Method, device and equipment for carrying and transferring users in different networks and storage medium

Abstract

The embodiment of the invention provides a carrying method and device for different network users, equipment and a storage medium. The method comprises the steps that a network relation attribute graph is constructed based on communication social event data, obtained in a single period, of home network users; the nodes in the network relation attribute graph comprise a node corresponding to the home network user and nodes corresponding to other users who have communication social events with the home network user; learning the network relation attribute graph based on a graph attention network (GAT) model to obtain feature data of each node in the network relation attribute graph; obtaining feature data of each node in more than two periods based on a multi-head attention mechanism; determining different network family members of the home network user based on the feature data of each node in the more than two periods; and determining a marketing scheme of the different network family members based on the business data of the home network user.

Description

Technical Field

[0001] The present application relates to the field of communication technologies, and in particular, to a method, apparatus, device, and storage medium for carrying and transferring users of different networks. Background Art

[0002] With the increasingly fierce competition in the telecommunications market, the existing users will show mobility, and "carrying phone numbers to transfer networks" will become the norm. Major operators have also entered the offensive and defensive battles of existing user operations and incoming marketing, which requires corresponding identification and analysis of this type of users (incoming users). Therefore, the mining and reaching of incoming users have become urgent problems to be solved in incoming marketing.

[0003] Currently, since methods such as incoming marketing have been gradually cancelled, the way to reach users of different networks is mainly achieved through methods such as business halls and community stalls. By users consulting questions and helping users query services, different network users are discovered for incoming marketing or other business marketing. However, these methods are difficult to fully cover different network users and cannot effectively reach for marketing; on the other hand, the proportion of passive marketing (waiting for different network users to enter the hall) is very low and cannot bring effective user growth. Summary of the Invention

[0004] In view of this, embodiments of the present application are expected to provide a method, apparatus, device, and storage medium for carrying and transferring users of different networks.

[0005] The technical solution of the embodiments of the present application is implemented as follows:

[0006] The embodiments of the present application provide a method for carrying and transferring users of different networks, the method including:

[0007] Constructing a network relationship attribute graph based on the communication and social event data of the home network users obtained within a single period; the nodes in the network relationship attribute graph include: the nodes corresponding to the home network users and the nodes corresponding to other users who have communication and social events with the home network users;

[0008] Learning the network relationship attribute graph based on the Graph Attention Network (GAT) model to obtain the feature data of each node in the network relationship attribute graph;

[0009] Obtaining the feature data of each node within two or more periods based on the multi-head attention mechanism;

[0010] Determining the different network family members of the home network users based on the feature data of each node within the two or more periods;

[0011] Determining a marketing plan for the different network family members based on the service data of the home network users.

[0012] Among them, the communication and social event data includes: communication and social event data between at least two users; the at least two users include: at least one user within the network and at least one user outside the network;

[0013] The communication and social event data further includes: the associated weight between the user within the network and the other users pre-configured.

[0014] Among them, the feature data of each node at least includes:

[0015] The association relationship between each node and other nodes; the association relationship between each node and other nodes includes: the attention score between each node and other nodes; among them,

[0016] A communication and social event occurs between the user corresponding to each node and the user corresponding to other nodes.

[0017] Among them, the learning of the network relationship attribute graph based on the Graph Attention Network (GAT) model to obtain the feature data of each node in the network relationship attribute graph includes:

[0018] Project the initial feature data of each node into a preset representation space based on a weight matrix in the network relationship attribute graph;

[0019] Calculate the corresponding attention score for each pair of neighbor nodes in the network relationship attribute graph; among them, the two nodes in a pair of neighbor nodes respectively correspond to two users who have a communication and social event; the attention score is used to represent the importance of one node to another node;

[0020] Normalize all the attention scores to obtain the normalized attention scores;

[0021] Use the normalized attention scores to perform weighted summation on the feature data of the neighbor nodes of each node, and the new node feature data obtained is the feature data of each node in the network relationship attribute graph.

[0022] Among them, the obtaining of the feature data of each node in more than two periods based on the multi-head attention mechanism includes:

[0023] For a single node, assume that a graph attention layer has K independent heads, and perform independent attention calculations for each head to obtain K new node feature sets; among them, all the node feature data in each period corresponds to a graph attention layer;

[0024] In the first n - 1 graph attention layers, use the averaging method to integrate the outputs of (n - 1)*k heads; the n represents the number of periods;

[0025] In the nth graph attention layer, according to the event decay characteristics and combined with the intensity temporal decay function, the outputs of the K heads are spliced in a splicing manner to obtain the final node features of the single node.

