Symbol network data analysis method and device based on period-boosted symbol graph convolutional network

Abstract

The invention discloses a symbol network data analysis method and device based on a periodic boosted symbol graph convolutional network. The method comprises the following steps: constructing an initial boosted symbol graph convolutional network; according to the initial boosted symbol graph convolutional network, constructing a periodic boosted symbol graph convolutional network; reading symbol network data, and constructing an initial node feature matrix according to the symbol network data; and inputting the initial node feature matrix into the period promotion symbol graph convolutional network to obtain a target node representation of the symbol network data. The method can improve the accuracy of symbol network data analysis, and can be widely applied to the technical field of graph machine learning.

Description

Technical Field

[0001] The present invention relates to the field of graph machine learning technology, and in particular to a method and device for analyzing symbolic network data based on a periodic lifting symbolic graph convolutional network. Background Art

[0002] In recent years, graph neural networks (GNNs), especially graph convolutional networks (GCNs), have attracted increasing attention. GCNs have shown their effectiveness in various network analysis tasks and can effectively learn low-dimensional node representations, such as link prediction, recommendation systems, community detection, node classification, graph classification, and visualization.

[0003] Graph convolutional networks (GCNs) have reached a high level of maturity in the application of unsigned networks. However, real networks are not limited to unsigned networks. With the increasing popularity of social media, people have begun to focus on studying the structure and evolution of signed social networks, and are committed to understanding and revealing the factional structure and development laws in social groups. In signed social networks, the links between individuals can be in positive and negative forms, representing friendship, trust or hostility and conflict.

[0004] As a key theoretical foundation in sociological research, structural balance theory is of great significance for understanding the structure and evolution of symbolic networks. Structural balance theory believes that the establishment of a network is based on the principles of "my friend's friend is my friend", "my friend's enemy is my enemy", "my enemy's friend is my enemy" and "my enemy's enemy is my friend". SGCN uses structural balance theory to design an aggregation mechanism and successfully solves the confusion problem between positive and negative link features. However, SGCN encounters information loss in the process of aggregating features, which inevitably leads to serious over-smoothing problems. Summary of the invention

[0005] In view of this, the main purpose of the embodiments of the present invention is to provide a symbolic network data analysis method and device based on a periodic lifting symbolic graph convolutional network, in order to solve at least one of the problems of the prior art and improve the accuracy of symbolic network data analysis.

[0006] To achieve the above purpose, an embodiment of the present invention provides a method for analyzing signed network data based on a periodic lifting signed graph convolutional network, the method comprising:

[0007] Construct an initial boosted signed graph convolutional network;

[0008] According to the initial lifted signed graph convolutional network, construct a periodic lifted signed graph convolutional network;

[0009] Reading symbolic network data, and constructing an initial node feature matrix according to the symbolic network data;

[0010] The initial node feature matrix is ​​input into the periodic lifting signed graph convolutional network to obtain the target node representation of the signed network data.

[0011] In some embodiments, constructing an initial lifting signed graph convolutional network comprises the following steps:

[0012] Setting a first aggregator of the initial lifted signed graph convolutional network;

[0013] Construct the first aggregation function;

[0014] According to the first aggregator and the first aggregation function, the initial lifted signed graph convolutional network is obtained.

[0015] In some embodiments, setting the first aggregator of the initial lifted signed graph convolutional network comprises the following steps:

[0016] Setting two of the first aggregators for the first layer neural network of the initial lifted signed graph convolutional network;

[0017] Obtaining a target number of the first aggregators set in a current layer neural network of the initial lifted signed graph convolutional network;

[0018] The first aggregators are set to twice the target number for the next layer of neural network of the initial lifted signed graph convolutional network.

[0019] In some embodiments, constructing the first aggregation function comprises the following steps:

[0020] Constructing a first weight matrix and a second weight matrix according to the number of layers of the initial lifted signed graph convolutional network;

[0021] Aggregate positive neighbor information to obtain positive neighbor aggregation features;

[0022] Aggregate negative neighbor information to obtain negative neighbor aggregation features;

[0023] The first aggregation function is obtained according to the first weight matrix, the second weight matrix, the positive neighbor aggregation feature and the negative neighbor aggregation feature in combination with a nonlinear activation function.

