Link Prediction Method and Device for Resisting Edge-Type Interference in Heterogeneous Social Networks

Abstract

The present invention provides a link prediction method and apparatus for resisting edge type interference in a heterogeneous social network. The method includes: constructing an initial heterogeneous social network link prediction model, where the initial heterogeneous social network link prediction model includes an edge feature extraction sub-model and multiple types of expert prediction sub-models; obtaining a data set in the heterogeneous social network and generating a heterogeneous social network graph according to the data set, where the data set includes multiple nodes, multiple edge samples, and multiple edge types; extracting edge features of each edge sample in the heterogeneous social network graph based on the edge feature extraction sub-model to generate an edge feature set; training the multiple types of expert prediction sub-models based on the edge feature set and a preset loss function to obtain a trained and complete target heterogeneous social network link prediction model; predicting a to-be-predicted edge link based on the target heterogeneous social network link prediction model. The present invention can predict multiple link types and new link types.

Description

Technical Field

[0001] The present invention relates to the technical field of link prediction in heterogeneous social networks, and particularly to a link prediction method and device for resisting edge type interference in heterogeneous social networks. Background Art

[0002] With the explosive growth of global social interactions through online services (such as LinkedIn, WeChat, Twitter, and Facebook) in recent years, social network analysis has achieved great success in various fields, from inferring social relationships, predicting volunteering trends, and maximizing the impact of participation from social activities. Link prediction is a long-standing research in social network analysis, focusing on predicting missing edges or future edges based on the observed network structure. In particular, there is an increasing focus on link prediction research in heterogeneous social networks. Heterogeneous social networks are usually represented as graphs containing different types of nodes and edges, where nodes represent individuals and edges represent social interactions. For fixed nodes, the main task of link prediction in heterogeneous social networks is to explain the formation of different types of edges. Different from homogeneous social networks, the formation of different types of edges is usually driven by different evolutionary mechanisms to express different type characteristics. It is difficult to simultaneously explain the formation of different types of edges by fitting different evolutionary mechanisms.

[0003] However, in heterogeneous social networks, it is difficult to collect enough verified edges of one type in real time and train a model for unverified edges of the same type. Different types of edges have type-specific characteristics that cannot be shared to characterize other edge types. Therefore, the challenge of edge type interference has emerged: existing models trained on verified edges from different types tend to learn type-specific knowledge, and their type-specific prediction labels may conflict with unverified edges, resulting in edge type interference, leading to the technical problems of being unable to predict multiple link types and new link types in heterogeneous social networks.

[0004] Therefore, there is an urgent need to propose a link prediction method and device for resisting edge type interference in heterogeneous social networks, which is used to resist edge type interference during link prediction and achieve the technical effect of predicting multiple link types and new link types. Summary of the Invention

[0005] In view of this, it is necessary to provide a link prediction method and device for resisting edge type interference in heterogeneous social networks to solve the technical problem in the prior art that due to the existence of edge type interference, it is impossible to predict multiple link types and new link types in heterogeneous social networks.

[0006] On the one hand, the present invention provides a link prediction method for resisting edge type interference in heterogeneous social networks, including:

[0007] Construct an initial heterogeneous social network link prediction model, where the initial heterogeneous social network link prediction model includes an edge feature extraction sub-model and multiple types of expert prediction sub-models;

[0008] Obtain a data set in the heterogeneous social network, and generate a heterogeneous social network graph according to the data set. The data set includes multiple nodes and multiple edge samples and multiple edge types formed by the multiple nodes;

[0009] Extract the edge features of each edge sample in the heterogeneous social network graph based on the edge feature extraction sub-model to generate an edge feature set;

[0010] Train the multiple types of expert prediction sub-models based on the edge feature set and a preset loss function to obtain a trained and complete target heterogeneous social network link prediction model;

[0011] Predict the edge link to be predicted based on the target heterogeneous social network link prediction model.

[0012] In some possible implementation manners, the multiple edge samples include a first edge sample formed by a first node and a second node; extracting the edge features of each edge sample in the heterogeneous social network graph includes:

[0013] Extract the first node feature of the first node and the second node feature of the second node;

[0014] Determine the edge feature of the first edge sample according to the first node feature and the second node feature.

[0015] In some possible implementation manners, the edge feature of the first edge sample is:

[0016] r e =X u *X v

[0017] In the formula, r e is the edge feature of the first edge sample; X u is the first node feature; X v is the second node feature; * is the Hadamard product operation symbol.

