Methods, apparatus, and computer equipment for identifying target users based on internet hotspots

Abstract

This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, and computer device for identifying target users based on network hotspots. The method includes: generating a user relationship graph based on network hotspot connection data; obtaining user attribute features corresponding to each user node in the user relationship graph; for each user node in the user relationship graph, sampling neighboring user nodes connected to the target user node according to node correlation priority; weighting and summing the user attribute features corresponding to the sampled neighboring user nodes and the user attribute features corresponding to the target user node, and normalizing the weighted and summed result to obtain neighbor aggregated attribute features; scoring the users corresponding to the target user node based on the neighbor aggregated attribute features, and identifying users whose scores reach a certain threshold as diffused users. This method can improve the accuracy of user diffusion.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, and computer device for identifying target users based on network hotspot connection data. Background Technology

[0002] With the continuous development of wireless networks, the number of network hotspots is also increasing, as are the number of users connecting to them. This process generates network hotspot connection data, and this massive amount of data contains a wealth of user-related information. Therefore, professionals use this network hotspot connection data to identify similar target users for targeted operations.

[0003] Traditional user diffusion schemes typically calculate the similarity between users based on connection frequency, connection time, and location information in network hotspot connection data using distance metrics to identify target users. However, this approach may result in the target users being searched not being the actual users needed. For example, in public hotspot areas, even if connection frequency, connection time, and location information are similar, it may not accurately reflect the similarity between users, thus reducing the accuracy of user diffusion. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, and computer device for identifying target users based on network hotspots that can improve the accuracy of user dissemination, addressing the aforementioned technical problems.

[0005] A method for identifying target users based on network hotspots, the method comprising:

[0006] A user relationship graph is generated based on network hotspot connection data; connected user nodes in the user relationship graph indicate that the corresponding users are connected to the same network hotspot.

[0007] Obtain the user attribute features corresponding to each user node in the user relationship graph;

[0008] For each user node in the user relationship graph, neighboring user nodes connected to the targeted user node are sampled first according to node association.

[0009] The user attribute features corresponding to the sampled neighbor user nodes are weighted and summed with the user attribute features corresponding to the targeted user nodes. The weighted and summed result is then normalized to obtain the neighbor aggregated attribute features.

[0010] Based on the neighbor aggregation attribute features, the users corresponding to the targeted user nodes are scored, and the users whose scores reach the score conditions are regarded as diffused users.

[0011] A target user identification device based on network hotspots, the device comprising:

[0012] The generation module is used to generate a user relationship graph based on network hotspot connection data; the connected user nodes in the user relationship graph indicate that the corresponding users are connected to the same network hotspot;

[0013] The acquisition module is used to acquire the user attribute features corresponding to each user node in the user relationship graph;

[0014] The sampling module is used to sample neighboring user nodes connected to each user node in the user relationship graph according to node association priority.

[0015] The aggregation module is used to weight and sum the user attribute features corresponding to the sampled neighbor user nodes and the user attribute features corresponding to the targeted user nodes, and normalize the weighted and summed result to obtain the neighbor aggregated attribute features.

[0016] The diffusion module is used to score the users corresponding to the targeted user nodes based on the neighbor aggregation attribute features, and to take the users whose scores reach the score conditions as the diffused users.

[0017] In one embodiment, the user relationship graph is an isomorphic graph formed based on the user nodes to reflect user relationships; the generation module is further configured to extract network hotspot identifiers, user identifiers, and network connection relationships from the network hotspot connection data; generate a bipartite graph with the network hotspot identifiers as network nodes and the user identifiers as user nodes, and the network nodes and user nodes are connected according to the network connection relationships; connect user nodes in the bipartite graph that are connected to the same network hotspots, and filter out the network nodes from the bipartite graph that are connected to the user nodes to obtain the isomorphic graph.

[0018] In one embodiment, the aggregation module is further configured to obtain the edge weights between the target user node and each sampled neighboring user node; weight the user attribute features corresponding to each sampled neighboring user node according to the obtained edge weights to obtain weighted neighboring user attribute features; sum the user attribute features corresponding to the target user node and the weighted neighboring user attribute features, and normalize the summed result to obtain the neighbor aggregated attribute features.

[0019] In one embodiment, the edge weight is obtained through an edge weight calculation step; the apparatus further includes:

[0020] The first calculation module is used to determine the frequency of a user connecting to a network hotspot within a preset time period for the user corresponding to the targeted user node; sum the frequencies corresponding to users connecting to the same network hotspot to obtain a sum value; and use the sum value as the edge weight between user nodes corresponding to users connecting to the same network hotspot.

[0021] In one embodiment, the user relationship graph is a homogeneous graph formed based on the user nodes to reflect user relationships; the generation module is further configured to extract user identifiers from the network hotspot connection data; use the extracted user identifiers as the user nodes, and connect target user nodes that connect to the same network hotspot among all the user nodes, thereby forming the homogeneous graph.

[0022] In one embodiment, the user attribute features include numerical attribute features and non-numerical attribute features; the device further includes:

[0023] The processing module is used to perform binning processing on the numerical attribute features to obtain discretized numerical attribute features;

[0024] The encoding module is used to encode the discretized numerical attribute features and the non-numerical attribute features respectively, and combine the encoding results into a feature matrix;

[0025] The aggregation module is further configured to aggregate the feature matrix corresponding to the sampled neighboring user nodes with the feature matrix corresponding to the targeted user node.

[0026] In one embodiment, the aggregation module is further configured to: obtain the attention parameters of the targeted user node, and the attention parameters between the targeted user node and sampled neighboring user nodes; calculate a first product between the user attribute features corresponding to the targeted user node, the corresponding attention parameters, and a preset parameter matrix; calculate a second product between the user attribute features corresponding to the sampled neighboring user nodes connected to the targeted user node, the corresponding attention parameters, and a preset parameter matrix; sum the first product and the second product corresponding to the targeted user node, and normalize the summation result to obtain the neighbor aggregated attribute features.

[0027] In one embodiment, the attention parameter is obtained based on an attention parameter calculation step, and the apparatus further includes:

[0028] The second calculation module is used to calculate the product of the parameter matrix and the user attribute features corresponding to the targeted user node to obtain the first weighted user attribute features; calculate the product of the parameter matrix and the user attribute features corresponding to the sampled neighboring user nodes to obtain the second weighted user attribute features; perform a nonlinear transformation on the first weighted user attribute features and the second weighted user attribute features through an attention parameter network; and normalize the result of the nonlinear transformation to obtain the attention parameters.

[0029] In one embodiment, the diffusion module is further configured to classify the neighbor aggregation attribute features using a classification model; score the users corresponding to the targeted user nodes according to the classification results; and sort the users corresponding to the targeted user nodes according to the scores obtained; and select users whose ranking reaches a preset ranking as diffusion users.

[0030] In one embodiment, the classification model is obtained by processing an initial classification model; the apparatus may further include:

[0031] The training module is used to generate a sample user relationship graph based on network hotspot connection data samples. Connected sample user nodes in the sample user relationship graph indicate that the corresponding users are connected to the same network hotspot. The sample user nodes include seed user nodes and random user nodes. An attribute feature training set is obtained based on each sample user node in the sample user relationship graph. From the attribute feature training set, the attribute feature training data corresponding to each sample user node and its corresponding neighbor user nodes are aggregated to obtain trained aggregated attribute features. The initial classification model is trained using the trained aggregated attribute features. Training stops when the feature similarity between the diffused users predicted by the initial classification model and the seed users reaches a similarity threshold.

[0032] In one embodiment, the training module is further configured to: extract a subset of users from the seed users; use the extracted seed users as positive samples and the extracted subset of users and the random users as negative samples; perform prediction processing on the training aggregated attribute features corresponding to the subset of users using the initial classification model, and determine a classification threshold based on the obtained prediction values; and perform prediction processing on the training aggregated attribute features of the random users using the initial classification model to obtain target negative samples; and train the initial classification model after prediction processing using the training aggregated attribute features corresponding to the positive samples and the target negative samples.

[0033] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program performing the following steps:

[0034] A user relationship graph is generated based on network hotspot connection data; connected user nodes in the user relationship graph indicate that the corresponding users are connected to the same network hotspot.