[0024] In some embodiments, constructing a periodic lifted signed graph convolutional network according to the initial lifted signed graph convolutional network comprises the following steps:

[0025] Construct the second aggregation function;

[0026] One of the initial lifted signed graph convolutional networks is set as a cycle, several of the initial lifted signed graph convolutional networks are connected, and combined with the second aggregation function to obtain the periodic lifted signed graph convolutional network.

[0027] In some embodiments, setting one of the initial lifted signed graph convolutional networks as a cycle and connecting a plurality of the initial lifted signed graph convolutional networks comprises the following steps:

[0028] Setting the first neural network layer of the initial lifted signed graph convolutional network of the current cycle to a cascaded neural network layer;

[0029] Concatenating the cascade layer neural network of the initial lifted signed graph convolutional network of the current cycle with the output features of the initial lifted signed graph convolutional network of the previous cycle;

[0030] The initial lifted signed graph convolutional network of the next cycle is used as the initial lifted signed graph convolutional network of the current cycle, and the step of setting the first layer neural network of the initial lifted signed graph convolutional network of the current cycle as the cascade layer neural network is returned.

[0031] To achieve the above purpose, another aspect of an embodiment of the present invention provides a signed network data analysis device based on a periodic lifting signed graph convolutional network, the device comprising:

[0032] The first module is used to build the initial lifting symbol graph convolutional network;

[0033] A second module is used to construct a periodic lifted signed graph convolutional network based on the initial lifted signed graph convolutional network;

[0034] The third module is used to read the symbolic network data and construct an initial node feature matrix according to the symbolic network data;

[0035] The fourth module is used to input the initial node feature matrix into the periodic lifting signed graph convolutional network to obtain the target node representation of the signed network data.

[0036] To achieve the above-mentioned purpose, another aspect of an embodiment of the present invention provides an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the aforementioned symbolic network data analysis method based on the periodic lifting symbolic graph convolutional network.

[0037] To achieve the above-mentioned purpose, another aspect of an embodiment of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the aforementioned symbolic network data analysis method based on the periodic lifting symbolic graph convolutional network.

[0038] To achieve the above-mentioned purpose, another aspect of an embodiment of the present invention provides a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device can read the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the aforementioned symbolic network data analysis method based on a periodic lifting symbolic graph convolutional network.

[0039] The embodiments of the present invention include at least the following beneficial effects: The present invention provides a symbolic network data analysis method and device based on a periodic lifting symbolic graph convolutional network. The scheme constructs an initial lifting symbolic graph convolutional network; constructs a periodic lifting symbolic graph convolutional network based on the initial lifting symbolic graph convolutional network; reads symbolic network data, and constructs an initial node feature matrix based on the symbolic network data; inputs the initial node feature matrix into the periodic lifting symbolic graph convolutional network to obtain a target node representation of the symbolic network data, which can improve the accuracy of symbolic network data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0041] Figure 1 It is a flow chart of a method for analyzing signed network data based on a periodic lifting signed graph convolutional network provided by an embodiment of the present invention;

[0042] Figure 2 is a schematic diagram of a symbolic graph neural network algorithm framework provided by an embodiment of the present invention;

[0043] Figure 3 is a schematic diagram of a symbolic network provided by an embodiment of the present invention;

[0044] Figure 4 is a schematic diagram of an ISGCN provided by an embodiment of the present invention;

[0045] Figure 5 is a schematic diagram of aggregator settings in ISGCN provided by an embodiment of the present invention;

[0046] Figure 6 is a schematic diagram of aggregator settings in an m-periodic ISGCN provided by an embodiment of the present invention;

[0047] Figure 7 It is a schematic diagram of the hardware structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present invention, and they are only examples of devices and methods consistent with some aspects of the embodiments of the present invention as detailed in the attached claims.

[0049] It should be noted that, although the functional modules are divided in the system schematic diagram and the logical order is shown in the flow chart, in some cases, the steps shown or described may be performed in a different order than the module division in the system or the flow chart. The terms "first / S100" and "second / S200" in the specification and claims and the above-mentioned drawings may be used to describe various concepts in this article, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiment of the present invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determination".

[0050] The terms "at least one", "multiple", "each", "any", etc. used in the present invention, at least one includes one, two or more, multiple includes two or more, each refers to each of the corresponding multiple, and any refers to any one of the multiple.