[0018] In some possible implementation manners, the loss function is:

[0019]

[0020]

[0021] C (k)(e) = -[y e log(P (k) (e)) + (1 - y e )log(1 - P (k) (e))]

[0022]

[0023] Q (k,k') (e) = -[P (k) (e)logP (k') (e) + (1 - P (k) (e))log(1 - P (k') (e))]

[0024] In the formula, is the loss function; is the prediction loss function of the k-th type of expert prediction sub-model; K is the total number of multiple types of expert prediction sub-models; λ is the loss function weight value; is the relevant loss function; S k is the edge feature subset with the k-th type of edge type in the edge feature set; C (k) (e) is the cross-entropy loss function of the k-th type of expert prediction sub-model; y e is the set threshold; P (k) (e) is the prediction probability of the k-th type of expert prediction sub-model on whether the edge sample e exists; S is the edge feature set; Q (k,k') (e) is the relevant cross-entropy loss function between the k-th type of expert prediction sub-model and the k'-th type of expert prediction sub-model; P (k) is the probability threshold; P (k') (e) is the prediction probability of the k'-th type of expert prediction sub-model on whether the edge sample e exists.

[0025] In some possible implementation manners, the target heterogeneous social network link prediction model further includes an elastic prediction sub-model; predicting the to-be-predicted edge link based on the target heterogeneous social network link prediction model includes:

[0026] Determining the target reliability weights of various types of expert prediction sub-models in the target heterogeneous social network link prediction model based on the elastic prediction sub-model;

[0027] Extracting the to-be-predicted edge features of the to-be-predicted edge link based on the edge feature extraction sub-model, and determining the similarity between the to-be-predicted edge features and each type of edge features in the edge feature set;

[0028] Predict the to-be-predicted edge feature based on the multiple types of expert prediction sub-models, and correspondingly obtain multiple prediction probabilities;

[0029] Determine the existence probability of the to-be-predicted edge link according to the target reliability weight, the similarity, and the prediction probability.

[0030] In some possible implementation manners, determining the target reliability weights of various types of expert prediction sub-models in the target heterogeneous social network link prediction model based on the elastic prediction sub-model includes:

[0031] Obtain the initial reliability weights of various types of expert prediction sub-models;

[0032] Determine the votes of various types of expert prediction sub-models based on the initial reliability weights;

[0033] Update the initial reliability weights based on the votes to obtain the target reliability weights.

[0034] In some possible implementation manners, the target reliability weight is:

[0035]

[0036]

[0037] In the formula, is the target reliability weight of the k-th type of expert prediction sub-model; A k is the number of prediction times of the k-th type of expert prediction sub-model; vote(a e ) is the vote of the k-th type of expert prediction sub-model; a e is the prediction label of the edge sample e; is the identification set of multiple types of expert prediction sub-models for predicting a e ; A k' is the number of prediction times of the k'-th type of expert prediction sub-model; w k is the initial reliability weight of the k-th type of expert prediction sub-model; w k' is the initial reliability weight of the k'-th type of expert prediction sub-model; μ is a regulation parameter.

[0038] In some possible implementation manners, the similarity is:

[0039]

[0040] In the formula, is the similarity between the to-be-predicted edge feature and the k-th type of edge feature; J d is the feature vector of the to-be-predicted edge feature; J (k)is the feature vector of the k-th type of edge feature; |J d | is the modulus of the feature vector of the edge feature to be predicted; |J (k) | is the modulus of the feature vector of the k-th type of edge feature.

[0041] In some possible implementation manners, the existence probability of the edge link to be predicted is:

[0042]

[0043] In the formula, P (d) is the existence probability of the edge link to be predicted; P (k) (d) is the prediction probability of whether the k-th type of expert prediction sub-model exists for the edge link d to be predicted.

[0044] On the other hand, the present invention also provides a link prediction device for resisting edge type interference in a heterogeneous social network, including:

[0045] An initial model construction unit for constructing an initial heterogeneous social network link prediction model, where the initial heterogeneous social network link prediction model includes an edge feature extraction sub-model and multiple types of expert prediction sub-models;

[0046] A network diagram generation unit for obtaining a data set in a heterogeneous social network and generating a heterogeneous social network diagram according to the data set, where the data set includes multiple nodes and multiple edge samples and multiple edge types formed by the multiple nodes;

[0047] An edge feature extraction unit for extracting edge features of each edge sample in the heterogeneous social network diagram based on the edge feature extraction sub-model to generate an edge feature set;

[0048] A model training unit for training the multiple types of expert prediction sub-models based on the edge feature set and a preset loss function to obtain a trained and complete target heterogeneous social network link prediction model;

[0049] A link prediction unit for predicting an edge link to be predicted based on the target heterogeneous social network link prediction model.

