A method and related device for determining community characteristics based on relationship network

Abstract

The embodiment of the present application discloses a method for determining community features based on a relationship network, the method comprising: supervised training of a neural network model using a first training sample, and determining a community feature extraction model based on the hidden layer of the trained neural network model, wherein the first training sample includes a community identifier of the community to which a sample node in a sample relationship network belongs, and the account type of the sample node; and determining the community features corresponding to the node to be identified using a community feature extraction model based on the community identifiers of multiple communities to which the node to be identified belongs, wherein the community features are used to identify the network relationships of the node to be identified in the relationship network. The present application can determine a community feature extraction model from the neural network model obtained through training, so that the community features of the node to be identified extracted using the community feature extraction model can have a data structure suitable for a classification model and also carry more information related to the account type.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a method and related device for determining community characteristics based on a relationship network. Background Art

[0002] With the widespread use of the internet, users can now fulfill their social needs, shopping, entertainment, and financial transactions through their online activities. Users' online activities connect with each other, forming a variety of relationship networks. For example, friendships can form social networks, while transaction relationships can form financial networks.

[0003] A user's online account can be considered a node in a relationship network. The connections between nodes reflect past network activity and indicate the degree of connection between them. Based on the degree of connection between nodes, the relationship network can be divided into multiple communities. A community is a collection of closely connected nodes in a relationship network, similar to a clustering of unstructured graph data. Furthermore, the same node often belongs to multiple communities.

[0004] It can be seen from this that the multiple communities to which a node belongs can reflect the network relationships of the accounts corresponding to the nodes. If the community features used to express the network relationships of accounts can be accurately obtained based on the community, it can play a significant role in application scenarios such as account identification and account classification. Summary of the Invention

[0005] In order to solve the above technical problems, the present application provides a community feature determination method based on a relational network. A neural network model for judging the account type of a node based on the community identification of the node can be obtained through supervised training, and a community feature extraction model can be determined based on the hidden layer of the neural network model. In this way, the community features of the node to be identified extracted using the community feature extraction model can have a data structure suitable for the classification model and carry more information related to the account type.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] In a first aspect, an embodiment of the present application provides a method for determining community characteristics based on a relationship network, the method comprising:

[0008] Determining a neural network model and a corresponding first training sample, wherein the first training sample includes a community identifier of a community to which a sample node in a sample relationship network belongs, and an account type of the sample node;

[0009] Performing supervised training on the neural network model using the first training sample, and determining a community feature extraction model based on a hidden layer of the trained neural network model;

[0010] determining a plurality of communities to which the nodes to be identified in the relationship network belong;

[0011] According to the community identifiers of the multiple communities, the community feature corresponding to the node to be identified is determined by the community feature extraction model, where the community feature is used to identify the network relationship of the node to be identified in the relationship network.

[0012] In a second aspect, an embodiment of the present application provides a community feature determination device based on a relationship network, the device comprising a first determination unit, a first training unit, a second determination unit, and a third determination unit:

[0013] The first determining unit is configured to determine a neural network model and a corresponding first training sample, wherein the first training sample includes a community identifier of a community to which a sample node in a sample relationship network belongs, and an account type of the sample node;

[0014] The first training unit is configured to perform supervised training on the neural network model using the first training sample, and determine a community feature extraction model based on a hidden layer of the trained neural network model;

[0015] The second determining unit is configured to determine a plurality of communities to which the node to be identified in the relationship network belongs;

[0016] The third determining unit is configured to determine, based on the community identifiers of the multiple communities, a community feature corresponding to the node to be identified by using the community feature extraction model, wherein the community feature is used to identify a network relationship between the node to be identified in the relationship network.

[0017] In a possible implementation, the fourth determining unit is specifically configured to:

[0018] Determining input data for a classification model based on the account characteristics corresponding to the node to be identified and the community characteristics;

[0019] The account type of the node to be identified is determined according to the input data through the classification model.

[0020] In a possible implementation, the apparatus further includes a fifth determining unit, a sixth determining unit, and a second training unit:

[0021] The fifth determining unit is configured to determine the sample community feature of the sample node based on the first training sample and the community feature extraction model;

[0022] The sixth determining unit is configured to determine a classification model and a corresponding second training sample, wherein the second training sample includes a sample account feature and a sample community feature of the sample node, and an account type of the sample node;

[0023] The second training unit is used to perform supervised training on the classification model using the second training samples to obtain the classification model.

[0024] In a possible implementation, the relationship network is an eigenvalue relationship network generated based on eigenvalue transfer, and the fourth determining unit is specifically configured to:

[0025] The account type of the node to be identified is determined to be a normal account or an abnormal account according to the account characteristics corresponding to the node to be identified and the community characteristics.

[0026] In a third aspect, an embodiment of the present application provides a computer device, the device comprising a processor and a memory:

[0027] The memory is used to store program code and transmit the program code to the processor;

[0028] The processor is configured to execute the community feature determination method based on a relationship network described in the first aspect according to the instructions in the program code.

[0029] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and the computer program is used to execute the community feature determination method based on the relationship network described in the first aspect.

[0030] As can be seen from the above technical solution, when it is necessary to determine the community characteristics corresponding to the node to be identified, the neural network model can first be supervised and trained using a first training sample. Since this first training sample includes the community identifier of the community to which the sample node belongs in the sample relationship network, as well as the account type of the sample node, during the training process, the hidden layer of the neural network model can extract the community characteristics corresponding to the sample node based on the community identifier of the sample node input into the input layer. This community characteristic is then passed to the output layer to identify the account type. Model training is performed under the supervision of the account type of the first training sample, so that the trained neural network model has the ability to determine the node account type based on the community identifier of the community to which the node belongs. Based on this training, the accuracy and performance of the hidden layer of the neural network model in extracting community characteristics can be improved. The community feature extraction model is determined based on the hidden layer of the trained neural network model. When it is necessary to extract the community features of the node to be identified, the multiple communities to which the node to be identified belongs in the relationship network can be determined first. Then, based on the community identifiers of the multiple communities, the community features corresponding to the node to be identified are determined through the community feature extraction model. The community features can identify the network relationships of the node to be identified in the relationship network, that is, the community characteristics of the node to be identified in the relationship network are retained. At the same time, the neural network model is also a type of classification model. Therefore, the community features obtained through the community feature extraction model are closer to classification business and have a data structure suitable for classification models. Since the subsequent application scenarios of community features, such as account identification and classification, often use classification models, the data structure of the community features can have better applicability in the classification model and can provide assistance in improving the classification accuracy of the classification model. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0032] Figure 1 A schematic diagram of a method for determining community characteristics based on a relationship network in an actual application scenario provided by an embodiment of the present application;

[0033] Figure 2 A flowchart of a method for determining community characteristics based on a relationship network provided in an embodiment of the present application;

[0034] Figure 3 A schematic diagram of a method for determining community characteristics based on a relationship network in an actual application scenario provided by an embodiment of the present application;

[0035] Figure 4 A schematic diagram of a method for determining community characteristics based on a relationship network in an actual application scenario provided by an embodiment of the present application;

[0036] Figure 5 A structural block diagram of a community feature determination device based on a relationship network provided in an embodiment of the present application;

[0037] Figure 6 A structural diagram of a computer device provided in an embodiment of the present application;

[0038] Figure 7 A structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0039] The embodiments of the present application are described below with reference to the accompanying drawings.