[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein are only for the purpose of describing the embodiments of the present invention and are not intended to limit the present invention.

[0052] In symbolic networks, people usually tend to agree with the opinions of those they trust and hold opinions that are contrary to those they do not trust. Therefore, it is important to distinguish between "friends" and "enemies" during aggregation. That is, when aggregating layer l, we need to aggregate B separately. iThe nodes in (l) and U i Nodes in (l). The Signed Graph Convolution Network (SGCN) uses the structural balance theory to set two aggregators at each layer, extending the GraphSAGE framework to symbolic networks and aggregating “friends” and “enemies” separately.

[0053] SGCN can be viewed as having two aggregators, B(l) and U(l), at each layer, each performing different tasks. An aggregator B(l) collects nodes u i of (Node u i The node features of the l-hop neighbor set of node u are collected by another aggregator U(l). i of (Node u i The node characteristics of the l-order enemy set). Figure 3 , taking the second-order path as an example, considering that although Tom is the enemy of Alice's enemy and John is the friend of Alice's friend, according to the structural balance theory, both Tom and John are classified as Alice's friends. However, there is still an essential difference between them. The latter has stronger emotions than the former, and the former is more inclined to friendship generated by interests. When designing the aggregation function, SGCN is not able to distinguish between these two situations.

[0054] In view of this, if Figure 2 As shown, in some embodiments, the influence of high-order paths in SGCN on symbol formation, the number of aggregators, and the influence of division of labor on the expression of neural networks are analyzed in depth. By expanding SGCN, increasing the number of aggregators, and refining the aggregation tasks of aggregators, an improved signed graph convolution network (ISGCN) is constructed to enhance the expression ability of the model. Then, according to the structural balance theory, an m-periodic improved signed graph convolutional neural network, namely the m-periodic ISGCN (m-periodic Improved Signed Graph Convolutional Network) architecture is designed, and it is capable of executing deep GCNs. Among them, 1-periodic ISGCN is equivalent to SGCN. Thereby solving the problem of exponential increase in computational complexity, through a more flexible m-periodic ISGCN architecture, the accuracy of analysis can be improved in symbolic network data analysis.

[0055] refer to Figure 1 , Figure 1It is an optional step flow chart of a method for analyzing signed network data based on a periodic lifting signed graph convolutional network provided by an embodiment of the present invention. The method may include but is not limited to steps S100 to S400:

[0056] Step S100, constructing an initial lifting signed graph convolutional network;

[0057] Step S200, constructing a periodic lifted signed graph convolutional network according to the initial lifted signed graph convolutional network;

[0058] Step S300, reading symbolic network data, and constructing an initial node feature matrix according to the symbolic network data;

[0059] Step S400, inputting the initial node feature matrix into the periodic lifting signed graph convolutional network to obtain the target node representation of the signed network data.

[0060] In step S100 of some embodiments, by improving SGCN and designing a new aggregation mechanism, an initial boosted signed graph convolutional network (ISGCN) can be obtained. Optionally, the SGCN framework can be built using pytorch. Exemplarily, the SGCN framework is built using pytorch, and SGCN is expanded. By increasing the number of aggregators, refining the aggregation tasks of the aggregators, etc., an initial boosted signed graph convolutional network is constructed, which enhances the expressive power of the model while avoiding an explosive surge in complexity.

[0061] In some embodiments, step S100 may include but is not limited to steps S110 to S130:

[0062] Step S110, setting a first aggregator of the initial lifted signed graph convolutional network;

[0063] Step S120, constructing a first aggregation function;

[0064] Step S130: obtaining the initial lifted signed graph convolutional network according to the first aggregator and the first aggregation function.

[0065] In some embodiments, step S110 may include but is not limited to steps S111 to S113:

[0066] Step S111, setting two of the first aggregators for the first layer neural network of the initial lifted signed graph convolutional network;

[0067] Step S112, obtaining a target number of the first aggregators set in the current layer neural network of the initial lifted signed graph convolutional network;

[0068] Step S113, setting twice the target number of the first aggregators for the next layer of the neural network of the initial lifted signed graph convolutional network.