[0035] Obtain the user attribute features corresponding to each user node in the user relationship graph;

[0036] For each user node in the user relationship graph, the neighboring user nodes connected to the targeted user node are sampled;

[0037] The user attribute features corresponding to the sampled neighbor user nodes are aggregated with the user attribute features corresponding to the targeted user nodes to obtain the neighbor aggregated attribute features.

[0038] Based on the neighbor aggregation attribute features, the users corresponding to the targeted user nodes are classified to determine the diffused users.

[0039] A computer-readable storage medium having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0040] A user relationship graph is generated based on network hotspot connection data; connected user nodes in the user relationship graph indicate that the corresponding users are connected to the same network hotspot.

[0041] Obtain the user attribute features corresponding to each user node in the user relationship graph;

[0042] For each user node in the user relationship graph, the neighboring user nodes connected to the targeted user node are sampled;

[0043] The user attribute features corresponding to the sampled neighbor user nodes are aggregated with the user attribute features corresponding to the targeted user nodes to obtain the neighbor aggregated attribute features.

[0044] Based on the neighbor aggregation attribute features, the users corresponding to the targeted user nodes are classified to determine the diffused users.

[0045] The aforementioned target user identification method, apparatus, computer equipment, and storage medium based on network hotspots organize the network hotspot connection data formed by users connecting to network hotspots into a graph, obtaining a user relationship graph. The existence of connections between user nodes in the user relationship graph expresses the relationships between users, thus avoiding manual intervention in constructing feature information and facilitating the rapid acquisition of user relationships. Furthermore, by aggregating the user attribute features corresponding to neighboring user nodes with those corresponding to the corresponding user nodes, neighbor aggregation attribute features that simultaneously characterize the similarity between network topology and user node features can be obtained. Users corresponding to user nodes are scored based on these neighbor aggregation attribute features, effectively identifying spreading users and improving the accuracy and coverage of targeted user diffusion. Moreover, before aggregation, neighboring user nodes connected to each user node are sampled, and then the user attribute features corresponding to the sampled neighboring user nodes are aggregated with those corresponding to the corresponding user nodes. This allows for the rapid acquisition of neighbor aggregation attribute features characterizing the similarity between network topology and user node features, enabling the rapid and accurate identification of spreading users. Attached Figure Description

[0046] Figure 1 This is an application environment diagram of a target user identification method based on network hotspots in one embodiment;

[0047] Figure 2 This is a flowchart illustrating a target user identification method based on network hotspots in one embodiment;

[0048] Figure 3 This is a schematic diagram of a bipartite graph structure with network hotspot identifiers and user identifiers as nodes in one embodiment;

[0049] Figure 4 This is a schematic diagram of the isomorphic graph with user identifiers as nodes in one embodiment;

[0050] Figure 5 This is a schematic diagram of the structure of an attention parameter network used to generate attention parameters in one embodiment;

[0051] Figure 6 This is a flowchart illustrating a target user identification method based on a network hotspot, using an application scenario of a user connecting to a WiFi hotspot as an example.

[0052] Figure 7 This is a flowchart illustrating the process of processing an initial classification model to obtain a classification model in one embodiment.

[0053] Figure 8 This is a flowchart illustrating a target user identification method based on network hotspots in another embodiment;

[0054] Figure 9 This is a schematic diagram illustrating the statistics of WiFi connection data in one embodiment;

[0055] Figure 10 This is a schematic diagram of the structure of a target user identification device based on network hotspots in one embodiment;

[0056] Figure 11 This is a structural block diagram of a target user identification device based on network hotspots in another embodiment;

[0057] Figure 12 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0059] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0060] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0061] Computer vision (CV) is a science that studies how to enable machines to "see." More specifically, it refers to machine vision, which uses cameras and computers to replace human eyes in recognizing and measuring targets, and then performs image processing to create images more suitable for human observation or transmission to instruments for detection. As a scientific discipline, computer vision studies related theories and technologies, attempting to build artificial intelligence systems capable of extracting information from images or multi-dimensional data. Computer vision technologies typically include image segmentation, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content / behavior recognition, 3D object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping (SLAM), and common biometric recognition technologies such as facial recognition and fingerprint recognition.

[0062] Machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, formulaic learning, and metric learning.

[0063] The solutions provided in this application involve technologies such as computer vision and machine learning in artificial intelligence, and are specifically illustrated through the following embodiments:

[0064] The target user identification method based on network hotspots provided in this application can be applied to, for example... Figure 1The application environment shown includes a terminal 102, a server 104, and a network hotspot 106. Terminal 102 and network hotspot 106 can be connected via a wireless network; terminal 102 and server 104, as well as server 104 and network hotspot 106, can be connected via a wireless network or data cable. Terminal 102 connects to the hotspot network emitted by network hotspot 106 and generates network hotspot connection data during the connection process, such as connection time, location, network hotspot identifier, and user identifier. After obtaining the network hotspot connection data, server 104 can generate a user relationship graph based on the data. Connected user nodes in the user relationship graph indicate that the corresponding users are connected to the same network hotspot. Server 104 obtains the user attribute features corresponding to each user node in the user relationship graph. For each user node in the user relationship graph, it samples the neighboring user nodes connected to the target user node. It aggregates the sampled user attribute features corresponding to the neighboring user nodes with the user attribute features corresponding to the target user node to obtain neighbor aggregated attribute features. Based on the neighbor aggregated attribute features, it classifies the users corresponding to the target user node to determine the spreading users.

[0065] The terminal 102 can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these.

[0066] Server 104 can be a standalone physical server or a server cluster consisting of multiple physical servers. It can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0067] Network hotspot 106 can refer to network devices that provide wireless LAN access to the Internet, such as wireless routers or wireless switches.

[0068] In one embodiment, such as Figure 2 As shown, a method for identifying target users based on network hotspots is provided, and this method is applied to... Figure 1 Taking server 104 as an example, the following steps are included:

[0069] S202, Generate a user relationship graph based on network hotspot connection data; connected user nodes in the user relationship graph indicate that the corresponding users are connected to the same network hotspot.

[0070] A network hotspot refers to a network device that provides wireless LAN access to the Internet. Network hotspot connection data refers to the data generated when a user connects to a network hotspot via a terminal. This connection data may include: connection time, location information, network hotspot identifier, user identifier, and network connection relationship. The connection time may include the start and end times of the connection. The network hotspot identifier is used to distinguish between network hotspots or the wireless LANs they emit; for example, an SSID (Service Set Identifier) ​​can be used to differentiate between different wireless LANs. The user identifier can be a label used to identify the user, such as a username, user ID (e.g., mobile phone number, social media account), or the terminal identifier of the terminal used by the user to connect to the network hotspot.

[0071] A user relationship graph can be defined as a graph that uses user identifiers as nodes and connects the user identifiers of users who have connected to the same network hotspot. The nodes in this user relationship graph are called user nodes.

[0072] When a terminal connects to a network hotspot, the hotspot records the connection time, the user ID corresponding to the terminal, and the location information of the terminal. It also records the hotspot's own identifier. A relationship is established between the user ID of the terminal connecting to the hotspot and the hotspot identifier (this relationship is the network connection relationship), thus obtaining the network hotspot connection data. After completing the information recording, the network hotspot connection data is saved locally or on a server.

[0073] In one embodiment, the server can obtain network hotspot connection data from network hotspots or obtain network hotspot connection data locally on the server. The server can then determine which network hotspot the terminal corresponding to the user identifier is connected to based on the network connection relationships within the network hotspot connection data. The obtained network hotspot connection data can be data generated within a preset time period, such as network hotspot connection data from the past week.

[0074] Specifically, the server can send a data retrieval request to the network hotspot, so that the network hotspot can obtain the corresponding network hotspot connection data according to the received data retrieval request, and then return the network hotspot connection data to the server.

[0075] User relationship graphs can be constructed in the following two ways:

[0076] Method 1: Construct a user relationship graph based on a bipartite graph formed by connecting data from network hotspots.