[0040] As the complexity of relationship networks continues to increase, accurately analyzing the nodes within them becomes increasingly difficult. Community features are an important reference for analyzing nodes within relationship networks. However, related technologies often face challenges with extracting these features, often resulting in high complexity and low accuracy.

[0041] In order to solve the above technical problems, the present application provides a community feature determination method based on a relational network. A neural network model for judging the account type of a node based on the community identification of the node can be obtained through supervised training, and a community feature extraction model can be determined based on the hidden layer of the neural network model. In this way, the community features of the node to be identified extracted using the community feature extraction model can have a data structure suitable for the classification model and carry more information related to the account type.

[0042] It is understandable that the method can be applied to a processing device, which is a processing device with a feature extraction function, for example, a terminal device or a server with a feature extraction function. The method can be executed independently by a terminal device or a server, or it can be applied to a network scenario in which a terminal device and a server communicate, and the terminal device and the server cooperate to run. Among them, the terminal device can be a mobile phone, a desktop computer, a personal digital assistant (PDA for short), a tablet computer and other devices. The server can be understood as an application server or a web server. In actual deployment, the server can be an independent physical server or a server cluster or distributed system composed of multiple physical servers. The terminal and the server can be directly or indirectly connected via wired or wireless communication, and this application is not limited here.

[0043] In addition, this application also relates to artificial intelligence (AI) technology. Artificial intelligence is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to achieve the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is the study of the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making.

[0044] Artificial intelligence technology is a comprehensive discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning / deep learning. This application primarily addresses machine learning technologies.

[0045] Machine learning (ML) is a multidisciplinary field that encompasses probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is at the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications span all areas of AI. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0046] In the embodiments of the present application, models involved in neural network models, classification models, etc. can be trained through machine learning technology.

[0047] To facilitate understanding of the technical solution of the present application, a method for determining community characteristics based on a relationship network provided in an embodiment of the present application will be introduced below in conjunction with an actual application scenario.

[0048] See also Figure 1 , Figure 1 A schematic diagram of a method for determining community characteristics based on a relationship network in an actual application scenario provided by an embodiment of the present application. In this actual application scenario, the processing device is a server 101, and the neural network model is a classification model having three layers: an input layer, a hidden layer, and an output layer.

[0049] To incorporate community features into determining node account types, it is first necessary to obtain community features that reflect the account type. In this practical application scenario, the neural network model can be supervised and trained using a first training sample. This first training sample includes the community identifier of the community to which the sample node belongs in the sample relationship network, as well as the account type of the sample node. After obtaining the first training sample, server 101 can input the first training sample into the input layer of the neural network model. The input layer can transmit the data in the first training sample to the hidden layer, which can then extract community features from the community identifier in the first training sample. This allows the extracted feature data to carry more valid data that reflects the account type of the sample node, such as the community characteristics of the sample node in the sample relationship network. After feature extraction, the hidden layer can transmit the obtained feature data to the output layer, which performs classification, determines the account type, and outputs it. During training, the neural network model can compare the account type output by the output layer with the account type of the sample node in the first training sample and adjust the model parameters based on the comparison results, thereby training a neural network model that can accurately determine node account types based on community identifiers.

[0050] It can be understood that since the hidden layer has the function of extracting community features from the community identifier, after training is completed, the server 101 can determine the community feature extraction model for extracting community features based on the hidden layer. In this actual application scenario, the server 101 can extract the model parameters of the input layer and the hidden layer in the neural network model and determine it as the community feature extraction model. Thus, after the community identifier corresponding to the node is input into the community feature extraction model, the community features corresponding to the node can be obtained. Since the community identifier of the community can identify the community to which the node belongs, and the neural network model is a model for determining the type of node account, the community features extracted by the hidden layer can not only reflect the network relationship of the node in the relationship network, but also carry relevant information for determining the account type.

[0051] When applying the community feature extraction model, the server 101 can first obtain a relationship network including multiple nodes. Based on the relationship network, the server 101 can determine the community to which each node belongs. When determining the community feature of a node to be identified, the server 101 can first determine the community to which the node to be identified belongs based on the relationship network, and then determine the community identifier corresponding to the community to which the node to be identified belongs. The server 101 can obtain the community feature corresponding to the node to be identified through the community identifier and the community feature extraction model. The community feature can carry relevant information for determining the account type of the node to be identified. Therefore, when the account type of the node to be identified is determined by the community feature, the judgment result obtained can be more accurate.

[0052] Furthermore, because this neural network model is a type of classification model, the data processed by it typically possesses a data structure suitable for classification models. This means that the community features extracted by the hidden layer are applicable to other classification models. When these community features are applied to other classification models for classification, this classification model can more quickly and accurately identify and process the node account, further improving classification efficiency and accuracy.

[0053] Next, a method for determining community characteristics based on a relationship network provided in an embodiment of the present application will be introduced with reference to the accompanying drawings.

[0054] See also Figure 2 , Figure 2 A flowchart of a method for determining community characteristics based on a relationship network provided in an embodiment of the present application, the method comprising:

[0055] S201: Determine a neural network model and a corresponding first training sample.

[0056] In order to be able to integrate community features into various node application scenarios, the processing device first needs to be able to obtain relatively accurate community features corresponding to each node. Among them, the node is used to reflect the user identification of the user in the relationship network, for example, it can be the network account used by the user in the relationship network; the relationship network is a network composed of the connection relationship between multiple nodes; the community refers to a node set composed of nodes with a relatively tight topological structure on the relationship network, and the community feature is used to reflect the network relationship of the node in the relationship network. By adding this community feature, the processing device can better combine the network relationship of the node in the relationship network when analyzing the node, so that the analysis result is more effective and accurate.

[0057] Among them, the relationship network can include various types, for example, a social network generated based on user communication on social software, a capital network generated based on capital transactions between users, etc. In the capital network, a node can be a capital account used by a user to manage funds.

[0058] It is understandable that in the application scenarios related to nodes, the application scenarios that classify and analyze the account types corresponding to the nodes account for a large proportion. Among them, the account type is used to identify the type of the node in the relationship network, for example, whether the node is an abnormal node, whether it is a safe node, etc. Therefore, in order to be able to more accurately judge the account type corresponding to the node in these application scenarios, the processing device can make the extracted node community features carry more valid data that can reflect the account type of the node. At the same time, in the above application scenarios, various classification models are usually used to classify and analyze the account types, and the data suitable for the classification model often has a unique data structure. Therefore, in order to further improve the applicability of community features in these classification models and enable the classification model to more accurately utilize the community features, the processing device can also extract community features based on the data structure of the classification model as much as possible.

[0059] Among them, as one of the classic classification models, the neural network model has the ability to quickly and efficiently classify and analyze samples based on certain sample characteristics. Therefore, in an embodiment of the present application, in order to obtain community features with the above characteristics, the processing device can first train a neural network model, and the trained neural network model is used to extract the community features of the node based on the relevant information of the node, and classify the account type of the node based on the community features. Thus, in the process of the trained neural network model working, some of the models therein can have the function of extracting community features based on the relevant information of the node. Since the trained neural network model is a classification model for classifying account types, the community features extracted by the model not only meet the data structure of the classification model, but also carry valid data related to the account type during the extraction process, thereby meeting the requirements for community features in this application.

