Network graph feature extraction method, device, electronic device, and storage medium

Abstract

The present disclosure relates to a network graph feature extraction method, device, electronic device, storage medium and computer program product, the method comprising: obtaining a relationship network graph corresponding to original interactive data; taking any object node in the relationship network graph as a target object node, and determining a set of adjacent nodes corresponding to the target object node; determining a vector representation corresponding to the target object node according to the connection relationship between each adjacent object node in the adjacent node set; dividing the relationship network graph into at least two sub-relationship networks according to the vector representation corresponding to each target object node, and extracting features from each sub-relationship network to obtain feature extraction results corresponding to each sub-relationship network, and using each feature extraction result as a multi-meaning feature corresponding to the original interactive data. The present method can obtain different sub-relationship networks for ambiguous description, thereby improving the extraction efficiency of relationship class features.

Description

Technical Field

[0001] The present disclosure relates to the field of Internet technology, and in particular to a network graph feature extraction method, device, electronic device, storage medium and computer program product. Background Art

[0002] With the development of information technology, feature engineering plays an important role in machine learning. Feature engineering mainly converts raw data into features suitable for modeling so that the algorithm program can understand and apply it to the calculation process, thereby improving the performance of the trained model by obtaining better training data features.

[0003] However, the features extracted in traditional machine learning are mainly static features. For relational features involving multiple entities, since relational features themselves are ambiguous, their feature values ​​will change according to different contextual scenarios. Taking user preferences for videos as an example, users show more interest in videos posted by unfamiliar authors, while they show more social preferences for videos posted by authors with close relationships such as family members, relatives and friends. Similar to the above-mentioned context-related and ambiguous relational features, it is currently difficult to extract them using traditional methods. Summary of the invention

[0004] The present disclosure provides a network graph feature extraction method, device, electronic device, storage medium and computer program product to at least solve the problem of difficulty in extracting relationship class features in related technologies. The technical solution of the present disclosure is as follows:

[0005] According to a first aspect of an embodiment of the present disclosure, a network graph feature extraction method is provided, comprising:

[0006] Obtaining a relationship network diagram corresponding to the original interactive data; the relationship network diagram is used to represent the interactive relationship between multiple data objects in the original interactive data, and the relationship network diagram includes object nodes representing each of the data objects and connection edges representing each of the interactive relationships;

[0007] Taking any object node in the relationship network diagram as a target object node, determining an adjacent node set corresponding to the target object node; the adjacent node set includes adjacent object nodes connected to the target object node, and the number of object nodes connected in series between the adjacent object node and the target object node is less than a preset threshold;

[0008] Determine the vector representation corresponding to the target object node according to the connection relationship between each adjacent object node in the adjacent node set;

[0009] According to the vector representation corresponding to each target object node, the relationship network diagram is divided into at least two sub-relationship networks, and feature extraction is performed on each of the sub-relationship networks to obtain feature extraction results corresponding to each of the sub-relationship networks, and each of the feature extraction results is used as a multi-meaning feature corresponding to the original interaction data.

[0010] In a possible implementation, taking any object node in the relationship network graph as a target object node and determining a set of adjacent nodes corresponding to the target object node includes:

[0011] Taking any object node in the relationship network diagram as a target object node, determining a directly connected node and an indirectly connected node corresponding to the target object node; the number of object nodes connected in series between the indirectly connected node and the target object node is less than a preset threshold;

[0012] The directly connected node and the indirectly connected node are regarded as the adjacent nodes.

[0013] In a possible implementation, determining the vector representation corresponding to the target object node according to the connection relationship between the adjacent object nodes in the adjacent node set includes:

[0014] Constructing a node sequence corresponding to the target object node according to the connection relationship between each adjacent object node in the adjacent node set;

[0015] Vector training is performed using the node sequence corresponding to each of the target object nodes to obtain the vector representation corresponding to each of the target object nodes.

[0016] In a possible implementation, constructing a node sequence corresponding to the target object node according to the connection relationship between adjacent object nodes in the adjacent node set includes:

[0017] Taking a certain object node in the adjacent node set as a sampling node;

[0018] Determine a next object node of the sampling node; the next object node is at least one object node among directly connected nodes of the sampling node;

[0019] Taking the next object node as the sampling node, and returning to the step of determining the next object node of the sampling node until the number of the sampling nodes meets a preset number threshold;

[0020] According to the node information corresponding to each of the sampling nodes, a node sequence corresponding to the target object node is constructed.

[0021] In a possible implementation, dividing the relationship network graph into at least two sub-relationship networks according to the vector representation corresponding to each of the target object nodes includes:

[0022] Determine the distance between any two of the object nodes in the relationship network diagram according to the vector representation corresponding to each of the target object nodes;

[0023] Dividing the relationship network diagram according to the distance between any two of the object nodes in the relationship network diagram to obtain the at least two sub-relationship networks;

[0024] Among them, the at least two sub-relationship networks include a first sub-relationship network and a second sub-relationship network, the distance between any two object nodes in the first sub-relationship network or the second sub-relationship network is less than a first distance threshold, and the distance between any object node in the first sub-relationship network and any object node in the second sub-relationship network is greater than a second distance threshold.

[0025] In a possible implementation, obtaining a relationship network diagram corresponding to the original interaction data includes:

[0026] Acquire an original relationship network diagram of the original interactive data; the original relationship network diagram is used to represent the interactive relationship between multiple original data objects of different data types, and the original relationship network diagram includes nodes representing each of the original data objects;

[0027] Determine the nodes to be deleted in the original relationship network diagram; the nodes to be deleted include original nodes of original data objects corresponding to data types other than the target data type;

[0028] The node to be deleted is deleted from the original relationship network diagram to obtain the relationship network diagram.

[0029] In a possible implementation, extracting features from each of the sub-relationship networks to obtain feature extraction results corresponding to each of the sub-relationship networks includes:

[0030] Acquire node attribute information corresponding to the node to be deleted; the node attribute information is used to characterize the interactive relationship between the object node in each of the sub-relationship networks and the node to be deleted;

[0031] Feature extraction is performed according to the node attribute information and each of the sub-relationship networks to obtain feature extraction results corresponding to each of the sub-relationship networks.