[0077] In one embodiment, the user relationship graph is a homogeneous graph formed based on user nodes to reflect user relationships; S202 may specifically include: the server extracting network hotspot identifiers, user identifiers, and network connection relationships from network hotspot connection data; generating a bipartite graph with network hotspot identifiers as network nodes and user identifiers as user nodes, and network nodes and user nodes connected according to network connection relationships; connecting user nodes in the bipartite graph that are connected to the same network hotspot, and filtering out network nodes from the bipartite graph that is connected to user nodes to obtain a homogeneous graph.

[0078] This user relationship can be used to reflect whether different users have connected to the same Wi-Fi hotspot through their respective devices. Users connecting to the same Wi-Fi hotspot share some degree of similarity. For example, users connecting to the same home Wi-Fi hotspot may be family members, while users connecting to the same company Wi-Fi hotspot may be colleagues. This user relationship can be used to extract valuable information about users.

[0079] An isomorphic graph can refer to a graph in which all nodes belong to the same category (i.e., all nodes belong to the user type). In other words, an isomorphic graph only contains user nodes and does not contain nodes of other types.

[0080] A bipartite graph is a graph that contains both user nodes and network nodes. This bipartite graph can be transformed into an isomorphic graph.

[0081] It should be noted that in a bipartite graph, each user node can represent a corresponding user or the user's terminal, and each network node represents a corresponding network hotspot. Correspondingly, in a isomorphic graph, each user node can represent a corresponding user or the user's terminal. In subsequent embodiments, if a user node connects to a network hotspot or a network node, it means that the user corresponding to that user node connects to the network hotspot corresponding to that network node through their terminal.

[0082] Specifically, after extracting network hotspot identifiers, user identifiers, and network connection relationships from the network hotspot connection data, the server can use the network hotspot identifier as the network node and the user identifier as the user node. Then, based on the network connection relationship, it determines the network hotspot connected to the user's terminal. Finally, it connects the user node to the network node corresponding to that network hotspot, thus obtaining a bipartite graph. This bipartite graph reveals which network hotspots a user connected to through their terminal within a certain time period, and which users connected to a particular network hotspot within that time period.

[0083] For example, Figure 3This is a bipartite graph based on the network nodes and user nodes connected according to network connectivity. The icons in this bipartite graph are just examples; other icons, such as dots or small circles, can be used. The graph shows which users are connected to which network hotspots. For example, the user corresponding to user node 1 connects to the network hotspot corresponding to network node 1 via a computer; similarly, the user corresponding to user node 2 connects to the network hotspot corresponding to network node 1 via a tablet, and so on. Furthermore, the graph also shows that the same user connects to different network hotspots at different times. For example, the user corresponding to user node 5 connects to the network hotspot corresponding to network node 1 via a tablet, and also connects to the network hotspot corresponding to network node 2.

[0084] In one embodiment, in this bipartite graph, the server can connect at least two user nodes that share the same network hotspot with edges, thereby associating these two user nodes. Furthermore, after connecting all user nodes that share the same network hotspot with edges, the network nodes can be filtered out of the bipartite graph, resulting in an isomorphic graph of the user nodes. It should be noted that filtering out network nodes from the bipartite graph also removes the corresponding edges from the bipartite graph.

[0085] For example, Figure 4 This is to create an isomorphic graph in a bipartite graph by connecting user nodes that share the same network hotspots and filtering out network nodes. Figure 4 It is Figure 3 In a bipartite graph, user nodes with the same network hotspots are connected, and then the network nodes are filtered out to obtain an isomorphic graph, for example, in... Figure 3 In the process, all user nodes 1-5 connected to network node 1 are connected by edges, meaning each user node 1-5 is paired with another user node. Then, network node 1 is removed from the bipartite graph, along with its corresponding edges. This process is repeated until all user nodes connected to the same network hotspot are connected, and the network hotspots and their corresponding edges are removed. The result is as follows: Figure 4 isomorphic graphs. In this Figure 4 In this context, for at least two user nodes that have a connection relationship, the users corresponding to those user nodes have connected to the same network hotspot through their terminals. For example, user node 1 and user nodes 2-5 have all connected to the same network hotspot, i.e., the network hotspot corresponding to network node 1, and so on.

[0086] Method 2: Directly construct a user relationship graph based on network hotspot connection data.

[0087] In one embodiment, the user relationship graph is a homogeneous graph formed based on user nodes to reflect user relationships; S202 may specifically include: the server extracting user identifiers from network hotspot connection data; using the extracted user identifiers as user nodes, and connecting target user nodes that connect to the same network hotspot among all user nodes, thereby forming a homogeneous graph.

[0088] For example, such as Figure 4 As shown, the server extracts all user identifiers from the network hotspot connection data, and then uses the extracted user identifiers as user nodes 1 to 23. It then generates a list of user nodes 1 to 23 connected to the same network hotspot using edges, thus obtaining... Figure 4 The isomorphic diagram shown.

[0089] S204, Obtain the user attribute features corresponding to each user node in the user relationship graph.

[0090] User attribute features can refer to various attribute features used to describe users, including but not limited to users' gender, age, education, hobbies, points of interest (i.e., places of interest), assets, devices, and place of origin.

[0091] In one embodiment, S204 may specifically include: the server can obtain the corresponding user attribute features from the local machine based on the user identifier corresponding to each user node; or, send an attribute feature acquisition request carrying the user identifier to the social server. When the user's authorization is obtained, the social server obtains the user attribute features based on the user identifier and then returns the acquired user attribute features to the server.

[0092] When users register for social application accounts or use social applications, they usually upload their user attribute characteristics. The social server will save the user attribute characteristics uploaded by the user. When the user authorizes the server, the user attribute characteristics can be sent to the server, so that the server can retrieve the corresponding user attribute characteristics from the local machine based on the user identifier.

[0093] In one embodiment, after obtaining the user attribute features corresponding to each user node, the server calculates the edge weights between connected user nodes in the user relationship graph, so as to weight the encoded user attribute features of the corresponding user nodes according to the edge weights, and then executes S206 and S208. Here, the edge weight can refer to the weight of the edge between connected user nodes.

[0094] The calculation steps for edge weights may include: for a user node, the server determines the frequency of the user's connection to the network hotspot within a preset time period; summing the frequencies of users connecting to the same network hotspot to obtain a sum value; and using the sum value as the edge weight between user nodes corresponding to users connecting to the same network hotspot.

[0095] Frequency can be the number of times a user connects to a network hotspot through their terminal within a preset time period.

[0096] For example, such as Figure 3 As shown, for all users corresponding to user nodes 1 to 23, i.e., users 1 to 23, the server determines the frequency of each user's connection to the network hotspot. If user 1 uses terminal 1 to connect to the network hotspot corresponding to network node 1 7 times within a week, and user 2 uses terminal 2 to connect to the network hotspot corresponding to network node 1 2 times within a week, then the edge weight between user node 1 and user node 2 is log_{i=1}^{i- ... t (cntA)+log t (cntB) = log t 7+log t 2, where t is 2, e, or 10, and e is the natural constant, approximately 2.718281828459045. That is, in Figure 4 In the isomorphic graph, the edge weight between user node 1 and user node 2 is log. t 7+log t 2. Similarly, it can be calculated that... Figure 4 The edge weights between user nodes.

[0097] In one embodiment, user attribute features include numerical attribute features and non-numerical attribute features; the method further includes: the server performing binning processing on the numerical attribute features to obtain discretized numerical attribute features; encoding the discretized numerical attribute features and non-numerical attribute features respectively, and combining the encoded results (i.e., the encoded user attribute features) into a feature matrix.

[0098] Numerical attribute features represent user attributes that are numerically related, such as a user's age. Binning involves dividing a continuous segment of numerical attribute features into several segments, with each segment considered a category. For example, a user's age can be divided into different age groups: 0-6, 6-12, 12-18, 18-24, and so on, thus obtaining different age groups.

[0099] In one embodiment, after the server obtains the discretized numerical attribute features, it encodes the discretized numerical and non-numerical attribute features using one-hot encoding. Then, it uses the aforementioned edge weights to weight the encoded results and combines the weighted results into a feature matrix, which is a weighted feature matrix.

[0100] S206, For each user node in the user relationship graph, sample the neighboring user nodes connected to the targeted user node according to the node association priority.