[0050] The beneficial effects of adopting the above embodiments are as follows: The link prediction method for resisting edge type interference in a heterogeneous social network provided by the present invention can extract type characteristics from different types of edge samples by setting multiple types of expert prediction sub-models, so as to enhance the performance of the target heterogeneous social network link prediction model for predicting multiple types of links, resist edge type interference, and then realize the prediction of multiple link types and newly emerging link types in the heterogeneous social network. Description of the Drawings

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for description in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0052] Figure 1 Schematic flowchart of an embodiment of the link prediction method for resisting edge type interference in the heterogeneous social network provided by the present invention;

[0053] Figure 2 Schematic structural diagram of an embodiment of the target heterogeneous social network link prediction model provided by the present invention;

[0054] Figure 3 For the present invention Figure 1 Schematic flowchart of an embodiment of S103 in the present invention;

[0055] Figure 4 For the present invention Figure 1 Schematic flowchart of an embodiment of S105 in the present invention;

[0056] Figure 5 For the present invention Figure 4 Schematic flowchart of an embodiment of S401 in the present invention;

[0057] Figure 6 Schematic structural diagram of an embodiment of the link prediction device for resisting edge type interference in the heterogeneous social network provided by the present invention. Detailed implementation manners

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0059] It should be understood that the schematic drawings are not drawn to actual scale. The flowcharts used in the present invention illustrate the operations implemented according to some embodiments of the present invention. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical context relationships may be reversed or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart or remove one or more operations from the flowchart under the guidance of the content of the present invention.

[0060] Some of the block diagrams shown in the drawings are functional entities and do not necessarily have to correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor systems and / or microcontroller systems.

[0061] Reference to "embodiment" in this document means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the present invention. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0062] An embodiment of the present invention provides a link prediction method and apparatus for resisting edge type interference in a heterogeneous social network, which will be described separately below.

[0063] Figure 1 It is a schematic flowchart of an embodiment of the link prediction method for resisting edge type interference in a heterogeneous social network provided by the present invention. Figure 2 It is a schematic structural diagram of an embodiment of the target heterogeneous social network link prediction model provided by the present invention. As Figure 1 and Figure 2 shown, the link prediction method for resisting edge type interference in a heterogeneous social network includes:

[0064] S101. Construct an initial heterogeneous social network link prediction model, where the initial heterogeneous social network link prediction model includes an edge feature extraction sub-model and multiple types of expert prediction sub-models;

[0065] S102. Obtain a data set in the heterogeneous social network, and generate a heterogeneous social network graph according to the data set. The data set includes multiple nodes, as well as multiple edge samples and multiple edge types formed by the multiple nodes;

[0066] S103. Extract the edge features of each edge sample in the heterogeneous social network graph based on the edge feature extraction sub-model to generate an edge feature set;

[0067] S104. Train multiple types of expert prediction sub-models based on the edge feature set and a preset loss function to obtain a trained and complete target heterogeneous social network link prediction model;

[0068] S105. Predict the edge link to be predicted based on the target heterogeneous social network link prediction model.

[0069] Compared with the prior art, the link prediction method for resisting edge type interference in the heterogeneous social network provided by the embodiment of the present invention can set a variety of types of expert prediction submodels, and can use the variety of types of expert prediction submodels to extract type characteristics from different types of edge samples, so as to enhance the performance of the target heterogeneous social network link prediction model for various types of link predictions, resist edge type interference, and further realize the prediction of various link types and newly emerging link types in the heterogeneous social network.

[0070] In a specific embodiment of the present invention, the edge feature extraction sub-module is node2vec.

[0071] Specifically, the feature representations of all nodes after embedding the heterogeneous social network graph are expressed as a matrix, and the rows of the matrix represent the multi-dimensional features of each node. In the embodiment of the present invention, the edge samples in the heterogeneous social network graph are represented as 128-dimensional vectors.

[0072] In some embodiments of the present invention, among the multiple edge samples, there is a first edge sample formed by a first node and a second node, then as Figure 3 shown, step S103 includes:

[0073] S301. Extract the first node feature of the first node and the second node feature of the second node;

[0074] S302. Determine the edge feature of the first edge sample according to the first node feature and the second node feature. Specifically, the edge feature of the first edge sample is:

[0075] r e =X u *X v

[0076] In the formula, r e is the edge feature of the first edge sample; X u is the first node feature; X v is the second node feature; * is the symbol of Hadamard product operation.