[0032] According to a second aspect of an embodiment of the present disclosure, a network graph feature extraction device is provided, comprising:

[0033] A relationship network graph acquisition unit is configured to execute acquisition of a relationship network graph corresponding to the original interactive data; the relationship network graph is used to represent the interactive relationship between multiple data objects in the original interactive data, and the relationship network graph includes object nodes representing each of the data objects and connection edges representing each of the interactive relationships;

[0034] The adjacent node set determining unit is configured to execute taking any object node in the relationship network diagram as a target object node and determining an adjacent node set corresponding to the target object node; the adjacent node set includes adjacent object nodes connected to the target object node, and the number of object nodes connected in series between the adjacent object node and the target object node is less than a preset threshold;

[0035] A node vector determination unit is configured to determine a vector representation corresponding to the target object node according to a connection relationship between adjacent object nodes in the adjacent node set;

[0036] The sub-relationship network division unit is configured to execute the division of the relationship network diagram into at least two sub-relationship networks according to the vector representation corresponding to each target object node, and perform feature extraction on each sub-relationship network to obtain feature extraction results corresponding to each sub-relationship network, and use each feature extraction result as a multi-meaning feature corresponding to the original interaction data.

[0037] In one possible implementation, the adjacent node set determination unit is specifically configured to execute taking any object node in the relationship network diagram as a target object node, determining the directly connected nodes and indirectly connected nodes corresponding to the target object node; the number of object nodes connected in series between the indirectly connected node and the target object node is less than a preset threshold; and taking the directly connected node and the indirectly connected node as the adjacent object nodes.

[0038] In one possible implementation, the node vector determination unit is specifically configured to construct a node sequence corresponding to the target object node according to the connection relationship between each adjacent object node in the adjacent node set; and perform vector training using the node sequence corresponding to each target object node to obtain a vector representation corresponding to each target object node.

[0039] In a possible implementation, the node vector determination unit is further configured to execute the steps of taking an object node in the adjacent node set as a sampling node; determining the next object node of the sampling node; the next object node being at least one object node among the directly connected nodes of the sampling node; taking the next object node as the sampling node, and returning to the step of determining the next object node of the sampling node until the number of sampling nodes meets a preset number threshold; and constructing a node sequence corresponding to the target object node according to the node information corresponding to each of the sampling nodes.

[0040] In a possible implementation, the sub-relationship network division unit is specifically configured to determine the distance between any two object nodes in the relationship network diagram according to the vector representation corresponding to each target object node; divide the relationship network diagram according to the distance between any two object nodes in the relationship network diagram to obtain the at least two sub-relationship networks; wherein the at least two sub-relationship networks include a first sub-relationship network and a second sub-relationship network, the distance between any two object nodes in the first sub-relationship network or the second sub-relationship network is less than a first distance threshold, and the distance between any object node in the first sub-relationship network and any object node in the second sub-relationship network is greater than a second distance threshold.

[0041] In one possible implementation, the relationship network diagram acquisition unit is specifically configured to execute acquisition of an original relationship network diagram of the original interactive data; the original relationship network diagram is used to characterize the interactive relationship between original data objects of multiple different data types, and the original relationship network diagram includes nodes representing each of the original data objects; determine the nodes to be deleted in the original relationship network diagram; the nodes to be deleted include the original nodes of the original data objects corresponding to data types other than the target data type; delete the nodes to be deleted from the original relationship network diagram to obtain the relationship network diagram.

[0042] In one possible implementation, the sub-relationship network division unit is specifically configured to execute acquisition of node attribute information corresponding to the node to be deleted; the node attribute information is used to characterize the interactive relationship between the object node in each of the sub-relationship networks and the node to be deleted; feature extraction is performed based on the node attribute information and each of the sub-relationship networks to obtain feature extraction results corresponding to each of the sub-relationship networks.

[0043] According to a third aspect of an embodiment of the present disclosure, there is provided an electronic device, including:

[0044] processor;

[0045] a memory for storing instructions executable by the processor;

[0046] Wherein, the processor is configured to execute the instructions to implement the network graph feature extraction method as described in any one of the above items.

[0047] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. When instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the network graph feature extraction method as described in any one of the above items.

[0048] According to a fifth aspect of an embodiment of the present disclosure, a computer program product is provided, wherein the computer program product includes instructions, and when the instructions are executed by a processor of an electronic device, the electronic device is enabled to execute the network graph feature extraction method as described in any one of the above items.

[0049] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects:

[0050] The scheme disclosed in the present invention obtains a relational network diagram corresponding to the original interactive data, the relational network diagram is used to characterize the interactive relationship between multiple data objects in the original interactive data, the relational network diagram includes object nodes representing each data object, and connecting edges representing each interactive relationship, then any object node in the relational network diagram is used as a target object node, and the adjacent node set corresponding to the target object node is determined, the adjacent node set includes adjacent object nodes connected to the target object node, and the number of object nodes connected in series between the adjacent object node and the target object node is less than a preset threshold, and the vector representation corresponding to the target object node is determined according to the connection relationship between each adjacent object node in the adjacent node set, and then according to the vector representation corresponding to each target object node, the relational network diagram is divided into at least two sub-relational networks, and features are extracted for each sub-relational network to obtain feature extraction results corresponding to each sub-relational network, and each feature extraction result is used as a multi-meaning feature corresponding to the original interactive data. In this way, the relational network diagram corresponding to the original interactive data can be effectively segmented to obtain meaningful sub-relational networks, and different sub-relational networks can be used to describe ambiguity, thereby improving the extraction efficiency of relational features.

[0051] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.

[0053] Figure 1 The present invention is a flowchart of a method for extracting network graph features according to an exemplary embodiment.

[0054] Figure 2 The diagram is a schematic diagram showing a network diagram division method according to an exemplary embodiment.