[0101] In this context, node correlation refers to the degree of association or closeness between user nodes in the user relationship graph. A higher degree of association or closeness indicates a stronger node correlation. During sampling, neighboring user nodes connected to the target user node are sampled according to the strength of the node correlation. That is, neighboring user nodes with stronger node correlations are more easily sampled, and conversely, neighboring user nodes with weaker node correlations are more difficult to sample. The strength of node correlation can be affected by the number of edges between nodes, their weights, or the frequency of user nodes connecting to network hotspots. Higher numbers of edges, weights, or frequencies indicate stronger node correlations, and vice versa. A neighboring user node refers to a user node connected to a given user node, such as... Figure 4 As shown, for user node 1, its neighboring user nodes are user nodes 2 to 5.

[0102] In one embodiment, in the user relationship graph, the server samples the neighboring user nodes of each user node. For example, as... Figure 4 As shown, for user nodes 1 to 23 in the user relationship graph, the neighboring user nodes of the user node can be sampled in ascending order of node number. For example, sampling the neighboring user nodes 2 to 5 of user node 1 can yield the sampled neighboring user nodes 3 and 5. Similarly, sampling the neighboring user nodes 1 and 3 to 5 of user node 2 can yield the sampled neighboring user nodes 1 and 5, and so on.

[0103] During sampling, sampling can be performed according to edge weights; user nodes with larger edge weights have a higher probability of being sampled, while those with smaller edge weights have a lower probability. Alternatively, sampling can be performed according to the number of edges a user node has; user nodes with more edges have a higher probability of being sampled, while those with fewer edges have a lower probability. For example... Figure 4As shown, when sampling the neighboring user nodes of user node 1, user node 5 has the highest probability of being sampled because it has the most edges. Here, the number of edges refers to the number of connecting edges between user nodes in the user relationship graph. For example, if two users have both connected to the same two Wi-Fi hotspots, then the number of edges between the two user nodes is 2.

[0104] S208, the user attribute features corresponding to the sampled neighbor user nodes are weighted and summed with the user attribute features corresponding to the target user node, and the weighted and summed result is normalized to obtain the neighbor aggregated attribute features.

[0105] S208 above describes the feature aggregation process. Feature aggregation can be performed based on weights or attention. Therefore, user attribute features can be aggregated using these two methods, as detailed below:

[0106] Method 1: Aggregation based on weights.

[0107] In one embodiment, the server obtains the edge weights between the target user node and each sampled neighboring user node; it then weights the user attribute features corresponding to each sampled neighboring user node according to the obtained edge weights to obtain weighted neighboring user attribute features; finally, it sums the user attribute features corresponding to the target user node with the weighted neighboring user attribute features, and normalizes the summed result into a neighboring aggregated attribute feature. Alternatively, before aggregation, the edge weights can be used to weight the user attribute features corresponding to the target user node before performing the aggregation operation.

[0108] In one embodiment, when user attribute features are not encoded, the server can use one-hot encoding to encode the user attribute features corresponding to the sampled neighbor user nodes, and then weight the encoded user attribute features corresponding to the neighbor user nodes using the corresponding edge weights. Alternatively, the server can also use one-hot encoding to encode the user attribute features corresponding to the user nodes, and then weight the encoded user attribute features corresponding to the user nodes using the corresponding edge weights. Finally, the server aggregates the weighted neighbor user attribute features corresponding to the sampled neighbor user nodes with the weighted user attribute features corresponding to the target user node to obtain the aggregated neighbor attribute features.

[0109] In another embodiment, when the discretized numerical and non-numerical attribute features are encoded to obtain a feature matrix and stored, S208 may specifically include: the server obtaining the feature matrix corresponding to the sampled neighboring user node and the feature matrix corresponding to the target user node from the stored feature matrix, and then aggregating the feature matrix corresponding to the sampled neighboring user node and the feature matrix corresponding to the target user node.

[0110] Method 2: Aggregation based on attention.

[0111] In one embodiment, S208 may specifically include: the server obtaining the attention parameters of the target user node, and the attention parameters between the target user node and the sampled neighboring user nodes; calculating the first product between the user attribute features corresponding to the target user node, the corresponding attention parameters, and the preset parameter matrix; and calculating the second product between the user attribute features corresponding to the sampled neighboring user nodes connected to the target user node, the corresponding attention parameters, and the preset parameter matrix; summing the first product and the second product corresponding to the target user node, and normalizing the summation result to obtain the neighbor aggregated attribute features.

[0112] The above calculation process can be performed using the following aggregation function to obtain the neighbor aggregation attribute features. Specifically, the user attribute features corresponding to the neighbor user nodes sampled by the server and the user attribute features corresponding to the target user node are input into the aggregation function. The aggregation function calculates the first product, the second product, sums the first and second products, and normalizes the summation result to obtain the neighbor aggregation attribute features. The aggregation function is as follows:

[0113]

[0114] i represents the user node being targeted, j represents the sampled neighboring user nodes, and N... i This represents the total number of user nodes in the user relationship graph; a ij This represents the attention parameters between the target user node and its sampled neighboring user nodes, where W is the parameter matrix; when j ≠ i, This represents the user attribute characteristics corresponding to the j-th neighbor user node of the i-th user node; when j = i, This represents the user attribute features corresponding to the i-th user node. σ(.) can be the sigmoid function, or the softplus or softmax function, etc., used to normalize the summation result.

[0115] Before aggregation, the server can use one-hot encoding to encode the user attribute features corresponding to the sampled neighbor user nodes, and then encode the user attribute features corresponding to the user nodes, before performing aggregation.

[0116] Attention parameters are obtained based on the attention parameter calculation steps. In one embodiment, the attention parameter calculation steps may specifically include: the server calculating the product of the parameter matrix and the user attribute features corresponding to the target user node to obtain the first weighted user attribute features; calculating the product of the parameter matrix and the user attribute features corresponding to the sampled neighboring user nodes to obtain the second weighted user attribute features; performing a nonlinear transformation on the first weighted user attribute features and the second weighted user attribute features through an attention parameter network; and normalizing the result of the nonlinear transformation to obtain the attention parameters.

[0117] For example, such as Figure 5 As shown, the first weighted user attribute features Second weighted user attribute features The attention parameters are input into the attention parameter network, and the first weighted user attribute features and the second weighted user attribute features are nonlinearly transformed through the nonlinear transformation layer of the attention parameter network. Then, the result of the nonlinear transformation is normalized by the softmax function to obtain the attention parameter 'a'. ij .

[0118] S210: Based on the neighbor aggregation attribute features, score the users corresponding to the target user nodes, and use the users whose scores reach the score conditions as the diffused users.

[0119] In this context, "diffusion users" refers to target users identified from the users corresponding to user nodes in the user relationship graph and used for message recommendation.

[0120] In one embodiment, S210 may specifically include: the server classifying the neighbor aggregation attribute features using a classification model; scoring the users corresponding to the target user node based on the classification results, and ranking the users corresponding to the target user node according to the scores obtained; and designating users whose ranking reaches a preset rank as diffusion users. The classification model is obtained by training an initial classification model using the user attribute features of seed users and unknown users.

[0121] In one embodiment, the server identifies that the diffusion users and the seed users have similar attribute characteristics. Therefore, the diffusion users and the seed users have similar interests and preferences for items. This allows the server to obtain recommendation information related to the interests or favorite items of the seed users and recommend this information to the diffusion users.

[0122] In one embodiment, the server can also vectorize each node in the bipartite graph using a graph neural network or other graph embedding network to obtain network latent vectors; in addition, it can vectorize the user attribute features of each user node in the isomorphic graph using a graph neural network or other graph embedding network to obtain user latent vectors; the network latent vectors and user latent vectors are input into a binary classification model, so that the binary classification model classifies the users corresponding to each user node in the isomorphic graph based on the network latent vectors and user latent vectors to determine the diffused users.

[0123] To better understand the above-mentioned target user identification method based on online hotspots, combined with... Figure 6 The above-mentioned target user identification method based on network hotspots is described as follows: Figure 6 As shown, this embodiment provides an application scenario where a user connects to a WiFi hotspot. The target user identification method based on the network hotspot in this application scenario includes the following:

[0124] (1) Obtain WiFi hotspot connection data.