[0069] In order to explain the constructed ISGCN more clearly, in some embodiments, a set of sequences is defined Among them, each sequence Satisfy a ij ∈{+,-},1≤i≤2 l , 1≤j≤l. Let

[0070]

[0071] in, represents the l-th order balance path, And the sequence The number of signs “-” in is even; let

[0072]

[0073] in, represents an unbalanced path of order l, And the sequence The number of "-" signs in is an odd number. For example, and

[0074] Define the terminal node set The terminal node set is from node u i The set of symbol sequences of the starting and passing paths is of type ω, referred to as node u i The set of generalized neighbors of .

[0075] When l = 1,

[0076]

[0077] When l>1,

[0078]

[0079] in, Define ω(l-1) as the aggregator of the (l-1)th layer, which can aggregate nodes The features of the nodes are aggregated.

[0080] In steps S111 to S113 of some embodiments, 2 is set in the first layer of ISGCN. l Aggregators. For example, Figure 4 As shown, Figure 4The figure shows the process of ISGCN aggregating neighbor information in the symbolic network through the first aggregator. In the first layer of the neural network, two aggregators are set: +(1) and -(1), which aggregate the nodes respectively. and In the second layer of the neural network, four aggregators are set: ++(2), +-(2), -+(2) and --(2), which aggregate the nodes respectively. and Among them, represents the set of high-order neighbors of the corresponding node, and N represents the trivial 1st-order neighbors of the corresponding node. Compared with SGCN, by refining the task allocation of the aggregator, they can distinguish between node u i of "friends of friends" and "enemies of my enemies", and distinguishing between nodes u i The “enemies of friends” and “friends of enemies” of the model enhance the expressive power of the model.

[0081] In some embodiments, step S120 may include but is not limited to steps S121 to S124:

[0082] Step S121, constructing a first weight matrix and a second weight matrix according to the number of layers of the initial lifted signed graph convolutional network;

[0083] Step S122, aggregating positive neighbor information to obtain positive neighbor aggregation features;

[0084] Step S123, aggregating negative neighbor information to obtain negative neighbor aggregation features;

[0085] Step S124, obtaining the first aggregation function according to the first weight matrix, the second weight matrix, the positive neighbor aggregation feature and the negative neighbor aggregation feature in combination with a nonlinear activation function.

[0086] In steps S121 to S124 of some embodiments, for each layer of the initial lifted signed graph convolutional network, two sets of weight matrices can be defined, one for positive links and the other for negative links. For each layer and each node in the network, the feature average of its positive neighbors can be first calculated to obtain the average feature of the positive neighbors. This average feature of the positive neighbors is combined with the feature of the node itself to obtain the positive neighbor aggregate feature. According to the weight matrix used for the positive link and the positive neighbor aggregate feature, combined with the nonlinear activation function, the positive neighbor information can be aggregated. Similarly, the feature average of the negative neighbors is calculated to obtain the average feature of the negative neighbors. This average feature of the negative neighbors is combined with the feature of the node itself to obtain the negative neighbor aggregate feature. According to the weight matrix used for the negative link and the negative neighbor aggregate feature, combined with the nonlinear activation function, the negative neighbor information can be aggregated. By aggregating the positive neighbor information and aggregating the negative neighbor information, a first aggregation function can be obtained.

[0087] like Figure 5 As shown, in some embodiments, Represents the initial input node feature matrix, where each aggregator (green rectangle) takes the corresponding node feature matrix as input and outputs the corresponding node feature matrix of a specific path type. Represents user u i The initial node features, corresponding to X (0) Next, the recursive expression of ISGCN, i.e., the first aggregation function, is given.

[0088] When l = 1,

[0089]

[0090] Where σ(·) is a nonlinear activation function; is the weight matrix of the linear transformation; d out is the dimension of the node hidden representation; d in is the input node feature dimension; is the output of aggregator+(1); is the output of aggregator-(1); Is user u i Initial node features of Is user u j Initial node features of Is user u k The initial node features.

[0091] For l>1,

[0092]

[0093] in, is the weight matrix of the linear transformation; Is user u i The output result of the aggregator ω+(l); Is user u i The output of the aggregator ω-(l); Is user u i The output of the aggregator ω(l-1) at the l-1th layer; Is user u j The output of the aggregator ω(l-1) at the l-1th layer; Is user u k The output of the aggregator ω(l-1) at the l-1th layer.