[0026] The determining of the different network family members of the local network user based on the characteristic data of each node in the two or more periods includes:

[0027] Select a user from another network as the center point from the users corresponding to each node in the network relationship attribute graph;

[0028] Based on the attention scores between the node corresponding to the alien network user and different neighboring nodes, a preset number of intimate relationship users with the highest intimacy with the alien network user are screened out from the neighboring users of the alien network user;

[0029] Calculating the similarity of the nighttime permanent locations of the alien network user and the intimate relationship user based on the permanent location information, and if the similarity of the nighttime permanent locations of the two is high, determining that the intimate relationship user is a family member of the alien network user;

[0030] A local network user is selected from the family members of the remote network user, and the remote network user is determined to be a member of the local network user's family.

[0031] Wherein, determining the marketing plan for the family members of the different network based on the service data of the users of the local network includes:

[0032] Analyze the business data of the users of the network;

[0033] Determine a door-to-door marketing plan for the off-network family members of the local network user based on the service data;

[0034] The business data includes one or more of the following:

[0035] Broadband quality data;

[0036] Broadband complaint data;

[0037] Smart home business subscription data;

[0038] Networking service data.

[0039] The embodiment of the present application also provides a device for transferring users from different networks, the device comprising:

[0040] The first processing unit is used to construct a network relationship attribute graph based on the communication and social event data of the users of this network acquired in a single cycle; the nodes in the network relationship attribute graph include: nodes corresponding to the users of this network and nodes corresponding to other users with whom the users of this network have communication and social events; the network relationship attribute graph is learned based on the graph attention network GAT model to obtain feature data of each node in the network relationship attribute graph;

[0041] The second processing unit is used to obtain the feature data of each node in more than two cycles based on the multi-head attention mechanism; determine the off-network family members of the local network user based on the feature data of each node in more than two cycles; and determine the marketing plan for the off-network family members based on the business data of the local network user.

[0042] The embodiment of the present application also provides a device for transferring users from different networks, including: a first communication interface and a first processor; wherein:

[0043] The first processor is used to construct a network relationship attribute graph based on the communication and social event data of the users of the network acquired in a single cycle; the nodes in the network relationship attribute graph include: nodes corresponding to the users of the network and nodes corresponding to other users with whom the users of the network have communication and social events; the network relationship attribute graph is learned based on the graph attention network GAT model to obtain feature data of each node in the network relationship attribute graph;

[0044] Based on the multi-head attention mechanism, characteristic data of each node in more than two cycles are obtained; based on the characteristic data of each node in more than two cycles, the off-network family members of the local network user are determined; based on the business data of the local network user, the marketing plan for the off-network family members is determined.

[0045] The embodiment of the present application also provides a device for transferring users from one network to another, comprising: a first processor and a first memory for storing a computer program that can be run on the processor,

[0046] Wherein, the first processor is used to execute the steps of the above method when running the computer program.

[0047] An embodiment of the present application further provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0048] The method, device, equipment, and storage medium for migrating users between different networks provided by the embodiments of the present application construct a network relationship attribute graph based on the communication and social event data of users on the home network obtained within a single cycle; the nodes in the network relationship attribute graph include: the nodes corresponding to the home network users and the nodes corresponding to other users who have communication and social events with the home network users; learning the network relationship attribute graph based on the Graph Attention Networks (GAT) model to obtain the feature data of each node in the network relationship attribute graph; obtaining the feature data of each node within two or more cycles based on the multi-head attention mechanism; determining the family members of different networks of the home network users based on the feature data of each node within the two or more cycles; and determining the marketing plan for the family members of different networks based on the service data of the home network users. The embodiments of the present application automatically determine the home network users' family members of different networks through the learning of the network relationship attribute graph using the GAT model and the application of the multi-head attention mechanism in multiple cycles, so as to realize the marketing and mining of users of different networks through family relationships, greatly increasing the amount of user data of different networks obtained, making the marketing method more effective, and realizing the proactive marketing method of door-to-door marketing for the home network users' family members of different networks by analyzing the service data of the home network users, ultimately effectively increasing the number of migrated users. Description of the Drawings

[0049] Figure 1 Schematic flowchart of the method for migrating users between different networks according to the embodiments of the present application;

[0050] Figure 2 Schematic diagram of the user communication and social event according to the embodiments of the present application;

[0051] Figure 3 Schematic structural diagram of the device for migrating users between different networks according to the embodiments of the present application;

[0052] Figure 4 Schematic structural diagram of the equipment for migrating users between different networks according to the embodiments of the present application. Detailed Embodiments

[0053] The present application will be described below with reference to the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related invention, rather than limiting the invention. Additionally, it should be noted that for the sake of description, only parts related to the invention are shown in the drawings. It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The present application will be described in detail below with reference to the drawings and embodiments.