[0125] The WiFi hotspot connection data refers to the data generated when a user connects to a WiFi hotspot via their terminal.

[0126] (2) Construct a WiFi user bipartite graph based on the network hotspot connection data.

[0127] The WiFi user bipartite graph is a bipartite graph with user and WiFi hotspot identifiers as nodes.

[0128] (3) Convert the WiFi user bipartite graph into an isomorphic graph with user identifiers as nodes.

[0129] In a homogeneous graph, connected user nodes indicate that the corresponding users are connected to the same network hotspot.

[0130] (4) Collect the user attribute features corresponding to each user node in the isomorphic graph.

[0131] (5) Sample neighboring user nodes for each user node in the isomorphic graph and aggregate the user attribute features of the user node and its corresponding neighboring user nodes.

[0132] (6) Input the neighbor aggregation attribute features into the binary classification model for user classification processing to determine the diffused users.

[0133] (7) Push recommendation information to users.

[0134] The recommended information can be introductions or purchase links for products that the user is interested in.

[0135] WiFi hotspot connection data contains a large amount of user-related information. By adopting the solution of this application, the target user group can be expanded by using users who share WiFi hotspots, and then information can be pushed to the expanded users, thereby achieving precise coverage and improving the targeted coverage rate of users.

[0136] In the above embodiments, the network hotspot connection data formed by users connecting to network hotspots is organized in a graph to obtain a user relationship graph. The existence of connections between user nodes in the user relationship graph expresses the association between users, thus avoiding manual intervention in constructing feature information and facilitating the rapid acquisition of user relationships. Furthermore, the user attribute features corresponding to neighboring user nodes are aggregated with the user attribute features corresponding to the corresponding user node to obtain neighbor aggregated attribute features that simultaneously characterize the similarity between network topology and user node features. Users corresponding to user nodes are scored based on these neighbor aggregated attribute features, effectively identifying spreading users and improving the accuracy and coverage of targeted user diffusion. Moreover, before aggregation, the neighboring user nodes connected to each user node are sampled. Then, the user attribute features corresponding to the sampled neighboring user nodes are aggregated with the user attribute features corresponding to the corresponding user node, quickly obtaining neighbor aggregated attribute features that characterize the similarity between network topology and user node features, thereby enabling rapid and accurate identification of spreading users.

[0137] In one embodiment, the classification model is obtained by processing an initial classification model; such as... Figure 7 As shown, the steps for processing the initial classification model may specifically include:

[0138] S702, Generate a sample user relationship graph based on network hotspot connection data samples; connected sample user nodes in the sample user relationship graph indicate that the corresponding users are connected to the same network hotspot; sample user nodes include seed user nodes and random user nodes.

[0139] The specific process of generating the sample user relationship graph in S702 can be referred to S202 in the above embodiment.

[0140] S704, obtain the attribute feature training set based on each sample user node in the sample user relationship graph.

[0141] The attribute feature training set contains the user attribute features of the users corresponding to each sample user node. The specific process for obtaining user attribute features can be found in S204 of the above embodiment.

[0142] S706, from the attribute feature training set, aggregate the attribute feature training data corresponding to each sample user node and its corresponding neighbor user nodes to obtain the training aggregated attribute features.

[0143] Before aggregation, the corresponding neighboring user nodes can be sampled, and then the sampled user nodes are aggregated with the attribute feature training data corresponding to the sampled neighboring user nodes. The above sampling and aggregation processes can refer to S206 and S208 in the above embodiments.

[0144] S708 trains the initial classification model by training aggregated attribute features. When the feature similarity between the diffusion users and seed users predicted by the initial classification model reaches the similarity threshold, training stops.

[0145] In one embodiment, the server extracts a subset of users from the seed users; uses the extracted seed users as positive samples, and the extracted users and random users as negative samples; performs prediction processing on the training aggregated attribute features corresponding to the extracted users using an initial classification model, and determines a classification threshold based on the obtained prediction values; and performs prediction processing on the training aggregated attribute features of random users using the initial classification model to obtain target negative samples; S708 may specifically include: the server trains the initial classification model after prediction processing using the training aggregated attribute features corresponding to the positive samples and the target negative samples.

[0146] For example, first, randomly sample some users S from seed users P, and add these sampled users to random users U of unknown type. P′=PS is used as a positive sample, and N=U+S is used as a negative sample.

[0147] Secondly, the binary classification model M is initially trained based on the training aggregated attribute features corresponding to positive samples P' and negative samples N. Given that the number of positive samples P' is much smaller than that of negative samples N, the binary classification model M can use the Random Forest algorithm, without needing to normalize the user attribute features of positive samples P' and negative samples N.

[0148] Furthermore, during the initial training process, the classification threshold t is determined based on the predicted value obtained by the binary classification model M for the training aggregated attribute features corresponding to user S, and the reliable target negative sample N' is determined based on the predicted value obtained by the binary classification model M for the training aggregated attribute features corresponding to random user U.

[0149] Finally, a binary classification model is trained based on the training aggregated attribute features corresponding to the positive sample p' and the target negative sample N', thereby obtaining the trained binary classification model, which is then used as the final prediction model.

[0150] In the above embodiments, the attribute feature training data corresponding to each sample user node and its corresponding neighbor user node in the attribute feature training set are aggregated to obtain training aggregated attribute features that are aggregated with seed users and random users. Then, the initial classification model is trained using the training aggregated attribute features. The resulting classification model can classify diffuse users with similar characteristics to seed users. Using this classification model, diffuse users can be quickly and accurately identified.

[0151] As an example, such as Figure 8 As shown, this target user identification method based on network hotspots may include:

[0152] (1) Obtain WiFi connection data and construct a WiFi user bipartite graph based on the WiFi connection data.

[0153] WiFi connection data is collected by reporting from the WiFi Manager backend. Compared to users scanning WiFi data, WiFi connection data collected by reporting from the WiFi Manager backend is more likely to reflect the close connection between users and WiFi hotspots.

[0154] In addition, data within a certain time period (e.g., one week) is filtered out to construct a bipartite graph between users and WiFi hotspots, i.e., a WiFi user bipartite graph, such as... Figure 3 As shown in the diagram, in this WiFi user bipartite graph, the identifier of a WiFi hotspot or user is treated as a node. If a user has a connection record with a WiFi hotspot via their terminal, an edge is drawn between the corresponding user node and the network node. The frequency of user connections to WiFi hotspots within a certain time period is counted, and this frequency is used as the weight of the corresponding edge in the WiFi user bipartite graph. The higher the frequency, the stronger the relationship between the user and the WiFi hotspot.

[0155] like Figure 9 As shown, graphs generated from periods shorter than one week tend to be sparser, with fewer edges between users and insufficient information. Graphs from periods longer than one week, on the other hand, have more edges and may contain some noise. This is because we selected one week of connection data as a specific time period, resulting in a graph density similar to other social networks, and the one-week duration effectively covers user connection behavior.

[0156] (2) Project the WiFi user bipartite graph into an isomorphic graph with user identifiers as nodes.

[0157] When two users connect to the same Wi-Fi hotspot through their respective terminals, an edge is drawn between these two corresponding user nodes in the Wi-Fi user bipartite graph, resulting in the following: Figure 4 The isomorphic diagram shown.

[0158] The method for calculating edge weights in an isomorphic graph is as follows:

[0159] If user A and user B have both established connections with a certain WiFi hotspot 1, and their connection frequencies are cntA and cntB respectively, then there is an edge between the two user nodes corresponding to user A and user B (i.e., user node A and user node B), and the weight of the edge between these two user nodes is w. wifi-1 =log(cntA)+log(cntB).

[0160] By counting all the WiFi hotspots that user A and user B have jointly connected to, and summing their weights, we can obtain the weight w of the edge between user node A and user node B. AB =w wifi-1 +···+w wifi-n .

[0161] (3) Collect the user attribute features corresponding to each user node in the isomorphic graph.