[0055] Figure 3 The present invention is a flowchart showing the steps of obtaining a relationship network diagram according to an exemplary embodiment.

[0056] Figure 4 It is a schematic diagram of an ambiguous feature extraction process according to an exemplary embodiment.

[0057] Figure 5 The figure is a flowchart of another network graph feature extraction method according to an exemplary embodiment.

[0058] Figure 6 It is a block diagram of a network graph feature extraction device according to an exemplary embodiment.

[0059] Figure 7 The figure is a diagram showing the internal structure of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0060] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0061] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure.

[0062] It should also be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0063] Figure 1 is a flow chart of a network graph feature extraction method according to an exemplary embodiment. Figure 1 As shown, the method can be used in a server and includes the following steps.

[0064] In step S110, a relationship network diagram corresponding to the original interaction data is obtained;

[0065] Among them, the relationship network diagram can be used to represent the interactive relationship between multiple data objects in the original interactive data, and the multiple data objects belong to the same data type, for example, can be the interactive relationship between multiple user accounts.

[0066] As an example, through feature engineering, features can be extracted from the original interaction data obtained in the actual scenario to provide data support for subsequent model training. For example, the original interaction data can be obtained based on the interactive attention between users, the interactive situation of users watching videos, etc.

[0067] In practical applications, based on the original interaction data, a relationship network diagram corresponding to the original interaction data can be obtained. The relationship network diagram may include object nodes representing each data object and connection edges representing each interaction relationship. The relationship network diagram can then be used to further perform network diagram feature extraction, such as extracting relationship class features.

[0068] Specifically, the relationship network diagram can be a homogeneous network, in which all network nodes (i.e., object nodes) have the same function and belong to the same data type, and the network nodes can be switched with each other. For example, in a user follow-up relationship network, it can include network nodes representing each user account, and the positions of different network nodes can be replaced with each other.

[0069] In step S120, any object node in the relationship network diagram is used as a target object node, and a set of adjacent nodes corresponding to the target object node is determined;

[0070] The adjacent node set may include adjacent object nodes connected to the target object node, and the number of object nodes serially connected between the adjacent object node and the target object node is less than a preset threshold.

[0071] After obtaining the relationship network diagram, any object node in the relationship network diagram can be used as the target object node, and then, according to the adjacent connection status of the target object node in the relationship network diagram, the object nodes whose number of object nodes connected in series with the target object node is greater than or equal to 0 and less than a preset threshold can be used as adjacent object nodes, and the adjacent node set corresponding to the target object node can be obtained.

[0072] In one example, the preset threshold may be a preset degree of neighbor relationship between object nodes. For example, object nodes of 1 to N degree neighbor relationships corresponding to the target object node may be selected to obtain an adjacent node set.

[0073] For example, in a relationship network diagram, if two object nodes are connected by an edge, then the two object nodes are adjacently connected, and the two object nodes have a first-degree adjacent relationship; if object node A has an adjacently connected object node B, and object node B has an adjacently connected object node C, then there is no direct adjacent relationship between object node A and object node C, and object node A and object node C have a second-degree adjacent relationship. Furthermore, when there are N object nodes in the connection path between two object nodes, the two object nodes have an N+1-degree adjacent relationship.

[0074] In step S130, according to the connection relationship between each adjacent object node in the adjacent node set, the vector representation corresponding to the target object node is determined;

[0075] In a specific implementation, the vector representation of the learning node can be determined by describing the co-occurrence relationship between the nodes in the adjacent node set according to the connection relationship between the adjacent object nodes in the adjacent node set, and then the vector representation corresponding to the target object node can be determined.

[0076] In an optional embodiment, based on the set of adjacent nodes corresponding to the target object node, node sampling can be performed using a random walk method. By using a depth-first traversal algorithm that can repeatedly visit visited nodes, a certain number of node access sequences can be obtained. The node access sequence can then be used for vector learning to obtain a vector representation corresponding to the target object node.

[0077] In step S140, the relationship network diagram is divided into at least two sub-relationship networks according to the vector representation corresponding to each target object node, and feature extraction is performed on each sub-relationship network to obtain feature extraction results corresponding to each sub-relationship network, and each feature extraction result is used as a multi-meaning feature corresponding to the original interaction data.

[0078] In practical applications, by taking any object node in the relationship network diagram as the target object node to determine the vector representation corresponding to each target object node, the vector representation corresponding to each object node in the relationship network diagram can be obtained, and then the relationship network diagram can be divided according to the vector representation corresponding to each object node to obtain at least two sub-relationship networks, and then feature extraction can be performed based on each sub-relationship network, and the obtained feature extraction results can be used as multi-meaning features corresponding to the original interaction data.

[0079] In one example, community discovery is performed based on each object node in a relationship network graph. The relationship network graph can be split into multiple sub-relationship networks, that is, multiple closely connected communities can be found in the relationship network graph, and the same node can be determined to which community or multiple communities based on the connection relationship. For example, by segmenting the original social graph to obtain meaningful sub-networks, an ambiguous description can be performed based on different sub-networks, thereby achieving ambiguous representation of nodes in the graph relationship network. By performing data calculation and processing on the sub-networks, computing resource consumption can be reduced and network performance can be improved.

[0080] In another example, taking the user social network as an example, the user social network can be divided into different social circle networks. Since each user node has user attribute information, feature extraction can be performed on each social circle network to meet the needs of ambiguity, such as a family relationship social circle including the user's family, a classmate relationship social circle including the user's classmates, an interest relationship social circle including accounts followed by the user, etc. Different social circle networks corresponding to each user account can be obtained for different user accounts.