[0094] In step S130 of some embodiments, by setting the first aggregator and constructing the first aggregation function, a complete framework of the initial lifting signed graph convolutional network can be obtained. For each node, feature aggregation is performed using the aggregation function through the initial lifting signed graph convolutional network. In the first layer, SGCN and ISGCN have the same form, but in the second layer, compared with SGCN, ISGCN has two additional learnable parameter matrices. In the subsequent layers, more additional learnable parameter matrices can be obtained, and more complex node representations can be learned. These additional parameters allow the network to have higher flexibility and expressiveness when processing positive and negative links, thereby enhancing the representation ability of the model.

[0095] In step S200 of some embodiments, the structural balance theory is used to reduce the number of aggregators. By setting an initial lifted signed graph convolutional network to a period and combining it with the corresponding aggregation function, a periodic lifted signed graph convolutional network can be constructed, which can effectively reduce the computational complexity, reduce the computational cost, and ensure the rationality of the aggregation process.

[0096] In some embodiments, step S200 may include but is not limited to steps S210 to S220:

[0097] Step S210, constructing a second aggregation function;

[0098] Step S220: setting one of the initial lifted signed graph convolutional networks as a period, connecting several of the initial lifted signed graph convolutional networks, and combining the second aggregation function to obtain the periodic lifted signed graph convolutional network.

[0099] In some embodiments, based on the balance theory, a balanced path is defined, where a balanced path is composed of an even number of negative links, and similarly, an unbalanced path is composed of an odd number of negative links.

[0100] In step S210 of some embodiments, a recursive expression of a periodic lifting signed graph convolutional network, i.e., a second aggregation function, is constructed. Taking an m-periodic lifting signed graph convolutional network as an example, L is defined to represent the number of layers of the m-periodic lifting signed graph convolutional network, where L≥m.

[0101] For a positive integer t satisfying t≥1, when l=1,

[0102]

[0103] in, is the weight matrix of the linear transformation.

[0104] When l = tm + 1,

[0105]

[0106] in, is the weight matrix of the linear transformation; BF is the aggregation result of a balanced aggregator; UF is the aggregation result of an unbalanced aggregator; They are user u j Corresponding The output of the aggregator; They are user u j Corresponding The output of the aggregator; They are user u i correspond The output of the aggregator; Is user u i Corresponding The output of the aggregator.

[0107] When (t-1)m+2≤l≤tm,

[0108]

[0109] in, And r = (l-1) mod m; W ω+(l) , is the weight matrix of the linear transformation.

[0110] In some embodiments, the step of setting one of the initial lifted signed graph convolutional networks as a cycle and connecting a plurality of the initial lifted signed graph convolutional networks in step S220 may include but is not limited to steps S221 to S223:

[0111] Step S221, setting the first neural network layer of the initial lifting signed graph convolutional network of the current cycle to a cascade layer neural network;

[0112] Step S222, concatenating the cascade layer neural network of the initial lifted signed graph convolutional network of the current cycle with the output features of the initial lifted signed graph convolutional network of the previous cycle;

[0113] Step S223, using the initial lifted signed graph convolutional network of the next cycle as the initial lifted signed graph convolutional network of the current cycle, and returning to the step of setting the first layer neural network of the initial lifted signed graph convolutional network of the current cycle as a cascade layer neural network.

[0114] In steps S221 to S223 of some embodiments, Figure 6As shown, an m-layer neural network is set as one cycle, that is, an initial lifting signed graph convolutional network is set as one cycle. The (tm+1)th layer is designated as a cascade layer, that is, for the tm+1th cycle, the neural network of the (tm+1)th layer is set as the neural network of the cascade layer, where t≥1. Through the structural balance theory, the output features corresponding to the positive (tm+1)-order path in the last layer of the previous cycle are obtained, and the output features are connected to the cascade layer (that is, the (tm+1)th layer) of the current cycle. Then, in the network of the current cycle, these output features are compressed as input features of the new "+" path aggregator in the next cycle. Similarly, it connects the feature output of the unbalanced path to the output of the new "-" path aggregator in the next cycle. The adjustable parameter m represents the number of layers in one cycle.

