layers of the relationship stack; A-linear φ(t) Represents a linear projection.

Citation Information

Patent Citations

  • Deep heterogeneous graph embedding model based on feature fusion

    CN114565053A

  • Heterogeneous graph representation method and system based on heterogeneous graph conversion and attention mechanism

    CN117725267A

  • Heterogeneous graph embedding method and system based on relation mining

    CN117807275A

Cited By

  • Heterogeneous graph completion method based on node perception attention and related equipment

    CN122133768A