[0081] In the above network graph feature extraction method, the relationship network graph corresponding to the original interactive data is obtained, and then any object node in the relationship network graph is used as the target object node to determine the adjacent node set corresponding to the target object node. According to the connection relationship between each adjacent object node in the adjacent node set, the vector representation corresponding to the target object node is determined, and then according to the vector representation corresponding to each target object node, the relationship network graph is divided into at least two sub-relationship networks, and features are extracted for each sub-relationship network to obtain feature extraction results corresponding to each sub-relationship network, and each feature extraction result is used as a multi-meaning feature corresponding to the original interactive data. In this way, the relationship network graph corresponding to the original interactive data can be effectively segmented to obtain meaningful sub-relationship networks, and different sub-relationship networks can be used to describe ambiguity, thereby improving the efficiency of extracting relationship features.

[0082] In an exemplary embodiment, any object node in a relationship network diagram is used as a target object node to determine a set of adjacent nodes corresponding to the target object node, including: taking any object node in the relationship network diagram as a target object node to determine the directly connected nodes and indirectly connected nodes corresponding to the target object node; taking the directly connected nodes and the indirectly connected nodes as adjacent object nodes.

[0083] The number of object nodes connected in series between the indirect connection node and the target object node is less than a preset threshold.

[0084] In a specific implementation, any object node in a relationship network diagram can be used as a target object node, and the directly connected nodes and indirectly connected nodes corresponding to the target object node can be determined based on the adjacent connections of the target object node in the relationship network diagram, and then the directly connected nodes and indirectly connected nodes can be used as adjacent object nodes.

[0085] For example, the directly connected nodes corresponding to the target object node and the indirectly connected nodes whose number of object nodes connected in series with the target object node is less than a preset threshold can be obtained, and then the adjacent node set corresponding to the target object node can be obtained based on the directly connected nodes and the indirectly connected nodes.

[0086] The technical solution of this embodiment, by taking any object node in the relationship network diagram as the target object node, determining the directly connected nodes and indirectly connected nodes corresponding to the target object node, and then taking the directly connected nodes and indirectly connected nodes as adjacent object nodes, can preset the adjacent relationship range to select the corresponding directly connected nodes and indirectly connected nodes, thereby providing data support for the subsequent determination of the vector representation.

[0087] In an exemplary embodiment, a vector representation corresponding to a target object node is determined based on a connection relationship between adjacent object nodes in an adjacent node set, including: constructing a node sequence corresponding to the target object node based on a connection relationship between adjacent object nodes in an adjacent node set; and performing vector training using the node sequence corresponding to each target object node to obtain a vector representation corresponding to each target object node.

[0088] In practical applications, the node sequence corresponding to the target object node can be constructed according to the connection relationship between each adjacent object node in the adjacent node set, and then the node sequence corresponding to each target object node can be used for vector training. The vector representation of the node is learned by describing the co-occurrence relationship between the nodes in the adjacent node set, and the vector representation corresponding to each target object node is obtained.

[0089] In one example, random walks can be used to sample nodes, and a certain number of node access sequences can be obtained based on the set of adjacent nodes corresponding to the target object node through a depth-first traversal algorithm that can repeatedly access visited nodes. The node access sequence can then be used for vector learning.

[0090] For example, for the node sampling method of random walk, for a certain object node, a node can be randomly sampled from its directly connected adjacent nodes as the next object node, and the process of randomly sampling the next node can be repeated according to the preset condition of sequence length until the obtained sequence length meets the preset condition, that is, until the number of sampled nodes meets the preset number threshold. For example, the sequence length can be preset to 3 or 4, that is, the number of sampled nodes is 3 or 4, or it can be other values, which are not specifically limited in this embodiment.

[0091] The technical solution of this embodiment constructs a node sequence corresponding to the target object node according to the connection relationship between each adjacent object node in the adjacent node set, and then uses the node sequence corresponding to each target object node to perform vector training to obtain the vector representation corresponding to each target object node. The vector representation of the node can be learned based on the co-occurrence relationship between the nodes in the adjacent node set, and the vector representation corresponding to each node can be accurately obtained.

[0092] In an exemplary embodiment, a node sequence corresponding to a target object node is constructed based on a connection relationship between adjacent object nodes in an adjacent node set, including: taking an object node in the adjacent node set as a sampling node; determining the next object node of the sampling node; taking the next object node as the sampling node, and returning to the step of determining the next object node of the sampling node until the number of sampling nodes meets a preset number threshold; and constructing a node sequence corresponding to the target object node based on the node information corresponding to each sampling node.

[0093] The next object node may be at least one object node among the directly connected nodes of the sampling node, such as randomly selecting a node from the adjacent nodes directly connected to the sampling node as the next object node.

[0094] In one example, a node sequence can be constructed according to an adjacent node set for vector training. Starting from an object node (i.e., a sampling node) in the adjacent node set, a random walk node sampling method is adopted. For the object node, a node is randomly sampled from its directly connected adjacent nodes as the next object node. The process of randomly sampling the next node can be repeated according to a preset condition of sequence length until the obtained sequence length meets the preset condition, that is, until the number of sampling nodes meets the preset number threshold. For example, the sequence length can be preset to 3 or 4, that is, the number of sampling nodes is 3 or 4, or it can be other values, which are not specifically limited in this embodiment.

[0095] In another example, during the node sampling process using a random walk method, the node information of the object node, such as node identification information, node order information, etc., can be recorded each time an object node is reached. Then, a node sequence corresponding to the target object node can be constructed based on the node information corresponding to each sampling node, and vector learning can be performed after obtaining a sufficient number of node sequences.

[0096] The technical solution of this embodiment is to use an object node in the adjacent node set as a sampling node, then determine the next object node of the sampling node, use the next object node as the sampling node, and return to the step of determining the next object node of the sampling node until the number of sampling nodes meets a preset number threshold, and then construct a node sequence corresponding to the target object node according to the node information corresponding to each sampling node. Node sampling can be performed based on a random walk method to construct a node sequence to provide data support for subsequent vector learning.

[0097] In an exemplary embodiment, the relationship network diagram is divided into at least two sub-relationship networks according to the vector representation corresponding to each target object node, including: determining the distance between any two object nodes in the relationship network diagram according to the vector representation corresponding to each target object node; dividing the relationship network diagram according to the distance between any two object nodes in the relationship network diagram to obtain at least two sub-relationship networks.