[0162] User attributes include gender (male / female), age (e.g., 0-6, 6-12, 12-18, 18-24, 24-30, 30-35, 35-45, 45-70, 70-100), education level (e.g., primary school, junior high school, high school, undergraduate, associate degree, master's degree, doctoral degree), assets, interests, points of interest (POI), frequently used applications (APP), smart devices used, and place of origin.

[0163] When these user attribute features are numerical, they need to be binned for discretization. Then, all user attribute features are one-hot encoded and stored as a feature matrix.

[0164] (4) Sample neighboring user nodes for each user node in the isomorphic graph.

[0165] Since many users connect to the same WiFi hotspot in the WiFi hotspot connection data, meaning that many user nodes in the isomorphic graph have thousands of neighboring user nodes (such as some public WiFi hotspots with a large number of connected users), in order to balance speed and efficiency, we first randomly sample the neighboring user nodes of each user node in the isomorphic graph, and use the sampling results as the object of the next feature aggregation step.

[0166] (5) Aggregate the user attribute features corresponding to each user node and its neighboring user nodes by using attention.

[0167] The neighbor features of each node are aggregated according to the following formula, where W is the parameter matrix to be learned, hj is the neighbor node features, hi is the original node features, hi' is the aggregated features, and σ is the sigmoid function.

[0168]

[0169] Among them, a ij Let a be the attention parameter between user node i and its neighboring user node j. ij The calculation method is as follows Figure 5 As shown, Figure 5 It is an attention parameter network, that is, a neural network used to calculate attention parameters. Its input is the user attribute features of user node i and its neighboring user nodes j. The input user attribute features are subjected to a non-linear transformation, and then the result of the non-linear transformation is normalized through a softmax layer to obtain the final 'a'. ij , where ∑ j α ij =1.

[0170] Optionally, in order to take into account the attributes of higher-order neighbors, the above aggregation operation will be performed multiple times. The input features for each aggregation are the aggregated attribute features after aggregation in the previous iteration, thereby obtaining the attribute features of higher-order neighbors.

[0171] (6) Train a binary classification model through PU-Learning.

[0172] For each seed package, using seed users as positive samples and random users as unknown samples, a binary classification model is learned. The specific steps are as follows:

[0173] Randomly sample some positive samples S from seed user P and add them to the sample U of unknown type. P' = PS is the positive sample and N = U + S is the negative sample.

[0174] A binary classification model M is trained based on samples P' and N. Since the number of samples P' is much smaller than that of samples N, the binary classification model can use the Random Forest algorithm without the need for feature normalization.

[0175] The threshold t is determined based on the predicted value of sample S by the binary classification model M, and the reliable negative sample N' is determined based on the predicted value of sample U by the binary classification model M.

[0176] A binary classification model M' is trained based on samples P' and N', and the trained binary classification model M' is used as the final WiFi recommendation model.

[0177] (7) Classify the neighbor aggregation attribute features using the trained binary classification model to identify the diffused users.

[0178] The binary classification model produced in step (6) scores and sorts the users corresponding to each user node in the isomorphic graph. The top candidate users sorted here will have similar attributes and WiFi connection behavior to the seed users, thereby achieving the purpose of spreading users.

[0179] The solutions described in the above embodiments can have the following beneficial effects:

[0180] This invention is applied to Tencent WiFi Manager product, which accurately calculates the similarity between users, enabling us to more effectively expand and recall user groups, improve the user coverage and accuracy of audience targeting, and enhance the effectiveness of advertising.

[0181] Furthermore, by using the obtained user latent vectors and WiFi latent vectors as inputs to the binary classification model, the high dimensionality and data sparsity problems faced by recommendation systems can be overcome, which is beneficial to improving recommendation performance.

[0182] It should be noted that the user information (including but not limited to user attribute characteristics, such as user personal information) and data (including but not limited to data used for analysis, data stored, data displayed) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0183] It should be understood that, although Figure 2 , 7 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 2 , 7 At least some of the steps in the process may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the steps or stages in other steps.

[0184] In one embodiment, such as Figure 10 As shown, a target user identification device based on network hotspots is provided. This device can be a software module, a hardware module, or a combination of both integrated into a computer device. Specifically, the device includes: a generation module 1002, an acquisition module 1004, a sampling module 1006, an aggregation module 1008, and a diffusion module 1010, wherein:

[0185] The generation module 1002 is used to generate a user relationship graph based on network hotspot connection data; connected user nodes in the user relationship graph indicate that the corresponding users are connected to the same network hotspot.

[0186] Module 1004 is used to obtain the user attribute features corresponding to each user node in the user relationship graph.

[0187] The sampling module 1006 is used to sample neighboring user nodes connected to each user node in the user relationship graph according to node correlation.

[0188] The aggregation module 1008 is used to weight and sum the user attribute features corresponding to the sampled neighbor user nodes and the user attribute features corresponding to the target user node, and normalize the weighted and summed result to obtain the neighbor aggregated attribute features.

[0189] The diffusion module 1010 is used to score the users corresponding to the target user node based on the neighbor aggregation attribute features, and to take the users whose scores reach the score conditions as the diffused users.

[0190] In one embodiment, the user relationship graph is an isomorphic graph formed based on user nodes to reflect user relationships; the generation module 1002 is further used to extract network hotspot identifiers, user identifiers, and network connection relationships from network hotspot connection data; generate a bipartite graph with network hotspot identifiers as network nodes and user identifiers as user nodes, and network nodes and user nodes are connected according to network connection relationships; connect user nodes in the bipartite graph that are connected to the same network hotspot, and filter out network nodes from the bipartite graph that is connected to user nodes to obtain an isomorphic graph.

[0191] In one embodiment, the aggregation module 1008 is further configured to obtain the edge weights between the target user node and each sampled neighbor user node; weight the user attribute features corresponding to each sampled neighbor user node according to the obtained edge weights to obtain the weighted neighbor user attribute features; sum the user attribute features corresponding to the target user node and the weighted neighbor user attribute features, and normalize the summed result to obtain the neighbor aggregated attribute features.

[0192] In one embodiment, the edge weight is obtained through an edge weight calculation step; such as Figure 11 As shown, the device also includes:

[0193] The first calculation module 1012 is used to determine the frequency of a user connecting to a network hotspot within a preset time period for a user corresponding to a user node; sum the frequencies of users connecting to the same network hotspot to obtain a sum value; and use the sum value as the edge weight between user nodes corresponding to users connecting to the same network hotspot.

[0194] In one embodiment, the user relationship graph is a homogeneous graph formed based on user nodes to reflect user relationships; the generation module 1002 is also used to extract user identifiers from network hotspot connection data; using the extracted user identifiers as user nodes, and connecting target user nodes that connect to the same network hotspot among all user nodes, thereby forming a homogeneous graph.

[0195] In one embodiment, user attribute features include numerical attribute features and non-numerical attribute features; such as Figure 11 As shown, the device also includes:

[0196] Processing module 1014 is used to perform binning processing on numerical attribute features to obtain discretized numerical attribute features;

[0197] The encoding module 1016 is used to encode the discretized numerical attribute features and non-numerical attribute features respectively, and combine the encoding results into a feature matrix;

[0198] The aggregation module 1008 is also used to aggregate the feature matrix corresponding to the sampled neighboring user nodes with the feature matrix corresponding to the target user node.

[0199] In one embodiment, the aggregation module 1008 is further configured to obtain the attention parameters of the target user node and the attention parameters between the target user node and the sampled neighboring user nodes; calculate the first product value between the user attribute features corresponding to the target user node, the corresponding attention parameters, and the preset parameter matrix; and calculate the second product value between the user attribute features corresponding to the sampled neighboring user nodes connected to the target user node, the corresponding attention parameters, and the preset parameter matrix; sum the first product value and the second product value corresponding to the target user node, and normalize the summation result to obtain the neighbor aggregated attribute features.

[0200] In one embodiment, the attention parameters are obtained based on the attention parameter calculation steps; such as... Figure 11 As shown, the device also includes:

[0201] The second calculation module 1018 is used to calculate the product of the parameter matrix and the user attribute features corresponding to the target user node to obtain the first weighted user attribute features; calculate the product of the parameter matrix and the user attribute features corresponding to the sampled neighbor user nodes to obtain the second weighted user attribute features; perform a nonlinear transformation on the first weighted user attribute features and the second weighted user attribute features through an attention parameter network; and normalize the result of the nonlinear transformation to obtain the attention parameters.