[0060] To obtain the trained neural network model, the processing device may first determine an initial neural network model and a corresponding first training sample. The first training sample includes a community identifier of the community to which the sample node in the sample relationship network belongs, and the account type of the sample node. The community identifier may be determined by the processing device based on the affiliation of the sample node in the sample network, and may be used to identify the community to which the sample node belongs in the sample relationship network. The affiliation indicates that the sample node is one of the nodes constituting the sample community. The sample relationship network may be any type of relationship network, and the sample node may be any node in the sample relationship network whose account type is known.

[0061] In addition to identifying the community within the relationship network, the community identifier can also carry other relevant community information to facilitate the model's determination of community characteristics. This community information can reflect some of the community's characteristics. For example, if the relationship network is a financial network, the community may be composed of user accounts with frequent financial transactions. The community identifier of this community can include information such as the transaction frequency and transaction time of users within the community, as well as the proportion of abnormal nodes within the community. An abnormal node refers to a node that engages in abnormal financial operations, such as a node that conducts a large number of asset transactions in a short period of time. Therefore, the neural network model can extract more accurate community characteristics based on these community characteristics, and can accurately determine the account type of the sample node.

[0062] S202: Performing supervised training on the neural network model using the first training sample, and determining a community feature extraction model based on the hidden layer of the trained neural network model.

[0063] After determining the first training sample, the processing device can use the first training sample to perform supervised training on the neural network model. Supervised training refers to a model training mode with input and expected output. In this embodiment of the application, the input can be the community identifier of the community to which the sample node belongs, and the expected output can be the account type of the sample node.

[0064] It is understood that a neural network model is generally composed of an input layer, a hidden layer, and an output layer. The input layer is used to receive data and transmit the data to the hidden layer; the hidden layer is used to process the data so that the processed data has a classification function that is more suitable for the neural network model; and the output layer is used to classify the processed data and output the classification result. During the training process of the embodiment of the present application, the input layer of the neural network model can receive the first training sample and transmit the first training sample to the hidden layer; the hidden layer can extract the community features of the sample node suitable for account type classification based on the community identifier of the community to which the sample node in the first training sample belongs, and transmit the community features to the output layer; the output layer can classify and analyze the account type corresponding to the sample node based on the community features and output the account type corresponding to the sample node. In the supervised training mode, the processing device can compare the account type of the sample node in the first training sample with the account type output by the neural network model, and feed the comparison result back to the neural network model, so that the neural network model can modify its own model parameters based on the comparison result so that the output account type is more in line with the actual account type of the sample node.

[0065] It can be seen that after training the neural network model in the above manner, the hidden layer of the neural network model can gradually have the ability to accurately extract the community characteristics of the sample node based on the community identifier of the community to which the sample node belongs, and in order to enable the output layer to accurately judge the account type of the sample node, the community characteristics can also have more valid data for reflecting the account type of the sample node. At the same time, since the neural network model is a classification model, and the hidden layer processes the data input by the input layer based on the data structure of the classification model, the community characteristics extracted by the hidden layer can have a data structure suitable for the classification model. Therefore, the community characteristics extracted by the hidden layer of the trained neural network model can meet the requirements of this application for community characteristics.

[0066] Based on this, in order to accurately extract community features suitable for account type classification in the form of a classification model data structure, the processing device can split the trained neural network model and determine the community feature extraction model based on the hidden layer of the trained neural network model. This is equivalent to stripping the original output layer from the neural network model and using the hidden layer of the neural network model as the output layer of the community feature model. Therefore, the community feature extraction model can be used to extract the community features corresponding to the node based on the community identifier of the community to which the node belongs.

[0067] S203: Determine multiple communities to which the node to be identified belongs in the relationship network.

[0068] After obtaining the community feature extraction model, the processing device can apply it to the community feature extraction of the node. For example, when the processing device needs to classify and analyze the account type of a node to be identified based on the community features of the node to be identified in the relationship network, it can first determine the multiple communities to which the node to be identified belongs in the relationship network. In one possible implementation, since the relationship network includes multiple nodes, and the community is composed of some or all of the multiple nodes, the processing device can determine the multiple communities to which the node to be identified belongs based on the node to be identified using the information of the nodes contained in each community.

[0069] To enable the community feature extraction model to more accurately extract community features, the processing device can selectively select the relationship network. For example, the content of community identifiers may vary across different types of relationship networks. For example, a community identifier in a financial network may include information about the flow of funds within the community, while a community identifier in a social network may include information about the chat frequency of nodes within the community. Because the neural network model is trained through supervised training based on a sample relationship network, the community feature extraction model may be more suitable for determining community features based on the community identifier of the sample relationship network type.

[0070] Based on this, to further improve the accuracy of community features, when using the community feature extraction model, the processing device can select nodes in a relationship network of the same or similar type as the sample relationship network as the nodes to be identified. For example, if the sample relationship network is a financial network, the relationship network in which the node to be identified resides can also be a financial network or a currency network.

[0071] At the same time, there are many methods for determining the belonging community, which can be roughly divided into two categories. One is the heuristic algorithm, which uses the propagation property, complete subgraph characteristics and matrix decomposition of complex networks to obtain the community to which the node belongs, such as the Speaker-Listener Label Propagation Algorithm (SLPA) algorithm, the Coefficient Page Rank Affinity Propagation (COPRA) algorithm, the Cluster Percolation Method (CPM) algorithm, and the Nonnegative Matrix Factorization (NMF) algorithm. The other is the Graph Convolutional Network (GCN) algorithm, which embeds the nodes and edges in the network into a low-dimensional space and converts them into structured data. Then, traditional non-graph clustering algorithms are used, such as the Network Embedding and K-means clustering algorithms based on graph neural networks, and the Density-Based Spatial Clustering of Applications (DSC) algorithm. withNoise, referred to as DBSCAN) algorithm.

[0072] In the embodiment of the present application, an improved partitioning algorithm based on COPRA may be used to determine the belonging community.

[0073] In addition, since the node's belonging community determined by different methods may also be different, in order to enable the community feature extraction model to more accurately determine the community features of the node to be identified, the processing device can use the same method as that used to determine the community to which the sample node belongs to determine the community to which the node to be identified belongs, thereby ensuring that the belonging community can fit the model characteristics of the community feature extraction model as much as possible.

[0074] S204: Determine, based on the community identifiers of the multiple communities, the community features corresponding to the node to be identified using a community feature extraction model.

[0075] After determining the multiple communities to which the node to be identified belongs, the processing device can determine community identifiers for the multiple communities and then use the community identifiers as input data for the community feature extraction model. Using the community feature extraction model, the processing device can determine community features corresponding to the node to be identified. These community features are used to identify the network relationships of the node to be identified in the relationship network. These network relationships can reflect the degree of association between the node to be identified and other nodes in the relationship network, as well as the degree of association between the node to be identified and various communities in the relationship network.