[0115] Through the periodic setting and the constructed second aggregation function, the complete framework of the periodic lifting signed graph convolutional network can be obtained. It can be obtained that SGCN is a special case of m-periodic ISGCN with m=1, while L-periodic ISGCN is equivalent to ISGCN. In some embodiments, the time complexity of ISGCN mainly depends on the aggregation process of node features. For each node, the aggregation function is used to aggregate features, and the time complexity of forward propagation in ISGCN can be approximately expressed as O(2 l nF), where l represents the number of layers, n represents the number of nodes, and F represents the dimension of node features. Similarly, the time complexity of the periodic lifting signed graph convolutional network is mainly determined by the node feature aggregation process. For example, taking the m-periodic lifting signed graph convolutional network as an example, for each node, an aggregation function is used to perform feature aggregation. Therefore, the time complexity of the forward propagation of the m-periodic lifting signed graph convolutional network can be approximately expressed as O(2 m |V|d out (Lr) / m), where r = Lmodm. By using the structural balance theory to reduce the number of aggregators and setting the m-layer neural network as a cycle, combined with the corresponding aggregation function, an m-cycle lifting signed graph convolutional network is constructed, which can effectively reduce the computational complexity and cost, and ensure the rationality of the aggregation process.

[0116] In some embodiments, the loss function L(θ W ,θ MLG ) is defined as follows:

[0117]

[0118] Where M represents a set of triples. In this set, given a triple (u i ,u j ,s)∈M, where s represents ui to u j Is there a positive (negative) link? If s = +1 (s = -1), it means u i and u j There is a positive (negative) link between them. On the other hand, if s = *, it means there is no link between them. The weight associated with each symbol s is denoted by τ s In order to balance the contribution to the overall goal, the parameter λ is introduced. The parameter matrix of the SGCN model is represented by θ W , and θ MLG represents the parameters of the MLG classifier. In addition, the triple set M(+,*) is defined as a subset of triples, in which each triple (u i ,u j ,u k )∈M(+,*) means from u i to u j There is a positive link, and from u j to u k There is no link. Similarly, we can define a triple set M(-,*) to represent triplets with negative links.

[0119] In steps S300 to S400 of some embodiments, symbolic network data is read, and an initial node feature matrix of the symbolic network data is constructed by a symbolic graph decomposition algorithm. The initial node feature matrix of the symbolic network data is used as the input of the initial lifted symbolic graph convolutional network and the m-periodic lifted symbolic graph convolutional network. After several layers of aggregation, the final node representation can be obtained for downstream tasks. Exemplarily, the final generated node representation can be directly used in subsequent tasks such as classification, clustering, link prediction, etc., especially in social network analysis or sentiment analysis. These node representations can accurately reflect the relationships and sentiment tendencies of nodes in the symbolic network, which helps to improve the accuracy of sentiment classification or the accuracy of community detection. For an undirected symbolic network The algorithm of the embodiment of the present invention aims to learn the low-dimensional node representation matrix of the symbolic network λ where d out is the dimension of the node embedding, the i-th row vector z of Z i Represents node u i This low-dimensional representation can be used for link sign prediction in symbolic networks.

[0120] Taking social network data analysis as an example, the symbolic network data analysis method based on periodic lifting symbolic graph convolutional network is applied to two social network datasets obtained from the SNAP website: Slashdot and Epinions. Slashdot is a symbolic network in which nodes represent users and links represent the relationship between two users. Users can establish positive (friends) and negative (hostile) relationships with other users. Epinions is a symbolic network that represents the trust relationship between users on Epinions.com. On this website, users can express trust or distrust to other users.

[0121] First, the data set is preprocessed, and the weights of the links in the signed network that are greater than 0 are modified to 1, indicating positive sign information, and the weights of the links that are less than 0 are modified to 11, indicating negative sign information. 80% of the positive links and negative links are selected as the training set, and the remaining 20% ​​of the positive links and negative links are selected as the test set, and the test set links are masked.