[0202] In one embodiment, the diffusion module 1010 is further configured to classify the neighbor aggregation attribute features using a classification model; score the users corresponding to the target user node according to the classification results; and sort the users corresponding to the target user node according to the scores obtained; and select users whose ranking reaches a preset ranking as diffusion users.

[0203] In the above embodiments, the network hotspot connection data formed by users connecting to network hotspots is organized in a graph to obtain a user relationship graph. The existence of connections between user nodes in the user relationship graph expresses the association between users, thus avoiding manual intervention in constructing feature information and facilitating the rapid acquisition of user relationships. Furthermore, the user attribute features corresponding to neighboring user nodes are aggregated with the user attribute features corresponding to the corresponding user node to obtain neighbor aggregated attribute features that simultaneously characterize the similarity between network topology and user node features. Users corresponding to user nodes are scored based on these neighbor aggregated attribute features, effectively identifying spreading users and improving the accuracy and coverage of targeted user diffusion. Moreover, before aggregation, the neighboring user nodes connected to each user node are sampled. Then, the user attribute features corresponding to the sampled neighboring user nodes are aggregated with the user attribute features corresponding to the corresponding user node, quickly obtaining neighbor aggregated attribute features that characterize the similarity between network topology and user node features, thereby enabling rapid and accurate identification of spreading users.

[0204] In one embodiment, the classification model is obtained by processing an initial classification model; such as... Figure 11 As shown, the device may further include:

[0205] Training module 1020 is used to generate a sample user relationship graph based on network hotspot connection data samples. Connected sample user nodes in the sample user relationship graph indicate that the corresponding users are connected to the same network hotspot. The sample user nodes include seed user nodes and random user nodes. An attribute feature training set is obtained based on each sample user node in the sample user relationship graph. From the attribute feature training set, the attribute feature training data corresponding to each sample user node and its corresponding neighbor user nodes are aggregated to obtain the training aggregated attribute features. The initial classification model is trained by training the aggregated attribute features. Training stops when the feature similarity between the diffused users predicted by the initial classification model and the seed users reaches a similarity threshold.

[0206] In one embodiment, the training module is further configured to: extract a subset of users from the seed users; use the extracted seed users as positive samples and the extracted users and random users as negative samples; perform prediction processing on the training aggregated attribute features corresponding to the extracted users using an initial classification model, and determine a classification threshold based on the obtained prediction values; and perform prediction processing on the training aggregated attribute features of random users using the initial classification model to obtain target negative samples; and train the initial classification model after prediction processing using the training aggregated attribute features corresponding to the positive samples and the target negative samples.

[0207] In the above embodiments, the attribute feature training data corresponding to each sample user node and its corresponding neighbor user node in the attribute feature training set are aggregated to obtain training aggregated attribute features that are aggregated with seed users and random users. Then, the initial classification model is trained using the training aggregated attribute features. The resulting classification model can classify diffuse users with similar characteristics to seed users. Using this classification model, diffuse users can be quickly and accurately identified.

[0208] Specific limitations regarding the target user identification device based on network hotspots can be found in the limitations of the target user identification method based on network hotspots mentioned above, and will not be repeated here. Each module in the aforementioned target user identification device based on network hotspots can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0209] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 12As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores user attribute characteristics. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a target user identification method based on network hotspots.

[0210] Those skilled in the art will understand that Figure 12 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0211] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0212] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0213] In one embodiment, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the computer device to perform the steps in the above method embodiments.

[0214] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0215] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0216] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for identifying target users based on network hotspots, characterized in that, The method includes: A user relationship graph is generated based on network hotspot connection data; the connected user nodes in the user relationship graph indicate that the corresponding users are connected to the same network hotspot; the user relationship graph only contains user nodes and does not contain other types of nodes. Obtain the user attribute features corresponding to each user node in the user relationship graph; For each user node in the user relationship graph, neighboring user nodes connected to the target user node are sampled first according to node correlation. The node correlation refers to the correlation or closeness between user nodes in the user relationship graph. The strength of the node correlation is affected by at least one of the following factors: the number of edges of the user node, the edge weight, or the frequency of the user node connecting to network hotspots. The attention parameters of the targeted user node and the attention parameters between the targeted user node and the sampled neighboring user nodes are obtained. The attention parameters are obtained by normalizing the result of the nonlinear transformation of the first weighted user attribute features and the second weighted user attribute features. The first weighted user attribute features are obtained by multiplying the parameter matrix with the user attribute features corresponding to the targeted user node. The second weighted user attribute features are obtained by multiplying the parameter matrix with the user attribute features corresponding to the sampled neighboring user nodes. For each sampled neighbor user node, when the sampled neighbor user node is the target user node, calculate the first product value between the user attribute features corresponding to the target user node, the corresponding attention parameters, and the preset parameter matrix; when the sampled neighbor user node is not the target user node, calculate the second product value between the user attribute features corresponding to the sampled neighbor user node, the corresponding attention parameters, and the preset parameter matrix. The first product value and each of the obtained second products are summed, and the summation result is normalized to obtain the neighbor aggregation attribute feature; Based on the neighbor aggregation attribute features, the users corresponding to the targeted user nodes are scored, and the users whose scores reach the score conditions are taken as the diffused users. Recommendation information is pushed to the aforementioned users.

2. The method according to claim 1, characterized in that, The user relationship graph is an isomorphic graph formed based on the user nodes to reflect user relationships; the generation of the user relationship graph with user identifiers as user nodes based on network hotspot connection data includes: Extract network hotspot identifiers, user identifiers, and network connection relationships from the network hotspot connection data; Generate a bipartite graph with the network hotspot identifier as the network node and the user identifier as the user node, and the network nodes and the user nodes are connected according to the network connection relationship; The user nodes that are connected to the same network hotspot in the bipartite graph are connected, and the network nodes are filtered out from the bipartite graph that connects the user nodes to obtain the isomorphic graph.

3. The method according to claim 1, characterized in that, The process of obtaining the neighbor aggregation attribute features also includes: Obtain the edge weights between the target user node and each sampled neighboring user node; The user attribute features corresponding to each sampled neighbor user node are weighted according to the obtained edge weights to obtain the weighted neighbor user attribute features. The user attribute features corresponding to the target user node are summed with the weighted neighbor user attribute features, and the summed result is normalized to obtain the neighbor aggregated attribute features.

4. The method according to claim 3, characterized in that, The edge weights are obtained through an edge weight calculation step, which includes: For the user corresponding to the target user node, determine the frequency with which the user connects to the network hotspot within a preset time period; The sum of the frequencies corresponding to users connecting to the same network hotspot is obtained. The sum is used as the edge weight between user nodes corresponding to users who are connected to the same network hotspot.

5. The method according to claim 1, characterized in that, The user relationship graph is an isomorphic graph formed based on the user nodes to reflect user relationships; the generation of the user relationship graph with user identifiers as user nodes based on network hotspot connection data includes: Extract user identifiers from the network hotspot connection data; The extracted user identifier is used as the user node, and target user nodes that are connected to the same network hotspot are connected among all the user nodes to form the isomorphic graph.

6. The method according to claim 1, characterized in that, The user attribute features include numerical attribute features and non-numerical attribute features; the method further includes: The numerical attribute features are binned to obtain discretized numerical attribute features; The discretized numerical attribute features and the non-numerical attribute features are encoded respectively, and the encoding results are combined into a feature matrix; The step of aggregating the user attribute features corresponding to the sampled neighbor user nodes with the user attribute features corresponding to the targeted user node includes: The feature matrices corresponding to the sampled neighboring user nodes are aggregated with the feature matrices corresponding to the targeted user nodes.

7. The method according to claim 1, characterized in that, The step of obtaining the user attribute features corresponding to each user node in the user relationship graph includes: Send an attribute feature retrieval request carrying a user identifier to the social server, so that the social server can obtain the user's attribute features based on the user identifier when it obtains the user's authorization; Receive the user attribute characteristics fed back by the social server.