[0076] Using these community characteristics, the processing device can analyze implicit characteristics of the node to be identified. Implicit characteristics are characteristics that cannot be directly obtained from the node's explicit information. Explicit information refers to node information that can be directly obtained without processing. For example, if the relationship network is a financial network, explicit information may include information such as the amount and frequency of fund transfers into and out of the node to be identified. Using this explicit information, the processing device can analyze some of the explicit characteristics of the node to be identified. For example, by analyzing the amount of funds transferred into and out of the node to be identified, the processing device can determine whether the node to be identified has any abnormal financial operations.

[0077] However, relying solely on explicit characteristics can make it difficult to accurately judge the node to be identified. Some nodes to be identified may not contain any abnormalities in their own information. For example, in a financial network, the number and frequency of transfers of a node to be identified may be within the normal range, but the node to be identified may frequently have financial transactions with some abnormal nodes or some abnormal communities. Therefore, in actual situations, the probability that the node to be identified is an abnormal node is also high. In this case, the processing device can analyze the implicit characteristics of the node to be identified. For example, by determining the community characteristics corresponding to the node to be identified, the implicit characteristics of the node to be identified, such as the community characteristics, can be analyzed. This allows the implicit characteristics of the node to be identified to be analyzed and judged when the community characteristics are applied to other application scenarios.

[0078] As can be seen from the above technical solution, when it is necessary to determine the community characteristics corresponding to the node to be identified, the neural network model can first be supervised and trained using a first training sample. Since this first training sample includes the community identifier of the community to which the sample node belongs in the sample relationship network, as well as the account type of the sample node, during the training process, the hidden layer of the neural network model can extract the community characteristics corresponding to the sample node based on the community identifier of the sample node input into the input layer. This community characteristic is then passed to the output layer to identify the account type. Model training is performed under the supervision of the account type of the first training sample, so that the trained neural network model has the ability to determine the node account type based on the community identifier of the community to which the node belongs. Based on this training, the accuracy and performance of the hidden layer of the neural network model in extracting community characteristics can be improved. The community feature extraction model is determined based on the hidden layer of the trained neural network model. When it is necessary to extract the community features of the node to be identified, the multiple communities to which the node to be identified belongs in the relationship network can be determined first. Then, based on the community identifiers of the multiple communities, the community features corresponding to the node to be identified are determined through the community feature extraction model. The community features can identify the network relationships of the node to be identified in the relationship network, that is, the community characteristics of the node to be identified in the relationship network are retained. At the same time, the neural network model is also a type of classification model. Therefore, the community features obtained through the community feature extraction model are closer to classification business and have a data structure suitable for classification models. Since the subsequent application scenarios of community features, such as account identification and classification, often use classification models, the data structure of the community features can have better applicability in the classification model and can provide assistance in improving the classification accuracy of the classification model.

[0079] It is understandable that the number of hidden layers of the neural network model may also vary depending on the accuracy requirements for data processing. In one possible implementation, the neural network model may include an input layer, an output layer, and k hidden layers located between the input layer and the output layer. The value of k can be adjusted according to the data accuracy requirements for the model. After receiving the relevant data, the input layer can transmit the data to the first hidden layer. The first hidden layer can extract the community characteristics of the node from the data and pass the community characteristics to the subsequent hidden layers for layer-by-layer iterative processing, ultimately obtaining a more accurate and more prominent community characteristic. It can be seen that the partial model composed of any one of the k hidden layers and the model layer before the hidden layer has the function of extracting the community characteristics corresponding to the node based on the community identifier of the community to which the node belongs.

[0080] Based on this, in one possible implementation, the processing device can determine a target hidden layer from the k hidden layers. The target hidden layer can be any one of the k hidden layers. Subsequently, the processing device can generate the community feature extraction model based on the model structure of the neural network model between the input layer and the target hidden layer, with the output layer of the community feature extraction model being the target hidden layer. Thus, in the community feature extraction model, after inputting the community identifier of the community to which the node to be identified belongs, the output layer of the model can output the community feature corresponding to the node.

[0081] Among them, the types of neural network models can also include multiple types, for example, they can include Recurrent Neural Network (RNN) model, Long Short-Term Memory (LSTM) model, Variational Autoencoder (VAE) model, etc. In one possible implementation, the neural network model can be a Multi-Layer Perceptron (MLP) model, and the k hidden layers of the MLP model can iteratively process the extracted community features in the above manner. As mentioned above, each hidden layer can process the data output by the previous model more accurately, that is, the closer the hidden layer is to the output layer, the higher the accuracy of the processed community features, so that the output layer can judge the account type of the node based on a high-precision community feature.

[0082] Therefore, in order to enable the community feature extraction model determined by the hidden layer based on the MLP model to output community features with higher accuracy, the processing device can use the hidden layer closest to the output layer among the k hidden layers as the target hidden layer, so that the community feature extraction model can output the community features of the node to be identified after iterative processing of the k hidden layers.

[0083] In addition to improving model accuracy, to further enhance the accuracy of community features, the processing device can also improve the data input to the model. It is understandable that with the increasing popularity of network technology, the number of users using the network is constantly increasing. The complexity of the relationship network corresponding to each user is also rapidly increasing. Therefore, in the relationship network, each node may belong to multiple communities. For example, when the relationship network is a message exchange network established based on chat software, the node can be the user account in the chat software. The user may have added multiple group chats and exchange messages with multiple users. Each group chat and each user may belong to a different community. When determining the community to which the node belongs, because the node has directly exchanged messages with user nodes in these group chats and other user nodes, the node information included in multiple communities may contain nodes corresponding to the user, which means that it is possible to determine the communities to which the node belongs. Even if the node has only had a few exchanges with a certain node, or the communication frequency is low, it may still be determined to belong to a community.

[0084] This shows that directly determining the communities to which a node belongs based on the affiliation relationship may result in a large number of communities, making it difficult to highlight the primary community to which the node belongs in the relationship network. Consequently, when the community identifiers of these communities are directly applied to the community feature extraction model, the determined community features may not clearly highlight the community characteristics of the node. Based on this, in order to further highlight the community characteristics of the node to be identified and find the primary community connections of the node to be identified in the relationship network, after determining the multiple communities to which the node to be identified belongs, the processing device can first determine the affiliation probabilities of the node to be identified corresponding to each of the multiple communities. This affiliation probability can reflect the closeness of the connection between the node to be identified and the community.

[0085] It is understood that the closer the connection between the node to be identified and a community, the more likely that community is to reflect the community characteristics of the node to be identified in the relationship network. Therefore, the processing device can set a threshold condition for the probability of belonging, and then, based on the probability of belonging, determine a target community from multiple communities whose probability of belonging is greater than the threshold condition. The target community is the community with a closer relationship with the node to be identified among the multiple communities. The processing device can determine the community characteristics corresponding to the node to be identified based on the community identifier of the target community using the community feature extraction model.

[0086] For example, in the aforementioned chat software-based relationship network, the probability of belonging can be determined by the frequency of message exchanges between the node to be identified and multiple communities. The processing device can use threshold conditions to determine which communities have a high frequency of message exchanges with the node to be identified as target communities. These target communities are the primary communities in which the user exchanges messages within the relationship network. Therefore, the community characteristics determined based on these target communities can further highlight the network relationships of the node to be identified within the relationship network, more clearly reflecting the community characteristics of the node to be identified.