[0077] In some embodiments of the present invention, the loss function is:

[0078]

[0079]

[0080] C (k) (e)=-[y e log(P (k) (e))+(1 - y e )log(1 - P (k) (e))]

[0081]

[0082] Q (k,k') L(e) = -[P (k) (e) log P (k') (e) + (1 - P (k) (e)) log(1 - P (k') (e))]

[0083] Wherein, is the loss function; is the prediction loss function of the k-th type of expert prediction sub-model; K is the total number of multiple types of expert prediction sub-models; λ is the loss function weight value; is the correlation loss function; S k is the edge feature subset with the k-th type of edge type in the edge feature set; C (k) (e) is the cross-entropy loss function of the k-th type of expert prediction sub-model; y e is the set threshold; P (k) (e) is the prediction probability of the k-th type of expert prediction sub-model on whether the edge sample e exists; S is the edge feature set; Q (k,k') (e) is the correlation cross-entropy loss function between the k-th type of expert prediction sub-model and the k'-th type of expert prediction sub-model; P (k) is the probability threshold; P (k') (e) is the prediction probability of the k'-th type of expert prediction sub-model on whether the edge sample e exists.

[0084] It should be understood that: both the set threshold and the probability threshold can be adjusted according to the actual application scenario and experience. For example, both the set threshold and the probability threshold are 0.5.

[0085] It should also be understood that: when the loss function is minimized, the training of the initial heterogeneous social network link prediction model is completed, and the target heterogeneous social network link prediction model is obtained.

[0086] Among them, the way of descending the model parameters during training can be stochastic gradient descent.

[0087] It should be noted that: as Figure 2 shown, each type of expert prediction sub-model includes three convolutional neural network layers, one fully connected layer, and the softmax function.

[0088] And it should be understood that: the number of types of multiple types of expert prediction sub-models is the same as the number of types of multiple edge types, that is, multiple types of expert prediction sub-models are constructed corresponding to the number of edge types in the dataset.

[0089] Considering the differences in the generation of different types of expert prediction sub - models, in order to further improve the prediction accuracy of the target heterogeneous social network link prediction model for new link types, in some embodiments of the present invention, as Figure 2 shown, the target heterogeneous social network link prediction model further includes an elastic prediction sub - model. Then, as Figure 4 shown, step S105 includes:

[0090] S401. Determine the target reliability weights of various types of expert prediction sub - models in the target heterogeneous social network link prediction model based on the elastic prediction sub - model;

[0091] S402. Extract the to - be - predicted edge features of the to - be - predicted edge link based on the edge feature extraction sub - model, and determine the similarity between the to - be - predicted edge features and each type of edge feature in the edge feature set;

[0092] S403. Predict the to - be - predicted edge features based on multiple types of expert prediction sub - models, and correspondingly obtain multiple prediction probabilities;

[0093] S404. Determine the existence probability of the to - be - predicted edge link according to the target reliability weights, similarity, and prediction probabilities.

[0094] In the embodiments of the present invention, by setting an elastic prediction sub - model to determine the target reliability weights of various types of expert prediction sub - models, and determining multiple prediction probabilities and similarities, the existence probability of specific types of edge links can be aggregated in the prediction stage, further improving the ability of the target heterogeneous social network link prediction model to resist edge - type interference and the prediction accuracy for new link types.

[0095] Moreover, the embodiments of the present invention propose target reliability weights and type similarities, and fuse their existence possibilities in the elastic prediction mechanism, improving the stability of the model prediction results.

[0096] In some embodiments of the present invention, step S401 can be specifically to determine the target reliability weights by the Truth Discovery method. Specifically, as Figure 5 shown, step S401 includes:

[0097] S501. Obtain the initial reliability weights of various types of expert prediction sub - models;

[0098] S502. Determine the votes of various types of expert prediction sub - models based on the initial reliability weights;

[0099] S503. Update the initial reliability weights based on the votes to obtain the target reliability weights.

[0100] In the embodiments of the present invention, by updating the initial reliability weights, the target reliability weights are obtained, which can improve the rationality of various types of expert prediction submodels when predicting different types of links.

[0101] It should be understood that the initial reliability weights of various types of expert prediction submodels are equal.