[0098] As an example, at least two sub-relationship networks may include a first sub-relationship network and a second sub-relationship network, the distance between any two object nodes in the first sub-relationship network or the second sub-relationship network is less than a first distance threshold, the distance between any object node in the first sub-relationship network and any object node in the second sub-relationship network is greater than a second distance threshold, and the first distance threshold may be less than or equal to the second distance threshold.

[0099] In a specific implementation, the distance between any two object nodes in the relationship network diagram can be determined according to the vector representation corresponding to each target object node. For example, since each object node in the relationship network diagram has a corresponding vector representation, the distance between any two object nodes can be calculated in the following manner:

[0100]

[0101] Among them, i is the node number corresponding to any two object nodes, and n is the total number of object nodes.

[0102] In one example, the relationship network graph can be divided according to the distance between any two object nodes in the relationship network graph to obtain at least two sub-relationship networks. For example, by using a spectral clustering algorithm, a weighted undirected graph (i.e., a relationship network graph) can be divided into two or more optimal sub-graphs (i.e., sub-relationship networks), such as Figure 2 In the two division methods shown, this embodiment can adopt Figure 2 In the middle division method 1, the internal parts of the divided subgraphs can be made as similar as possible, while the distances between the subgraphs are as far as possible (i.e., the distance between any two object nodes in the first sub-relationship network or the second sub-relationship network is less than the first distance threshold, and the distance between any object node in the first sub-relationship network and any object node in the second sub-relationship network is greater than the second distance threshold), thereby achieving the purpose of clustering.

[0103] The technical solution of this embodiment determines the distance between any two object nodes in the relationship network diagram according to the vector representation corresponding to each target object node, and then divides the relationship network diagram according to the distance between any two object nodes in the relationship network diagram to obtain at least two sub-relationship networks. The relationship network diagram can be effectively segmented to obtain meaningful sub-relationship networks, which is conducive to the subsequent extraction of multiple meaningful features based on the sub-relationship networks.

[0104] In an exemplary embodiment, if Figure 3 As shown, obtaining the relationship network diagram corresponding to the original interaction data may include the following steps:

[0105] In step S310, an original relationship network diagram of original interaction data is obtained;

[0106] Among them, the original relationship network diagram can be used to characterize the interactive relationship between original data objects of multiple different data types. The original relationship network diagram may include nodes representing each original data object, and each original data object may belong to multiple different data types. For example, taking the user video watching relationship network as an example, it can be the interactive relationship between multiple user accounts and between different user accounts and watching videos.

[0107] In practical applications, the original relationship network diagram of the original interaction data can be obtained based on the original interaction data. The original relationship network diagram may include nodes representing each original data object. For example, the user video viewing relationship network may include nodes representing multiple user accounts, nodes for the videos watched by each user account, and connecting edges representing each interaction relationship.

[0108] In one example, the original relationship network diagram can be a heterogeneous network, that is, all network nodes (that is, original data objects) in the original relationship network diagram can be divided into two or more data types according to function and utility. For example, in the user video watching relationship network, it can include network nodes representing each user account and network nodes representing the videos watched by each user account.

[0109] In step S320, the nodes to be deleted in the original relationship network graph are determined;

[0110] After obtaining the original relationship network diagram, the target data type can be obtained. The target data type can be used to indicate that the original relationship network diagram is preprocessed to obtain the relationship network diagram. Then, according to the target data type, the original nodes of the original data objects corresponding to data types other than the target data type can be taken as nodes to be deleted to preprocess the original relationship network diagram.

[0111] In step S330, the node to be deleted is deleted from the original relationship network diagram to obtain a relationship network diagram.

[0112] In the specific implementation, the original relationship network diagram can be split into two parts, one is the relationship network body for the core nodes, and the other is the attribute information associated with the core nodes. For example, taking the user watching video relationship network as an example, the social relationship between user nodes can be used as the relationship network body of the core nodes, and then the original user watching video relationship network can be split into a relationship network containing only user nodes and an attribute network containing only video nodes.

[0113] In one example, the video node (i.e., the node to be deleted) can be deleted by retaining the user nodes in the original heterogeneous network and the edges connecting the user nodes, and removing the edges connecting the user nodes and the video nodes. The video node information can be stored, such as the video-related information corresponding to the video node, including the number of video playbacks, the video duration, etc.

[0114] For example, in order to generalize common scenarios, heterogeneous networks can be made compatible, that is, by preprocessing the original relationship network graph, the heterogeneous original relationship network graph can be converted into a homogeneous relationship network graph.

[0115] The technical solution of this embodiment obtains the original relationship network diagram of the original interactive data, then determines the nodes to be deleted in the original relationship network diagram, and then deletes the nodes to be deleted from the original relationship network diagram to obtain a relationship network diagram, which is compatible with heterogeneous networks and improves generalization capabilities.

[0116] In an exemplary embodiment, feature extraction is performed on each sub-relationship network to obtain feature extraction results corresponding to each sub-relationship network, including: obtaining node attribute information corresponding to the node to be deleted; the node attribute information is used to characterize the interactive relationship between the object node in each sub-relationship network and the node to be deleted; feature extraction is performed based on the node attribute information and each sub-relationship network to obtain feature extraction results corresponding to each sub-relationship network.

[0117] In practical applications, feature extraction can be performed by restoring the node attribute information corresponding to the node to be deleted on the sub-relationship network. Since the node attribute information can be used to characterize the interactive relationship between the object node in each sub-relationship network and the node to be deleted, feature extraction can be performed based on the node attribute information and each sub-relationship network to obtain the feature extraction results corresponding to each sub-relationship network.