[0087] After obtaining the community features corresponding to the node to be identified, the processing device can apply the community features in multiple application scenarios. In one possible implementation, the processing device can use the community features to determine the account type corresponding to the node to be identified. In order to further improve the accuracy of account type determination, the processing device can also analyze the account type based on the community features and other features related to the node to be identified. For example, the processing device can first determine the account features corresponding to the node to be identified. The account features are used to reflect the account characteristics of the node to be identified itself in the relationship network. This part of the account characteristics can be reflected without the community relationship of the node to be identified. For example, when the relationship network is a financial network, the node to be identified can be the user's financial account in the financial network. The processing device can identify the financial account and analyze whether it has abnormal behavior, such as whether it logs in from an abnormal location, transfers an abnormal amount of funds, etc., and determine the account features corresponding to the node to be identified based on the identification results.

[0088] Furthermore, the content of these account characteristics may also vary depending on the type of relationship network. For example, if the relationship network is a social network generated by chat software, the account characteristics may include the account login location, chat time, message sending method, etc. Thus, determining the node account type based on these account characteristics can make the determined account type more closely aligned with the specific function of the relationship network, thereby making the account type more effective and accurate.

[0089] The processing device can determine the account type of the node to be identified based on the account characteristics and community characteristics corresponding to the node to be identified, so that the obtained account type can integrate the account characteristics of the node to be identified itself and the community characteristics in the relationship network, and can show both the explicit characteristics of the node to be identified, that is, the account characteristics that can be directly obtained by identifying the behavior of the node to be identified itself, and the implicit characteristics of the node to be identified, that is, the community characteristics obtained by analyzing the community relationship of the node to be identified.

[0090] For example, in the above-mentioned capital network, explicit characteristics can be characteristics obtained by identifying the behavior of the capital account itself, which can be the amount of transferred funds, login behavior, etc.; implicit characteristics can be some characteristics that cannot be directly obtained by relying solely on the information of the account itself. For example, the behavior of a certain capital account itself may not be abnormal, but in a certain community to which the capital account belongs, there are multiple abnormal capital accounts. At this time, the account type determined by this method can integrate the community characteristics of the capital account and determine the capital account as an account with a certain probability of abnormality, thereby being able to mine the implicit characteristics of the capital account and making the determined account type more accurate.

[0091] When determining the account type of the node to be identified using the above conditions, various methods can be used. As mentioned above, the community features obtained by the community feature extraction model have a data structure suitable for a classification model. Therefore, to make account type determination more efficient and accurate, in one possible implementation, the processing device can determine the account type of the node to be identified using the classification model.

[0092] The processing device can determine the input data of the classification model based on the account characteristics and the community characteristics of the node to be identified. For example, the community characteristics and the account characteristics can be reflected in the form of vectors. The vector (embedding) is a multi-dimensional feature representation of the data. It can reduce the dimension of complex non-Euclidean data to low-dimensional data and obtain a low-dimensional and information-rich feature vector. In the community feature extraction model, the model can convert the community identifiers of multiple communities input by the processing device into low-dimensional community feature vectors through data processing, so that it can retain the community characteristics of the node to be identified and can be more conveniently and effectively applied to the classification model. The processing device can combine the account feature vector and the community feature vector of the node to be identified, and determine the combined vector as the input data of the classification model.

[0093] The processing device can use the classification model to determine the account type of the node to be identified based on the input data. It is understandable that due to the rich variety of relationship network types, the types of account types corresponding to the node to be identified may also vary in different relationship network types. For example, when the relationship network is a social network established based on social software, the account types may include accounts with normal logins and accounts with abnormal logins; when the relationship network is a financial network, the account types may include accounts with normal financial operations and accounts with abnormal financial operations.

[0094] In one possible implementation, the relationship network may be a feature value relationship network generated based on feature value transfer. The feature values ​​may include multiple types, such as currency feature values ​​in games or real currency feature values ​​in funding software. To ensure the asset security of each node in the feature value relationship network, the processing device may determine the account type corresponding to the node.

[0095] In this eigenvalue relationship network, account types can include normal accounts and abnormal accounts. Normal accounts refer to account types with low risk of eigenvalue transfer operations involving this node, and abnormal accounts refer to account types with high risk of eigenvalue transfer operations involving this node. The processing device can determine whether the account type of the node to be identified is a normal account or an abnormal account based on the account type and community characteristics corresponding to the node to be identified, so that the node can be processed accordingly based on the account type. For example, when the processing device determines that the account type of the node is an abnormal account, the eigenvalue transfer function of the node can be suspended to prevent the user's asset loss.

[0096] To accurately determine the account type of the node to be identified using the classification model, the processing device can perform more detailed model training on the classification model to improve model accuracy. As mentioned above, the community features determined by the feature extraction model have a data structure suitable for the classification model and carry valid data that is conducive to account type determination. Therefore, to improve training effectiveness, the processing device can use the community features determined by the community feature extraction model to perform training to obtain the classification model.

[0097] In one possible implementation, the processing device may determine the sample community features of the sample node based on the first training sample and the community feature extraction model. The processing device may then determine a classification model and a corresponding second training sample, where the second training sample includes the sample account features and the sample community features of the sample node, as well as the account type of the sample node.

[0098] The processing device can perform supervised training on the classification model through the second training sample to obtain the above-mentioned classification model, which is used to determine the account type corresponding to the node to be identified based on the account characteristics and community characteristics corresponding to the node to be identified. During the supervised training process, the processing device can input the second training sample into the classification model, analyze the account type output by the classification model, compare whether it is consistent with the account type of the sample node in the second training sample, and feed back the comparison result to the classification model, so that the classification model can adjust the parameters in the model in a targeted manner, so that the output account type is as close as possible to the account type of the actual sample node. After the training is completed, the trained classification model can be used as the classification model to determine the account type.

[0099] In the related art, there are actually other methods for determining community features. For example, the bag-of-words model (one-hot) can be used to encode the community to which the node belongs. If it belongs to a certain community, it is encoded as 1, otherwise the community is encoded as 0. Although this method is simple, it will cause dimensionality explosion and feature sparsity. In a relationship network including hundreds of millions of nodes, millions or tens of millions of communities may be generated, which means there will be millions or tens of millions of dimensions, which is not desirable. In addition, there are also community algorithms that directly study community feature vectors (Community embedding), but these algorithms need to combine or unify node vector (Node embedding) algorithms and clustering algorithms. The complexity generated by coupling these algorithms is high, and there are too many hyperparameters, which makes them less practical.

[0100] As shown in the following table, the comparison results of various community feature extraction methods are provided:

[0101]

[0102] As can be seen from the table, the algorithm provided in the embodiment of the present application can not only better reflect the community characteristics of the node, but also avoid the complex iteration problems caused by the node vector. By using the more mature neural network model technology, a good community feature extraction model can be obtained through supervised training.

[0103] Next, a community feature determination method based on a relationship network provided in an embodiment of the present application will be introduced in conjunction with an actual application scenario.