[0122] In the above-mentioned embodiment of social network data analysis, the method of the present invention is compared with Signed Spectral Embedding (SSE), Signed Network Embedding (SiNE), SIgned Directed Network Embedding (SIDE) and SGCN by using two evaluation indicators, AUC and F1. The detailed description of the data set is shown in the data set statistics in Table 1:

[0123] Table 1

[0124] Dataset node Positive Link Negative Link % Positive Link Slashdot 82140 425072 124130 77.40 Epinions 131828 717667 123705 85.30

[0125] In the above example of social network data analysis, L layers of ISGCN were implemented, where L∈{2,3,6}, and L layers of 2-periodic ISGCN and 3-periodic ISGCN were implemented, where L=6. For all GNNs methods, the final embedding of the SSE model is used as the input feature matrix X (0) , and concatenate the representations output by the last layer to form the final features of the node. For each dataset, 80% of the data is randomly divided into a training set, and the remaining 20% ​​is used as a test set. The SGCN model includes 2 layers and 6 layers, and the size of both input and output channels is set to 64. The learning rate is set to 0.01 and the number of iterations is set to 200. Labeled links are used as training data, and the performance is evaluated using two indicators, F1 and AUC. Generally, higher F1 and AUC values ​​indicate better performance. For the parameters of the loss function, λ is set to 5. Obviously, using richer node features as the initial input matrix X of m-periodic ISGCN(0) The expressive power of m-periodic ISGCN (e.g., a more advanced symbolic graph embedding algorithm) can still be enhanced. However, the present invention focuses on extending the SGCN architecture. Therefore, in order to make a fair comparison, the embodiment of the present invention adopts the same benchmark setting as SGCN. And thus, friend recommendations between users based on link weights and link symbol information between nodes are obtained. If the predicted link weight is greater than 0, the users tend to be in a friend relationship and can be recommended. If the predicted link weight is less than 0, the users tend to be in an enemy relationship and recommendation is not recommended.

[0126] The embodiment of the present invention further provides a symbolic network data analysis device based on a periodic lifting symbolic graph convolutional network, which can implement the above-mentioned symbolic network data analysis method based on a periodic lifting symbolic graph convolutional network, and the device includes:

[0127] The first module is used to build the initial lifting symbol graph convolutional network;

[0128] A second module is used to construct a periodic lifted signed graph convolutional network based on the initial lifted signed graph convolutional network;

[0129] The third module is used to read the symbolic network data and construct an initial node feature matrix according to the symbolic network data;

[0130] The fourth module is used to input the initial node feature matrix into the periodic lifting signed graph convolutional network to obtain the target node representation of the signed network data.

[0131] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0132] The embodiment of the present invention further provides an electronic device, which includes a processor and a memory, wherein the memory stores a computer program, and when the processor executes the computer program, the above-mentioned symbolic network data analysis method based on the periodic lifting symbolic graph convolutional network is implemented. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0133] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0134] refer to Figure 7 , Figure 7 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0135] The processor 501 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention;

[0136] The memory 502 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 502 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 502, and the processor 501 calls and executes the symbolic network data analysis method based on the periodic lifting symbolic graph convolutional network of the embodiment of the present invention;

[0137] Input / output interface 503, used to implement information input and output;

[0138] Communication interface 504, used to realize communication interaction between the device and other devices, which can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0139] A bus 505 that transmits information between the various components of the device (e.g., the processor 501, the memory 502, the input / output interface 503, and the communication interface 504);

[0140] The processor 501 , the memory 502 , the input / output interface 503 and the communication interface 504 are connected to each other in communication within the device via the bus 505 .

[0141] An embodiment of the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned symbolic network data analysis method based on the periodic lifting symbolic graph convolutional network.

[0142] It can be understood that the contents of the above method embodiments are all applicable to the present storage medium embodiments, the functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0143] The embodiment of the present invention also provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the aforementioned symbolic network data analysis method based on the periodic lifting symbolic graph convolutional network.

[0144] In summary, the method and device for analyzing signed network data based on periodic lifting signed graph convolutional network according to the embodiment of the present invention have the following advantages:

[0145] The embodiment of the present invention introduces a new aggregation mechanism for the symbolic graph convolutional neural network based on the structural balance theory in sociology, focusing on enhancing the expressive power of the symbolic graph neural network (SGCN). To this end, the aggregation mechanism of SGCN is deeply analyzed. Secondly, by optimizing the aggregator task, the expressive power of SGCN is enhanced, thereby introducing the boosted symbolic graph convolutional network (ISGCN) to solve the over-smoothing problem of the current SGCN. Subsequently, a generalized form of ISGCN, the m-periodic boosted symbolic graph convolutional network (m-periodic ISGCN), is proposed to solve the problem of exponential increase in computational complexity and provide a more flexible architecture. As a graph neural network algorithm, the embodiment of the present invention can be widely used in social network analysis, sentiment analysis, recommendation systems, etc.