[0118] For example, taking the user video watching relationship network as an example, for the different social circle networks (i.e., sub-relationship networks) obtained by division, such as the family relationship social circle including the user's family, the classmate relationship social circle including the user's classmates, the interest relationship social circle including the accounts followed by the user, etc., each sub-relationship network and the node attribute information corresponding to the video nodes deleted by each sub-relationship network can be combined to complete feature extraction. For example, based on the user A's family relationship social circle, classmate relationship social circle, and the interactive information such as the number of plays, duration, and likes of videos watched by user A in the family relationship social circle and classmate relationship social circle, the social preference characteristics of user A for videos posted by authors with close relationships such as family members, relatives and friends can be extracted; or based on the user A's interest relationship social circle and the interactive information such as the number of plays, duration, and likes of videos watched by user A in the interest relationship social circle, the interest preference characteristics of user A for videos posted by unfamiliar authors can be extracted, and then the user's preference for videos can be obtained. For example, the user shows more interest preference for videos posted by unfamiliar authors, but shows more social preference for videos posted by authors with close relationships such as family members, relatives and friends.

[0119] In one example, the features of each sub-relationship network can be summarized (ie, the feature extraction results corresponding to each sub-relationship network), and then the corresponding relationship class feature extraction results can be used to carry out targeted model training according to different business strategies.

[0120] The technical solution of this embodiment obtains the node attribute information corresponding to the node to be deleted, and then performs feature extraction based on the node attribute information and each sub-relationship network to obtain the feature extraction results corresponding to each sub-relationship network. It can perform ambiguous descriptions through different sub-relationship networks, thereby improving the efficiency of extracting relationship features.

[0121] In order to enable those skilled in the art to better understand the above steps, Figure 4 The embodiments of the present disclosure are illustrated by an example, but it should be understood that the embodiments of the present disclosure are not limited thereto.

[0122] 1. The original graph relationship network (i.e., the original relationship network diagram) can be preprocessed. By identifying the relationship nodes (i.e., the object nodes representing each data object in the relationship network diagram), the relationship network (i.e., the relationship network diagram corresponding to the original interactive data) can be obtained, and the node attributes (i.e., the nodes to be deleted and their node attribute information) can be stored;

[0123] 2. By performing community discovery on the relationship network, the relationship network can be split into N sub-relationship networks, where N is a preset value that can be adjusted according to specific business scenarios;

[0124] 3. For each sub-relationship network, feature extraction can be performed in combination with the node attributes corresponding to each sub-relationship network to obtain the polysemous features corresponding to each sub-relationship network (ie, feature extraction results).

[0125] Figure 5 is a flowchart of another network graph feature extraction method according to an exemplary embodiment. Figure 5 As shown, the method is used in a computer device such as a server, and includes the following steps.

[0126] In step S510, the original relationship network diagram of the original interactive data is obtained; the original relationship network diagram is used to characterize the interactive relationship between multiple original data objects of different data types, and the original relationship network diagram includes nodes representing each original data object. In step S520, the nodes to be deleted in the original relationship network diagram are determined; the nodes to be deleted include the original nodes of the original data objects corresponding to the data types other than the target data type; the nodes to be deleted are deleted from the original relationship network diagram to obtain the relationship network diagram. In step S530, the relationship network diagram corresponding to the original interactive data is obtained; the relationship network diagram is used to characterize the interactive relationship between multiple data objects in the original interactive data, and the relationship network diagram includes object nodes representing each data object, and connecting edges representing each interactive relationship. In step S540, any object node in the relationship network diagram is used as the target object node, and the adjacent node set corresponding to the target object node is determined; the adjacent node set includes adjacent object nodes connected to the target object node. In step S550, according to the connection relationship between each adjacent object node in the adjacent node set, a node sequence corresponding to the target object node is constructed. In step S560, vector training is performed using the node sequence corresponding to each target object node to obtain a vector representation corresponding to each target object node. In step S570, the relationship network diagram is divided into at least two sub-relationship networks based on the vector representation corresponding to each target object node, and feature extraction is performed on each sub-relationship network to obtain a feature extraction result corresponding to each sub-relationship network, and each feature extraction result is used as a multi-meaning feature corresponding to the original interaction data. It should be noted that the specific limitations of the above steps can be found in the specific limitations of a network diagram feature extraction method above, which will not be repeated here.

[0127] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0128] It can be understood that the same / similar parts between the various embodiments of the above method in this specification can refer to each other, and each embodiment focuses on the differences from other embodiments. For related points, please refer to the description of other method embodiments.

[0129] Based on the same inventive concept, the embodiment of the present disclosure also provides a network graph feature extraction device for implementing the above-mentioned network graph feature extraction method.

[0130] Figure 6 is a block diagram of a network graph feature extraction device according to an exemplary embodiment. Figure 6 , the device comprises:

[0131] The relationship network graph acquisition unit 601 is configured to execute acquisition of a relationship network graph corresponding to the original interaction data; the relationship network graph is used to represent the interaction relationship between multiple data objects in the original interaction data, and the relationship network graph includes object nodes representing each of the data objects and connection edges representing each of the interaction relationships;

[0132] The adjacent node set determining unit 602 is configured to execute taking any object node in the relationship network diagram as a target object node, and determining an adjacent node set corresponding to the target object node; the adjacent node set includes adjacent object nodes connected to the target object node, and the number of object nodes connected in series between the adjacent object node and the target object node is less than a preset threshold;

[0133] The node vector determining unit 603 is configured to determine the vector representation corresponding to the target object node according to the connection relationship between each adjacent object node in the adjacent node set;

[0134] The sub-relationship network division unit 604 is configured to execute the division of the relationship network diagram into at least two sub-relationship networks according to the vector representation corresponding to each of the target object nodes, and perform feature extraction on each of the sub-relationship networks to obtain feature extraction results corresponding to each of the sub-relationship networks, and use each of the feature extraction results as a multi-meaning feature corresponding to the original interaction data.

[0135] In one possible implementation, the adjacent node set determination unit 602 is specifically configured to execute taking any object node in the relationship network diagram as a target object node, determining the directly connected nodes and indirectly connected nodes corresponding to the target object node; the number of object nodes connected in series between the indirectly connected node and the target object node is less than a preset threshold; and taking the directly connected node and the indirectly connected node as the adjacent object nodes.