[0102] In some embodiments of the present invention, the target reliability weight is:

[0103]

[0104]

[0105] In the formula, is the target reliability weight of the k-th type of expert prediction submodel; A k is the number of predictions of the k-th type of expert prediction submodel; vote(a e ) is the vote of the k-th type of expert prediction submodel; a e is the predicted label of the marginal sample e; is the identification set of multiple types of expert prediction submodels for predicting a e ; A k' is the number of predictions of the k'-th type of expert prediction submodel; w k is the initial reliability weight of the k-th type of expert prediction submodel; w k' is the initial reliability weight of the k'-th type of expert prediction submodel; μ is the adjustment parameter.

[0106] In some embodiments of the present invention, the similarity is:

[0107]

[0108] In the formula, is the similarity between the edge feature to be predicted and the k-th type of edge feature; J d is the feature vector of the edge feature to be predicted; J (k) is the feature vector of the k-th type of edge feature; |J d | is the norm of the feature vector of the edge feature to be predicted; |J (k) | is the norm of the feature vector of the k-th type of edge feature.

[0109] In some embodiments of the present invention, the existence probability of the edge link to be predicted is:

[0110]

[0111] In the formula, P (d) is the existence probability of the edge link to be predicted; P (k)(d) is the prediction probability of whether the to-be-predicted edge link d exists for the k-th type of expert prediction sub-model.

[0112] To better implement the link prediction method for resisting edge type interference in the heterogeneous social network in the embodiments of the present invention, correspondingly, based on the link prediction method for resisting edge type interference in the heterogeneous social network, the embodiments of the present invention further provide a link prediction device for resisting edge type interference in the heterogeneous social network. As Figure 6 shown, the link prediction device 800 for resisting edge type interference in the heterogeneous social network includes:

[0113] An initial model construction unit 601, configured to construct an initial heterogeneous social network link prediction model, where the initial heterogeneous social network link prediction model includes an edge feature extraction sub-model and multiple types of expert prediction sub-models;

[0114] A network graph generation unit 602, configured to obtain a data set in the heterogeneous social network and generate a heterogeneous social network graph according to the data set. The data set includes multiple nodes and multiple edge samples and multiple edge types formed by the multiple nodes;

[0115] An edge feature extraction unit 603, configured to extract edge features of each edge sample in the heterogeneous social network graph based on the edge feature extraction sub-model, and generate an edge feature set;

[0116] A model training unit 604, configured to train multiple types of expert prediction sub-models based on the edge feature set and a preset loss function, and obtain a trained and complete target heterogeneous social network link prediction model;

[0117] A link prediction unit 605, configured to predict a to-be-predicted edge link based on the target heterogeneous social network link prediction model.

[0118] The link prediction device 800 for resisting edge type interference in the heterogeneous social network provided in the above embodiments can implement the technical solutions described in the embodiments of the link prediction method for resisting edge type interference in the heterogeneous social network. The specific implementation principles of the above modules or units can be referred to the corresponding content in the embodiments of the link prediction method for resisting edge type interference in the heterogeneous social network, and will not be elaborated here.

[0119] Those skilled in the art can understand that all or part of the processes for implementing the methods in the above embodiments can be completed by instructing relevant hardware (such as a processor, a controller, etc.) through a computer program, and the computer program can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium is a magnetic disk, an optical disk, a read-only memory, or a random access memory, etc.

[0120] The above has introduced in detail the link prediction method and device for resisting edge type interference in the heterogeneous social network provided by the present invention. Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A link prediction method for resisting interference of edge types in a heterogeneous social network, characterized in that Comprising: Constructing an initial heterogeneous social network link prediction model, where the initial heterogeneous social network link prediction model includes an edge feature extraction sub-model and multiple types of expert prediction sub-models; Obtaining a data set in the heterogeneous social network, and generating a heterogeneous social network graph according to the data set, where the data set includes multiple nodes and multiple edge samples and multiple edge types formed by the multiple nodes; Extracting edge features of each edge sample in the heterogeneous social network graph based on the edge feature extraction sub-model to generate an edge feature set; Training the multiple types of expert prediction sub-models based on the edge feature set and a preset loss function to obtain a trained and complete target heterogeneous social network link prediction model; Predicting a to-be-predicted edge link based on the target heterogeneous social network link prediction model; The loss function is: In the formula, is the loss function; is the prediction loss function of the k-th type of expert prediction sub-model; K is the total number of multiple types of expert prediction sub-models; is the loss function weight value; is the relevant loss function; is the edge feature subset with the edge type of the k-th type of edge in the edge feature set; is the cross-entropy loss function of the k-th type of expert prediction sub-model; is the set threshold; is the prediction probability of whether the edge sample e exists for the k-th type of expert prediction sub-model; S is the edge feature set; is between the k-th type of expert prediction sub-model and the -th type of expert prediction sub-model's relevant cross-entropy loss function; is the probability threshold; is the -th type of expert prediction sub-model's prediction probability of whether the edge sample e exists; The target heterogeneous social network link prediction model further includes an elastic prediction sub-model; the predicting the to-be-predicted edge link based on the target heterogeneous social network link prediction model includes: Determining target reliability weights of various types of expert prediction sub-models in the target heterogeneous social network link prediction model based on the elastic prediction sub-model; Extracting to-be-predicted edge features of the to-be-predicted edge link based on the edge feature extraction sub-model, and determining the similarity between the to-be-predicted edge features and each type of edge features in the edge feature set; Predicting the to-be-predicted edge features based on the multiple types of expert prediction sub-models to obtain multiple prediction probabilities correspondingly; Determining the existence probability of the to-be-predicted edge link according to the target reliability weights, the similarity, and the prediction probabilities.