8. The method according to any one of claims 1 to 7, characterized in that, The step of scoring users corresponding to the targeted user nodes based on the neighbor aggregation attribute features, and designating users whose scores reach a certain threshold as diffused users, includes: The neighbor aggregation attribute features are classified using a classification model; Based on the results of the classification process, the users corresponding to the targeted user nodes are scored, and the users corresponding to the targeted user nodes are sorted according to the scores obtained. Users who reach the preset ranking will be considered as spreading users.

9. The method according to claim 8, characterized in that, The classification model is obtained by processing the initial classification model; the processing of the initial classification model includes: A sample user relationship graph is generated based on network hotspot connection data samples; the connected sample user nodes in the sample user relationship graph indicate that the corresponding users are connected to the same network hotspot; the sample user nodes include seed user nodes and random user nodes; Obtain the attribute feature training set based on each of the sample user nodes in the sample user relationship graph; From the attribute feature training set, the attribute feature training data corresponding to each sample user node and its corresponding neighbor user node are aggregated to obtain the training aggregated attribute features; The initial classification model is trained using the aggregated attribute features. Training stops when the feature similarity between the diffusion users predicted by the initial classification model and the seed users reaches a similarity threshold.

10. The method according to claim 9, characterized in that, The method further includes: Remove a subset of users from the seed users; The seed users after extracting the aforementioned portion of users are used as positive samples, and the extracted portion of users and the random users are used as negative samples; The training aggregated attribute features corresponding to the positive samples and the negative samples are respectively input into the initial classification model for preliminary training, so as to predict the training aggregated attribute features corresponding to the partial users through the initial classification model, and determine the classification threshold based on the obtained prediction values; and, the training aggregated attribute features of the random users are predicted through the initial classification model to obtain the target negative samples. The step of training the initial classification model using the aggregated attribute features includes: The initial classification model, after prediction processing, is trained by using the training aggregated attribute features corresponding to the positive samples and the target negative samples.

11. A target user identification device based on network hotspots, characterized in that, The device includes: The generation module is used to generate a user relationship graph based on network hotspot connection data; the connected user nodes in the user relationship graph indicate that the corresponding users are connected to the same network hotspot, and the user relationship graph only contains user nodes and does not contain other types of nodes; The acquisition module is used to acquire the user attribute features corresponding to each user node in the user relationship graph; The sampling module is used to sample neighboring user nodes connected to each user node in the user relationship graph according to node correlation. The node correlation refers to the degree of correlation or closeness between user nodes in the user relationship graph. The strength of the node correlation is affected by at least one of the following factors: the number of edges of the user node, the edge weight, or the frequency of the user node connecting to network hotspots. An aggregation module is used to obtain the attention parameters of the targeted user node and the attention parameters between the targeted user node and sampled neighboring user nodes. The attention parameters are obtained by performing a nonlinear transformation on a first weighted user attribute feature and a second weighted user attribute feature, followed by normalization of the transformation result. The first weighted user attribute feature is obtained by multiplying the parameter matrix with the user attribute feature corresponding to the targeted user node, and the second weighted user attribute feature is obtained by multiplying the parameter matrix with the user attribute feature corresponding to the sampled neighboring user nodes. For each sampled neighboring user node, when the sampled neighboring user node is the targeted user node, a first product value is calculated between the user attribute feature corresponding to the targeted user node, the corresponding attention parameters, and a preset parameter matrix. When the sampled neighboring user node is not the targeted user node, a second product value is calculated between the user attribute feature corresponding to the sampled neighboring user node, the corresponding attention parameters, and a preset parameter matrix. The first product value and each of the obtained second product values ​​are summed, and the summation result is normalized to obtain the neighbor aggregated attribute features. The diffusion module is used to score the users corresponding to the target user nodes based on the neighbor aggregation attribute features, and to take the users whose scores reach the score conditions as the diffused users. The module is used to perform the following steps: push recommendation information to the disseminating users.

12. The apparatus according to claim 11, characterized in that, The user relationship graph is an isomorphic graph formed based on the user nodes to reflect user relationships; the generation module is also used to extract network hotspot identifiers, user identifiers and network connection relationships from the network hotspot connection data; and generate a bipartite graph with the network hotspot identifiers as network nodes and the user identifiers as user nodes, and the network nodes and user nodes are connected according to the network connection relationships; The user nodes that are connected to the same network hotspot in the bipartite graph are connected, and the network nodes are filtered out from the bipartite graph that connects the user nodes to obtain the isomorphic graph.

13. The apparatus according to claim 11, characterized in that, The aggregation module is further configured to obtain the edge weights between the target user node and each sampled neighboring user node; to weight the user attribute features corresponding to each sampled neighboring user node according to the obtained edge weights, thereby obtaining the weighted neighboring user attribute features; to sum the user attribute features corresponding to the target user node and the weighted neighboring user attribute features, and to normalize the summation result to obtain the neighbor aggregated attribute features.

14. The apparatus according to claim 13, characterized in that, The edge weights are obtained through an edge weight calculation step, and the device further includes: The first calculation module is used to determine the frequency of a user connecting to a network hotspot within a preset time period for the user corresponding to the targeted user node; sum the frequencies corresponding to users connecting to the same network hotspot to obtain a sum value; and use the sum value as the edge weight between user nodes corresponding to users connecting to the same network hotspot.

15. The apparatus according to claim 11, characterized in that, The user relationship graph is an isomorphic graph formed based on the user nodes to reflect user relationships; The generation module is further configured to extract user identifiers from the network hotspot connection data; use the extracted user identifiers as user nodes, and connect target user nodes that connect to the same network hotspot among all the user nodes to form the isomorphic graph.

16. The apparatus according to claim 11, characterized in that, The user attribute features include numerical attribute features and non-numerical attribute features; the device further includes: The processing module is used to perform binning processing on the numerical attribute features to obtain discretized numerical attribute features; The encoding module is used to encode the discretized numerical attribute features and the non-numerical attribute features respectively, and combine the encoding results into a feature matrix; The aggregation module is further configured to aggregate the feature matrix corresponding to the sampled neighboring user nodes with the feature matrix corresponding to the targeted user node.

17. The apparatus according to claim 11, characterized in that, The acquisition module is further configured to send an attribute feature acquisition request carrying a user identifier to the social server, so that the social server can acquire user attribute features based on the user identifier when it obtains the user's authorization; and receive the user attribute features fed back by the social server.

18. The apparatus according to any one of claims 11 to 17, characterized in that, The diffusion module is further configured to classify the neighbor aggregation attribute features using a classification model; score the users corresponding to the targeted user nodes based on the classification results; and sort the users corresponding to the targeted user nodes according to the scores obtained; and select users whose sorting ranking reaches a preset ranking as diffusion users.

19. The apparatus according to claim 18, characterized in that, The classification model is obtained by processing the initial classification model; the device further includes: The training module is used to generate a sample user relationship graph based on network hotspot connection data samples. Connected sample user nodes in the sample user relationship graph indicate that the corresponding users are connected to the same network hotspot. The sample user nodes include seed user nodes and random user nodes. An attribute feature training set is obtained based on each sample user node in the sample user relationship graph. From the attribute feature training set, the attribute feature training data corresponding to each sample user node and its corresponding neighbor user nodes are aggregated to obtain trained aggregated attribute features. The initial classification model is trained using the trained aggregated attribute features. Training stops when the feature similarity between the diffused users predicted by the initial classification model and the seed users reaches a similarity threshold.

20. The apparatus according to claim 19, characterized in that, The training module is further configured to extract a subset of users from the seed users; use the extracted seed users as positive samples, and use the extracted subset of users and the random users as negative samples; The training aggregated attribute features corresponding to the positive samples and the negative samples are respectively input into the initial classification model for preliminary training, so as to predict the training aggregated attribute features corresponding to the partial users through the initial classification model, and determine the classification threshold based on the obtained prediction values; and, the training aggregated attribute features of the random users are predicted through the initial classification model to obtain target negative samples; the initial classification model after prediction processing is trained using the training aggregated attribute features corresponding to the positive samples and the target negative samples.

21. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 10.

22. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 10.

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

Images