[0136] In one possible implementation, the node vector determination unit 603 is specifically configured to construct a node sequence corresponding to the target object node according to the connection relationship between each adjacent object node in the adjacent node set; and perform vector training using the node sequence corresponding to each target object node to obtain a vector representation corresponding to each target object node.

[0137] In a possible implementation, the node vector determination unit 603 is further specifically configured to execute the steps of taking an object node in the adjacent node set as a sampling node; determining the next object node of the sampling node; the next object node being at least one object node among the directly connected nodes of the sampling node; taking the next object node as the sampling node, and returning to the step of determining the next object node of the sampling node until the number of sampling nodes meets a preset number threshold; and constructing a node sequence corresponding to the target object node according to the node information corresponding to each of the sampling nodes.

[0138] In one possible implementation, the sub-relationship network division unit 604 is specifically configured to determine the distance between any two object nodes in the relationship network diagram according to the vector representation corresponding to each target object node; divide the relationship network diagram according to the distance between any two object nodes in the relationship network diagram to obtain the at least two sub-relationship networks; wherein the at least two sub-relationship networks include a first sub-relationship network and a second sub-relationship network, the distance between any two object nodes in the first sub-relationship network or the second sub-relationship network is less than a first distance threshold, and the distance between any object node in the first sub-relationship network and any object node in the second sub-relationship network is greater than a second distance threshold.

[0139] In one possible implementation, the relationship network diagram acquisition unit 601 is specifically configured to execute the acquisition of the original relationship network diagram of the original interactive data; the original relationship network diagram is used to characterize the interactive relationship between original data objects of multiple different data types, and the original relationship network diagram includes nodes representing each of the original data objects; determine the nodes to be deleted in the original relationship network diagram; the nodes to be deleted include the original nodes of the original data objects corresponding to the data types other than the target data type; delete the nodes to be deleted from the original relationship network diagram to obtain the relationship network diagram.

[0140] In one possible implementation, the sub-relationship network division unit 604 is specifically configured to execute acquisition of node attribute information corresponding to the node to be deleted; the node attribute information is used to characterize the interactive relationship between the object node in each of the sub-relationship networks and the node to be deleted; feature extraction is performed based on the node attribute information and each of the sub-relationship networks to obtain feature extraction results corresponding to each of the sub-relationship networks.

[0141] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0142] Each module in the above network graph feature extraction device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module above.

[0143] Figure 7 is a block diagram of an electronic device 700 for implementing a network graph feature extraction method according to an exemplary embodiment. For example, the electronic device 700 may be a server. Figure 7 , the electronic device 700 includes a processing component 720, which further includes one or more processors, and a memory resource represented by a memory 722 for storing instructions that can be executed by the processing component 720, such as an application. The application stored in the memory 722 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 720 is configured to execute the instructions to perform the above method.

[0144] The electronic device 700 may further include a power supply component 724 configured to perform power management of the electronic device 700, a wired or wireless network interface 726 configured to connect the electronic device 700 to a network, and an input / output (I / O) interface 728. The electronic device 700 may operate based on an operating system stored in the memory 722, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, or the like.

[0145] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, such as a memory 722 including instructions, and the instructions can be executed by a processor of the electronic device 700 to perform the above method. The storage medium can be a computer-readable storage medium, for example, the computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0146] In an exemplary embodiment, a computer program product is further provided. The computer program product includes instructions. The instructions can be executed by a processor of the electronic device 700 to complete the above method.

[0147] It should be noted that the above-mentioned devices, electronic devices, computer-readable storage media, computer program products, etc. may also include other implementation methods according to the description of the method embodiments. The specific implementation methods can refer to the description of the relevant method embodiments, which will not be described one by one here.

[0148] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

[0149] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A network graph feature extraction method, characterized in that: The method comprises: Obtaining a relationship network diagram corresponding to the original interactive data; the relationship network diagram is used to represent the interactive relationship between multiple data objects in the original interactive data, and the relationship network diagram includes object nodes representing each of the data objects and connection edges representing each of the interactive relationships; Taking any object node in the relationship network diagram as a target object node, determining an adjacent node set corresponding to the target object node; the adjacent node set includes adjacent object nodes connected to the target object node, and the number of object nodes connected in series between the adjacent object node and the target object node is less than a preset threshold; Determine the vector representation corresponding to the target object node according to the connection relationship between each adjacent object node in the adjacent node set; According to the vector representation corresponding to each of the target object nodes, the relationship network diagram is divided into at least two sub-relationship networks, and feature extraction is performed on each of the sub-relationship networks to obtain feature extraction results corresponding to each of the sub-relationship networks, and each of the feature extraction results is used as a multi-meaning feature corresponding to the original interaction data; including: obtaining node attribute information corresponding to the node to be deleted in the original relationship network diagram of the original interaction data; the node attribute information is used to characterize the interaction relationship between the object node in each of the sub-relationship networks and the node to be deleted; feature extraction is performed based on the node attribute information and each of the sub-relationship networks to obtain feature extraction results corresponding to each of the sub-relationship networks.

2. The method according to claim 1, characterized in that The step of taking any object node in the relationship network graph as a target object node and determining a set of adjacent nodes corresponding to the target object node includes: Taking any object node in the relationship network diagram as a target object node, determining a directly connected node and an indirectly connected node corresponding to the target object node; the number of object nodes connected in series between the indirectly connected node and the target object node is less than a preset threshold; The directly connected node and the indirectly connected node are regarded as the adjacent nodes.

3. The method according to claim 1, characterized in that The step of determining the vector representation corresponding to the target object node according to the connection relationship between the adjacent object nodes in the adjacent node set includes: Constructing a node sequence corresponding to the target object node according to the connection relationship between each adjacent object node in the adjacent node set; Vector training is performed using the node sequence corresponding to each of the target object nodes to obtain the vector representation corresponding to each of the target object nodes.