[0104] See also Figure 3 , Figure 3 A schematic diagram of a method for determining community features based on a relational network in an actual application scenario provided by an embodiment of the present application. In this actual application scenario, the neural network model can be an MLP model with three hidden layers, and the processing device is a server capable of feature extraction. Figure 3 The training process of the community feature extraction model and the classification model used to classify account types in this practical application scenario is demonstrated. First, the server can determine the sample node identifier as a sample. Then, based on the sample node identifier, community algorithms such as SLPA and Copra are used to determine the corresponding community of the sample node.

[0105] Subsequently, the server can determine N communities with a higher probability of belonging from multiple belonging communities, and determine the community identifiers of these communities, which are recorded as Top1_comID, Top1_comID...TopN_comID, where com is the abbreviation of community. The server can normalize these community identifiers, and then use the account type corresponding to the sample node as the training label, and the normalized community identifier as input to supervise the training to obtain a community feature extraction model. The output layer of the community feature extraction model is the last hidden layer of the original MLP model, and the input layer is the input layer of the original MLP model. The community feature extraction model includes the model part between the input layer and the last hidden layer of the original MLP.

[0106] After obtaining the community feature extraction model, the server can use it to determine the community features corresponding to the sample node. Subsequently, the server can use the community features and other features corresponding to the sample node, such as account features, as well as the sample node itself and its corresponding account type, as training samples to perform supervised training on the classification model, resulting in a classification model that can be used to determine the node's account type based on multiple features.

[0107] Some of the training formulas in the model are as follows:

[0108]

[0109] Vector community =NN MLP (Vector input )

[0110] InputFeature classification =Vector community ∥Vector other_feature

[0111] Among them, Vector input represents the input feature vector, Norm(Top_CommunityIDs) represents the normalized vector of each community ID, Indicates that the normalized community ID is determined based on the maximum value of each community ID in the TopN communities and the community ID to which it belongs. N is the number of TopN communities, i is any one of the N communities, and Vector community represents community characteristics, NN MLP (Vector input ) indicates that the input feature vector is processed by the MLP model, InputFeature classification Represents the input vector of the classification model, Vector community∥Vector other_feature Represents the concatenation vector of the community feature vector and other features, where the other features may be account features, etc.

[0112] After training the community feature extraction model and the classification model, in the application process, such as Figure 4 As shown, the server can obtain a node to be identified in a relationship network with the same network type as the sample node, first obtain the belonging community based on the identifier of the node to be identified using the community algorithm, then determine the community features of the node to be identified through the community feature extraction model, and finally input the community features, other features and the identifier of the node to be identified into the classification model to determine the account type corresponding to the node to be identified.

[0113] This application also compares the method of determining account types based on community characteristics with other methods. The comparison results are shown in the following table:

[0114] method AUC value Accuracy Recall F1 value accuracy A 91.15％ 72.17％ 92.40％ 81.04％ 79.59％ B 75.30％ 54.45％ 87.34％ 67.08％ 59.52％ C 75.39％ 62.78％ 84.81％ 72.15％ 69.08％ D 74.87％ 52.66％ 92.94％ 67.23％ 57.21％ A+B 91.64％ 72.35％ 92.76％ 81.29％ 79.84％ A+C 93.04％ 78.26％ 91.13％ 83.85％ 84.21％ A+D 93.35％ 79.15％ 91.32％ 84.54％ 84.80％

[0115] Among them, method A uses the account characteristics of the node to determine the account type, method B uses the node vector to determine the account type, method C uses the community algorithm and the node vector to perform weighted aggregation to determine the account type, and method D uses the community algorithm and the neural network model to determine the account type. As can be seen from the table, the data of methods A+D are all the highest or relatively high among all methods. That is, the parameters such as the accuracy of the account type determined by combining the community characteristics obtained by the technical solution of this application with the account characteristics of the node are all at a high level. This shows that the community characteristics obtained by the technical solution of this application can be better utilized in application scenarios where account type analysis is required.

[0116] Based on the method for determining community characteristics based on a relationship network described in the above embodiment, the present application also provides a device for determining community characteristics based on a relationship network, such as Figure 5 As shown, Figure 5 This is a structural block diagram of a community feature determination device 500 based on a relationship network provided in an embodiment of the present application. The device 500 includes a first determination unit 501, a first training unit 502, a second determination unit 503, and a third determination unit 504:

[0117] A first determining unit 501 is configured to determine a neural network model and a corresponding first training sample, wherein the first training sample includes a community identifier of a community to which a sample node in a sample relationship network belongs, and an account type of the sample node;

[0118] A first training unit 502 is configured to perform supervised training on the neural network model using the first training sample, and determine a community feature extraction model based on a hidden layer of the trained neural network model;

[0119] A second determining unit 503 is configured to determine a plurality of communities to which the node to be identified in the relationship network belongs;

[0120] The third determining unit 504 is configured to determine, based on the community identifiers of the multiple communities, a community feature corresponding to the node to be identified by using the community feature extraction model, wherein the community feature is used to identify the network relationship of the node to be identified in the relationship network.

[0121] In a possible implementation, the neural network model includes an input layer, an output layer, and k hidden layers located between the input layer and the output layer, and the first training unit 502 is specifically configured to:

[0122] Determining a target hidden layer from the k hidden layers;

[0123] The community feature extraction model is generated based on the model structure of the neural network model between the input layer and the target hidden layer, and the output layer of the community feature extraction model is the target hidden layer.

[0124] In a possible implementation, the neural network model is a multilayer perceptron model, and the first training unit 502 is specifically configured to:

[0125] The hidden layer closest to the output layer among the k hidden layers is used as the target hidden layer.

[0126] In a possible implementation, the third determining unit 504 is specifically configured to:

[0127] Determining the belonging probabilities of the to-be-identified nodes to the multiple communities respectively;

[0128] Determining a target community whose belonging probability is greater than a threshold condition from the plurality of communities according to the belonging probability;

[0129] According to the community identifier of the target community, the community feature corresponding to the node to be identified is determined by the community feature extraction model.

[0130] In a possible implementation, the apparatus 500 further includes a fourth determining unit:

[0131] The fourth determining unit is configured to determine the account type of the node to be identified based on the account characteristics corresponding to the node to be identified and the community characteristics.

[0132] In a possible implementation, the fourth determining unit is specifically configured to:

[0133] Determining input data for a classification model based on the account characteristics corresponding to the node to be identified and the community characteristics;

[0134] The account type of the node to be identified is determined according to the input data through the classification model.

[0135] In a possible implementation, the apparatus 500 further includes a fifth determining unit, a sixth determining unit, and a second training unit:

[0136] a fifth determining unit, configured to determine the sample community feature of the sample node according to the first training sample and the community feature extraction model;

[0137] a sixth determining unit, configured to determine a classification model and a corresponding second training sample, wherein the second training sample includes a sample account feature and a sample community feature of the sample node, and an account type of the sample node;

[0138] The second training unit is used to perform supervised training on the classification model through the second training samples to obtain the classification model.

[0139] In a possible implementation, the relationship network is an eigenvalue relationship network generated based on eigenvalue transfer, and the fourth determining unit is specifically configured to:

[0140] The account type of the node to be identified is determined to be a normal account or an abnormal account according to the account characteristics corresponding to the node to be identified and the community characteristics.