2. The link prediction method for resisting edge type interference in a heterogeneous social network according to claim 1, characterized in that, The multiple edge samples include a first edge sample formed by a first node and a second node; the extracting edge features of each edge sample in the heterogeneous social network graph includes: Extracting a first node feature of the first node and a second node feature of the second node; Determining the edge feature of the first edge sample according to the first node feature and the second node feature.

3. The method for link prediction to resist edge type interference in a heterogeneous social network according to claim 2, wherein The edge feature of the first edge sample is: In the formula, is the edge feature of the first edge sample; is the first node feature; is the second node feature; is the symbol of Hadamard product operation.

4. The link prediction method for resisting edge type interference in a heterogeneous social network according to claim 1, wherein The determining target reliability weights of various types of expert prediction sub-models in the target heterogeneous social network link prediction model based on the elastic prediction sub-model includes: Obtaining initial reliability weights of various types of expert prediction sub-models; Determining votes of various types of expert prediction sub-models based on the initial reliability weights; Updating the initial reliability weights based on the votes to obtain the target reliability weights.

5. The link prediction method for resisting edge type interference in a heterogeneous social network according to claim 1, wherein The target reliability weights are: Wherein, is the target reliability weight of the k-th type of expert prediction sub-model; is the number of predictions of the k-th type of expert prediction sub-model; is the vote of the k-th type of expert prediction sub-model; is the predicted label of the marginal sample e; is the prediction identification set of multiple types of expert prediction sub-models; is the number of predictions of the k-th type of expert prediction sub-model; is the initial reliability weight of the k-th type of expert prediction sub-model; is the initial reliability weight of the k-th type of expert prediction sub-model; is the adjustment parameter.

6. The link prediction method for resisting edge type interference in a heterogeneous social network according to claim 5, wherein The similarity is: In the formula, is the similarity between the edge feature to be predicted and the edge feature of the k-th type; is the feature vector of the edge feature to be predicted; is the feature vector of the edge feature of the k-th type; is the norm of the feature vector of the edge feature to be predicted; is the norm of the feature vector of the edge feature of the k-th type.

7. The method for link prediction in a heterogeneous social network to resist edge type interference according to claim 6, wherein The existence probability of the to-be-predicted edge link is: In the formula, is the existence probability of the edge link to be predicted; is the prediction probability of whether the edge link d to be predicted exists by the k-th type of expert prediction sub-model.

8. A link prediction device for resisting interference of edge types in a heterogeneous social network, characterized in that, Applicable to the link prediction method for resisting edge type interference in the heterogeneous social network according to any one of claims 1-7, the device includes: An initial model construction unit for constructing an initial heterogeneous social network link prediction model, where the initial heterogeneous social network link prediction model includes an edge feature extraction sub-model and multiple types of expert prediction sub-models; A network graph generation unit, configured to obtain a data set in a heterogeneous social network and generate a heterogeneous social network graph according to the data set, where the data set includes a plurality of nodes and a plurality of edge samples and a plurality of edge types formed by the plurality of nodes; An edge feature extraction unit, configured to extract edge features of each edge sample in the heterogeneous social network graph based on the edge feature extraction sub-model, and generate an edge feature set; A model training unit, configured to train the multiple types of expert prediction sub-models based on the edge feature set and a preset loss function to obtain a trained and complete target heterogeneous social network link prediction model; A link prediction unit, configured to predict a to-be-predicted edge link based on the target heterogeneous social network link prediction model.

Images