[0146] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part for which is described as a larger operation is performed independently.

[0147] In addition, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features described may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present invention. More specifically, in view of the properties, functions, and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional skills of the engineer. Therefore, those skilled in the art can implement the present invention set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0148] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0149] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0150] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0151] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0152] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0153] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

[0154] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the described embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A method for analyzing signed network data based on periodic lifting signed graph convolutional network, characterized in that: The following steps are involved: Construct an initial boosted signed graph convolutional network; According to the initial lifted signed graph convolutional network, construct a periodic lifted signed graph convolutional network; Reading symbolic network data, and constructing an initial node feature matrix according to the symbolic network data; The initial node feature matrix is ​​input into the periodic lifting signed graph convolutional network to obtain the target node representation of the signed network data.

2. According to claim 1, a method for analyzing signed network data based on periodic lifting signed graph convolutional network is characterized in that: The construction of the initial lifting symbol graph convolutional network includes the following steps: Setting a first aggregator of the initial lifted signed graph convolutional network; Construct the first aggregation function; According to the first aggregator and the first aggregation function, the initial lifted signed graph convolutional network is obtained.

3. A method for analyzing signed network data based on a periodic lifting signed graph convolutional network according to claim 2, characterized in that: The step of setting the first aggregator of the initial lifted signed graph convolutional network comprises the following steps: Setting two of the first aggregators for the first layer neural network of the initial lifted signed graph convolutional network; Obtaining a target number of the first aggregators set in a current layer neural network of the initial lifted signed graph convolutional network; The first aggregators are set to twice the target number for the next layer of neural network of the initial lifted signed graph convolutional network.

4. The method for analyzing signed network data based on periodic lifting signed graph convolutional network according to claim 2, characterized in that: The constructing of the first aggregation function comprises the following steps: Constructing a first weight matrix and a second weight matrix according to the number of layers of the initial lifted signed graph convolutional network; Aggregate positive neighbor information to obtain positive neighbor aggregation features; Aggregate negative neighbor information to obtain negative neighbor aggregation features; The first aggregation function is obtained according to the first weight matrix, the second weight matrix, the positive neighbor aggregation feature and the negative neighbor aggregation feature in combination with a nonlinear activation function.

5. The method for analyzing signed network data based on periodic lifting signed graph convolutional network according to claim 1, characterized in that: The step of constructing a periodic lifted signed graph convolutional network according to the initial lifted signed graph convolutional network comprises the following steps: Construct the second aggregation function; One of the initial lifted signed graph convolutional networks is set as a cycle, several of the initial lifted signed graph convolutional networks are connected, and combined with the second aggregation function to obtain the periodic lifted signed graph convolutional network.

6. A method for analyzing signed network data based on periodic lifting signed graph convolutional network according to claim 5, characterized in that: The step of setting one of the initial lifted signed graph convolutional networks as a cycle and connecting a plurality of the initial lifted signed graph convolutional networks comprises the following steps: Setting the first neural network layer of the initial lifted signed graph convolutional network of the current cycle to a cascaded neural network layer; Concatenating the cascade layer neural network of the initial lifted signed graph convolutional network of the current cycle with the output features of the initial lifted signed graph convolutional network of the previous cycle; The initial lifted signed graph convolutional network of the next cycle is used as the initial lifted signed graph convolutional network of the current cycle, and the step of setting the first layer neural network of the initial lifted signed graph convolutional network of the current cycle as the cascade layer neural network is returned.

7. A symbolic network data analysis device based on a periodic lifting symbolic graph convolutional network, characterized in that: include: The first module is used to build the initial lifting symbol graph convolutional network; A second module is used to construct a periodic lifted signed graph convolutional network based on the initial lifted signed graph convolutional network; The third module is used to read the symbolic network data and construct an initial node feature matrix according to the symbolic network data; The fourth module is used to input the initial node feature matrix into the periodic lifting signed graph convolutional network to obtain the target node representation of the signed network data.

8. An electronic device, characterized in that: including a processor and a memory; The memory is used to store programs; The processor executes the program to implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The storage medium stores a program, and the program is executed by a processor to implement the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images