[0141] The present application also provides a computer device, which is described below with reference to the accompanying drawings. Figure 6 As shown, the embodiment of the present application provides a device, which may also be a terminal device. The terminal device may be any intelligent terminal including a mobile phone, a tablet computer, a personal digital assistant (PDA), a point of sales (POS), an in-vehicle computer, etc., taking a mobile phone as an example:

[0142] Figure 6 The block diagram shows a partial structure of a mobile phone related to the terminal device provided in the embodiment of the present application. Figure 6The mobile phone includes components such as a radio frequency (RF) circuit 610, a memory 620, an input unit 630, a display unit 640, a sensor 650, an audio circuit 660, a wireless fidelity (WiFi) module 670, a processor 680, and a power supply 690. It will be understood by those skilled in the art that Figure 6 The mobile phone structure shown in the figure does not constitute a limitation to the mobile phone, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0143] The following combination Figure 6 A detailed introduction to the various components of a mobile phone:

[0144] The RF circuit 610 can be used to receive and send signals during information transmission or calls. In particular, after receiving downlink information from the base station, it is sent to the processor 680 for processing; in addition, the designed uplink data is sent to the base station. Generally, the RF circuit 610 includes but is not limited to an antenna, at least one amplifier, a transceiver, a coupler, a low noise amplifier (LNA), a duplexer, etc. In addition, the RF circuit 610 can also communicate with the network and other devices through wireless communication. The above-mentioned wireless communication can use any communication standard or protocol, including but not limited to Global System of Mobile communication (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), email, Short Messaging Service (SMS), etc.

[0145] The memory 620 can be used to store software programs and modules. The processor 680 executes the various functional applications and data processing of the mobile phone by running the software programs and modules stored in the memory 620. The memory 620 may mainly include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as a sound playback function, an image playback function, etc.); the data storage area may store data created based on the use of the mobile phone (such as audio data, a phone book, etc.). In addition, the memory 620 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0146] The input unit 630 can be used to receive input digital or character information, and to generate key signal input related to the user settings and function control of the mobile phone. Specifically, the input unit 630 may include a touch panel 631 and other input devices 632. The touch panel 631, also known as a touch screen, can collect user touch operations on or near it (such as operations performed by the user using any suitable object or accessory such as a finger, stylus, etc. on or near the touch panel 631) and drive the corresponding connection device according to a pre-set program. Optionally, the touch panel 631 may include two parts: a touch detection device and a touch controller. Among them, the touch detection device detects the user's touch direction and detects the signal caused by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device and converts it into touch point coordinates, which are then sent to the processor 680. It can also receive commands sent by the processor 680 and execute them. In addition, the touch panel 631 can be implemented using various types such as resistive, capacitive, infrared and surface acoustic wave. In addition to the touch panel 631, the input unit 630 may further include other input devices 632. Specifically, the other input devices 632 may include, but are not limited to, one or more of a physical keyboard, function keys (such as volume control keys, switch keys, etc.), a trackball, a mouse, and a joystick.

[0147] The display unit 640 can be used to display information input by the user or information provided to the user and various menus of the mobile phone. The display unit 640 may include a display panel 641. Optionally, the display panel 641 may be configured in the form of a liquid crystal display (LCD), an organic light-emitting diode (OLED), etc. Further, the touch panel 631 may cover the display panel 641. When the touch panel 631 detects a touch operation on or near it, it is transmitted to the processor 680 to determine the type of touch event. Subsequently, the processor 680 provides corresponding visual output on the display panel 641 according to the type of touch event. Although in Figure 6 In the embodiment, the touch panel 631 and the display panel 641 are used as two independent components to realize the input and output functions of the mobile phone, but in some embodiments, the touch panel 631 and the display panel 641 can be integrated to realize the input and output functions of the mobile phone.

[0148] The mobile phone may also include at least one sensor 650, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor, wherein the ambient light sensor may adjust the brightness of the display panel 641 according to the brightness of the ambient light, and the proximity sensor may turn off the display panel 641 and / or the backlight when the mobile phone is moved to the ear. As a type of motion sensor, the accelerometer sensor can detect the magnitude of acceleration in all directions (generally three axes), and can detect the magnitude and direction of gravity when stationary. It can be used for applications that identify the posture of the mobile phone (such as horizontal and vertical screen switching, related games, magnetometer posture calibration), vibration recognition related functions (such as pedometer, tapping), etc.; as for other sensors that the mobile phone can also be configured with, such as gyroscopes, barometers, hygrometers, thermometers, infrared sensors, etc., they will not be described here.

[0149] Audio circuit 660, speaker 661, and microphone 662 provide an audio interface between the user and the phone. Audio circuit 660 converts received audio data into electrical signals and transmits them to speaker 661, which then converts them into sound signals for output. Microphone 662, on the other hand, converts collected sound signals into electrical signals, which are then received by audio circuit 660 and converted into audio data. The audio data is then processed by processor 680 and then transmitted to, for example, another phone via RF circuit 610. Alternatively, the audio data can be output to memory 620 for further processing.

[0150] WiFi is a short-range wireless transmission technology. The mobile phone can help users send and receive emails, browse the web and access streaming media through the WiFi module 670. It provides users with wireless broadband Internet access. Figure 6A WiFi module 670 is shown, but it is understandable that it is not an essential component of the mobile phone and can be omitted as needed without changing the essence of the invention.

[0151] Processor 680 is the control center of the mobile phone, connecting all parts of the mobile phone using various interfaces and circuits. It executes software programs and / or modules stored in memory 620 and accesses data stored in memory 620 to perform various functions and process data. Processor 680 may optionally include one or more processing units. Preferably, processor 680 integrates an application processor and a modem processor. The application processor primarily handles the operating system, user interface, and application programs, while the modem processor primarily handles wireless communications. It is understood that the modem processor may not be integrated into processor 680.

[0152] The mobile phone also includes a power supply 690 (such as a battery) for supplying power to various components. Preferably, the power supply can be logically connected to the processor 680 through a power management system, thereby managing charging, discharging, and power consumption through the power management system.

[0153] Although not shown, the mobile phone may also include a camera, a Bluetooth module, etc., which will not be described in detail here.

[0154] In this embodiment, the processor 680 included in the terminal device further has the following functions:

[0155] Determining a neural network model and a corresponding first training sample, wherein the first training sample includes a community identifier of a community to which a sample node in a sample relationship network belongs, and an account type of the sample node;

[0156] Performing supervised training on the neural network model using the first training sample, and determining a community feature extraction model based on a hidden layer of the trained neural network model;

[0157] determining a plurality of communities to which the nodes to be identified in the relationship network belong;

[0158] According to the community identifiers of the multiple communities, the community feature corresponding to the node to be identified is determined by the community feature extraction model, where the community feature is used to identify the network relationship of the node to be identified in the relationship network.

[0159] This application embodiment also provides a server, see Figure 7 As shown, Figure 7The structural diagram of the server 700 provided in the embodiment of the present application, the server 700 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPUs) 722 (for example, one or more processors) and memories 732, and one or more storage media 730 (for example, one or more massive storage devices) for storing application programs 742 or data 744. Among them, the memories 732 and the storage media 730 can be temporary storage or persistent storage. The program stored in the storage medium 730 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Furthermore, the central processing unit 722 can be configured to communicate with the storage medium 730 to execute a series of instruction operations in the storage medium 730 on the server 700.