4. The method according to claim 3, characterized in that The step of constructing a node sequence corresponding to the target object node according to the connection relationship between the adjacent object nodes in the adjacent node set includes: Taking a certain object node in the adjacent node set as a sampling node; Determine a next object node of the sampling node; the next object node is at least one object node among directly connected nodes of the sampling node; Taking the next object node as the sampling node, and returning to the step of determining the next object node of the sampling node until the number of the sampling nodes meets a preset number threshold; According to the node information corresponding to each of the sampling nodes, a node sequence corresponding to the target object node is constructed.

5. The method according to claim 1, characterized in that The step of dividing the relationship network diagram into at least two sub-relationship networks according to the vector representation corresponding to each of the target object nodes includes: Determine the distance between any two of the object nodes in the relationship network diagram according to the vector representation corresponding to each of the target object nodes; Dividing the relationship network diagram according to the distance between any two of the object nodes in the relationship network diagram to obtain the at least two sub-relationship networks; Among them, the at least two sub-relationship networks include a first sub-relationship network and a second sub-relationship network, the distance between any two object nodes in the first sub-relationship network or the second sub-relationship network is less than a first distance threshold, and the distance between any object node in the first sub-relationship network and any object node in the second sub-relationship network is greater than a second distance threshold.

6. The method according to claim 1, characterized in that The obtaining of the relationship network diagram corresponding to the original interaction data includes: Acquire an original relationship network diagram of the original interactive data; the original relationship network diagram is used to represent the interactive relationship between multiple original data objects of different data types, and the original relationship network diagram includes nodes representing each of the original data objects; Determine the nodes to be deleted in the original relationship network diagram; the nodes to be deleted include original nodes of original data objects corresponding to data types other than the target data type; The node to be deleted is deleted from the original relationship network diagram to obtain the relationship network diagram.

7. A network graph feature extraction device, characterized in that: The device comprises: A relationship network graph acquisition unit is configured to execute acquisition of a relationship network graph corresponding to the original interactive data; the relationship network graph is used to represent the interactive relationship between multiple data objects in the original interactive data, and the relationship network graph includes object nodes representing each of the data objects and connection edges representing each of the interactive relationships; The adjacent node set determining unit is configured to execute taking any object node in the relationship network diagram as a target object node and determining an adjacent node set corresponding to the target object node; the adjacent node set includes adjacent object nodes connected to the target object node, and the number of object nodes connected in series between the adjacent object node and the target object node is less than a preset threshold; A node vector determination unit is configured to determine a vector representation corresponding to the target object node according to a connection relationship between adjacent object nodes in the adjacent node set; A sub-relationship network division unit is configured to divide the relationship network diagram into at least two sub-relationship networks according to the vector representation corresponding to each of the target object nodes, and perform feature extraction on each of the sub-relationship networks to obtain feature extraction results corresponding to each of the sub-relationship networks, and use each of the feature extraction results as a multi-meaning feature corresponding to the original interaction data; Among them, the sub-relationship network division unit is specifically configured to execute the acquisition of node attribute information corresponding to the node to be deleted in the original relationship network diagram of the original interaction data; the node attribute information is used to characterize the interaction relationship between the object node in each of the sub-relationship networks and the node to be deleted; feature extraction is performed based on the node attribute information and each of the sub-relationship networks to obtain feature extraction results corresponding to each of the sub-relationship networks.

8. The device according to claim 7, characterized in that The adjacent node set determination unit is specifically configured to execute any object node in the relationship network diagram as a target object node, determine the directly connected nodes and indirectly connected nodes corresponding to the target object node; the number of object nodes connected in series between the indirectly connected node and the target object node is less than a preset threshold; and use the directly connected node and the indirectly connected node as the adjacent object nodes.

9. The device according to claim 7, characterized in that The node vector determination unit is specifically configured to construct a node sequence corresponding to the target object node according to the connection relationship between each adjacent object node in the adjacent node set; use the node sequence corresponding to each target object node to perform vector training to obtain the vector representation corresponding to each target object node.

10. The device according to claim 9, characterized in that The node vector determination unit is further specifically configured to execute the steps of taking an object node in the adjacent node set as a sampling node; determining the next object node of the sampling node; the next object node being at least one object node among the directly connected nodes of the sampling node; taking the next object node as the sampling node, and returning to the step of determining the next object node of the sampling node until the number of sampling nodes meets a preset number threshold; and constructing a node sequence corresponding to the target object node according to the node information corresponding to each of the sampling nodes.

11. The device according to claim 7, characterized in that The sub-relationship network division unit is specifically configured to determine the distance between any two object nodes in the relationship network diagram according to the vector representation corresponding to each target object node; divide the relationship network diagram according to the distance between any two object nodes in the relationship network diagram to obtain the at least two sub-relationship networks; wherein the at least two sub-relationship networks include a first sub-relationship network and a second sub-relationship network, the distance between any two object nodes in the first sub-relationship network or the second sub-relationship network is less than a first distance threshold, and the distance between any object node in the first sub-relationship network and any object node in the second sub-relationship network is greater than a second distance threshold.

12. The device according to claim 7, characterized in that The relationship network diagram acquisition unit is specifically configured to execute the acquisition of the original relationship network diagram of the original interactive data; the original relationship network diagram is used to characterize the interactive relationship between original data objects of multiple different data types, and the original relationship network diagram includes nodes representing each of the original data objects; determine the nodes to be deleted in the original relationship network diagram; the nodes to be deleted include the original nodes of the original data objects corresponding to data types other than the target data type; delete the nodes to be deleted from the original relationship network diagram to obtain the relationship network diagram.

13. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the network graph feature extraction method as described in any one of claims 1 to 6.

14. A computer-readable storage medium, characterized in that: When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the network graph feature extraction method as described in any one of claims 1 to 6.

15. A computer program product, comprising instructions, characterized in that: When the instruction is executed by a processor of an electronic device, the electronic device is enabled to execute the network graph feature extraction method as described in any one of claims 1 to 6.

Images