[0160] The server 700 may also include one or more power supplies 726, one or more wired or wireless network interfaces 750, one or more input and output interfaces 758, and / or one or more operating systems 741, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0161] The steps performed by the server in the above embodiment can be based on Figure 7 The server structure shown.

[0162] The embodiment of the present application further provides a computer-readable storage medium for storing a computer program, wherein the computer program is used to execute any one of the implementation methods of the community feature determination method based on the relationship network described in the aforementioned embodiments.

[0163] Those skilled in the art will understand that all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to program instructions, and the above-mentioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the above-mentioned storage medium can be at least one of the following media: read-only memory (English: read-only memory, abbreviated: ROM), RAM, magnetic disk or optical disk, etc., various media that can store program codes.

[0164] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0165] The above is merely one specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A community feature determination method based on a relationship network, characterized in that: The method comprises: Determining a neural network model and a corresponding first training sample, wherein the first training sample includes a community identifier of a community to which a sample node in a sample relationship network belongs, and an account type of the sample node; The neural network model is supervisedly trained using the first training sample, and a community feature extraction model is determined based on a hidden layer of the trained neural network model, wherein the hidden layer is used to extract community features from the community identifier in the first training sample, the community features extracted by the hidden layer reflect the network relationships of the nodes in the relationship network, and carry relevant information for determining the account type, the account type is used to identify the type of the node in the relationship network, and the community feature extraction model is used to extract the community features corresponding to the node based on the community identifier of the community to which the node belongs; determining a plurality of communities to which the nodes to be identified in the relationship network belong; Determining, based on the community identifiers of the multiple communities, a community feature corresponding to the node to be identified using the community feature extraction model, wherein the community feature is used to identify the network relationship of the node to be identified in the relationship network; Based on the classification model, the account type of the node to be identified is determined according to the account features and the community features corresponding to the node to be identified, wherein the training method of the classification model is: determining the sample community features of the sample node based on the first training sample and the community feature extraction model; determining the classification model and the corresponding second training sample, the second training sample including the sample account features and the sample community features of the sample node, and the account type of the sample node; and performing supervised training on the classification model through the second training sample to obtain the classification model.

2. The method according to claim 1, characterized in that The neural network model includes an input layer, an output layer, and k hidden layers located between the input layer and the output layer. The community feature extraction model is determined based on the hidden layers of the trained neural network model, including: Determining a target hidden layer from the k hidden layers; The community feature extraction model is generated based on the model structure of the neural network model between the input layer and the target hidden layer, and the output layer of the community feature extraction model is the target hidden layer.

3. The method according to claim 2, characterized in that The neural network model is a multi-layer perceptron model, and determining a target hidden layer from the k hidden layers includes: The hidden layer closest to the output layer among the k hidden layers is used as the target hidden layer.

4. The method according to claim 1, wherein The determining, based on the community identifiers of the multiple communities, the community feature corresponding to the node to be identified by using the community feature extraction model includes: Determining the belonging probabilities of the to-be-identified nodes to the multiple communities respectively; Determining a target community whose belonging probability is greater than a threshold condition from the plurality of communities according to the belonging probability; According to the community identifier of the target community, the community feature corresponding to the node to be identified is determined by the community feature extraction model.

5. The method according to claim 1, wherein The determining the account type of the node to be identified based on the classification model and the account characteristics corresponding to the node to be identified and the community characteristics includes: Determining input data for a classification model based on the account characteristics corresponding to the node to be identified and the community characteristics; The account type of the node to be identified is determined according to the input data through the classification model.

6. The method according to claim 1, wherein The relationship network is a feature value relationship network generated based on feature value transfer, and the determining of the account type of the node to be identified based on the account feature corresponding to the node to be identified and the community feature includes: The account type of the node to be identified is determined to be a normal account or an abnormal account according to the account characteristics corresponding to the node to be identified and the community characteristics.

7. A community feature determination device based on a relationship network, characterized in that: The apparatus comprises a first determining unit, a first training unit, a second determining unit, a fourth determining unit, and a third determining unit: The first determining unit is configured to determine a neural network model and a corresponding first training sample, wherein the first training sample includes a community identifier of a community to which a sample node in a sample relationship network belongs, and an account type of the sample node; The first training unit is configured to perform supervised training on the neural network model using the first training sample, and to determine a community feature extraction model based on a hidden layer of the trained neural network model, wherein the hidden layer is configured to extract community features from the community identifier in the first training sample, the community features extracted by the hidden layer reflecting the network relationships of the nodes in the relationship network and carrying relevant information for determining the account type, the account type being used to identify the type of the node in the relationship network, and the community feature extraction model being configured to extract the community features corresponding to the node based on the community identifier of the community to which the node belongs; The second determining unit is configured to determine a plurality of communities to which the node to be identified in the relationship network belongs; The third determining unit is configured to determine, based on the community identifiers of the multiple communities, a community feature corresponding to the node to be identified by using the community feature extraction model, wherein the community feature is used to identify the network relationship of the node to be identified in the relationship network; The fourth determination unit is used to determine the account type of the node to be identified based on the classification model and according to the account characteristics and the community characteristics corresponding to the node to be identified, wherein the training method of the classification model is: determining the sample community characteristics of the sample node based on the first training sample and the community characteristic extraction model; determining the classification model and the corresponding second training sample, the second training sample including the sample account characteristics and the sample community characteristics of the sample node, and the account type of the sample node; and performing supervised training on the classification model through the second training sample to obtain the classification model.

8. The device according to claim 7, characterized in that The neural network model includes an input layer, an output layer, and k hidden layers located between the input layer and the output layer. The first training unit is specifically used to: Determining a target hidden layer from the k hidden layers; The community feature extraction model is generated based on the model structure of the neural network model between the input layer and the target hidden layer, and the output layer of the community feature extraction model is the target hidden layer.

9. The device according to claim 8, characterized in that The neural network model is a multi-layer perceptron model, and the first training unit is specifically used for: The hidden layer closest to the output layer among the k hidden layers is used as the target hidden layer.

10. The device according to claim 7, characterized in that The third determining unit is specifically configured to: Determining the belonging probabilities of the to-be-identified nodes to the multiple communities respectively; Determining a target community whose belonging probability is greater than a threshold condition from the plurality of communities according to the belonging probability; According to the community identifier of the target community, the community feature corresponding to the node to be identified is determined by the community feature extraction model.

11. The device according to claim 7, characterized in that The fourth determining unit is specifically configured to: Determining input data for a classification model based on the account characteristics corresponding to the node to be identified and the community characteristics; The account type of the node to be identified is determined according to the input data through the classification model.

12. The device according to claim 7, characterized in that The relationship network is an eigenvalue relationship network generated based on eigenvalue transfer, and the fourth determining unit is specifically configured to: The account type of the node to be identified is determined to be a normal account or an abnormal account according to the account characteristics corresponding to the node to be identified and the community characteristics.

13. A computer device, characterized in that: The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is configured to execute the community feature determination method based on a relationship network according to instructions in the program code.

14. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the community feature determination method based on a relationship network according to any one of claims 1 to 6.

Images