[0054] The embodiments of the present application provide a method for migrating users between different networks, as Figure 1 shown, the method includes:

[0055] Step 101: Construct a network relationship attribute graph based on the communication and social event data of the users on the local network obtained within a single period; the nodes in the network relationship attribute graph include: the nodes corresponding to the users on the local network and the nodes corresponding to other users who have communication and social events with the users on the local network;

[0056] Step 102: Learn the network relationship attribute graph based on the GAT model to obtain the feature data of each node in the network relationship attribute graph;

[0057] Step 103: Obtain the feature data of each node within two or more periods based on the multi-head attention mechanism;

[0058] Step 104: Determine the off-network family members of the users on the local network based on the feature data of each node within the two or more periods;

[0059] Step 105: Determine the marketing plan for the off-network family members based on the service data of the users on the local network.

[0060] In the embodiments of the present application, the single period may include: one year, one quarter, one month, etc.

[0061] In the embodiments of the present application, the communication and social event data includes: the communication and social event data between at least two users; the at least two users include: at least one user on the local network and at least one off-network user;

[0062] The communication and social event data further includes: the pre-configured association weight between the users on the local network and the other users.

[0063] In one embodiment, the communication and social event may include one or more of the following: telephone communication, short message communication, 5G message communication.

[0064] In practical applications, as Figure 2 shown, users h2 - h7 are the users who have had communication and social events with user h1 (a user on the local network). The arrow represents the communication initiative and passivity direction. At the same time, the communication frequency and communication events can be assigned different weights. For example: from the perspective of user h1, the connection weight between user h1 and user h2 = the number of active text messages * the active text message weight + the number of passive text messages * the passive text message weight + the number of incoming calls * the incoming call weight + the number of outgoing calls * the outgoing call weight. Currently, short message communication is usually an expression of an external relationship, and the short message method of an intimate relationship has been replaced by social APPs. Therefore, the short message weight can be measured by a negative weight to measure the degree of intimacy.

[0065] In actual applications, it is also necessary to obtain the location data of users. Each user selects the top 5 data of the resident base stations as the location attributes of the user. For example, the resident base stations of h1 are (c1, c2, c3, c4, c5,...); obtaining the home broadband, communication packages, and consumption data of users can be used as the basic data for subsequent marketing.

[0066] In the embodiments of the present application, the feature data of each node at least includes:

[0067] The association relationship between each node and other nodes; the association relationship between each node and other nodes includes: the attention score between each node and other nodes; where

[0068] A communication and social event occurs between the user corresponding to each node and the user corresponding to other nodes.

[0069] In an embodiment of the present application, a network relationship attribute graph can be constructed based on the communication and social event data of users in the same network. The network relationship attribute graph includes: user nodes (such as user h1), user node attributes, and the edges between users; the edges between users represent that users have had communication and social events and generated social relationships;

[0070] The user node attributes in the network relationship attribute graph include the following information: whether the corresponding user is the main card user of the mobile phone, age, gender, permanent residence, etc.;

[0071] Users are connected through the communication and social events of users. For example, if a communication and social event occurs between user 1 and user 2, then the user nodes h1 and h2 are connected; for another example, if there is a text message sending and / or 5G message event between user 2 and user 3, then the user nodes h2 and h3 are connected; the connected edges can be directed edges, and the direction of the connected edge is determined by the active and passive parties.

[0072] As an embodiment, if there are a large number of users, the users can be divided by region, and a network relationship attribute graph between users can be constructed according to the region.

[0073] In one embodiment, the network relationship attribute graph can be expressed as: G = (V, E, W, A), where G represents the information of the heterogeneous attribute graph, V is the set of nodes (i.e., the user nodes), E is the set of edges (i.e., the edges between users), W is the weight matrix of the edges, and A represents the set of attributes of each node. For example: the edge connecting node h1 to node h2 is represented by e(h1, h2), and w h1h2 is the connection weight between node h1 and node h2, and the weight represents the degree of relationship between nodes. α in A h1 represents the attribute of node h1; the connection weight has been introduced above and will not be elaborated here.

[0074] In the embodiments of the present application, learning the network relationship attribute graph by the Graph Attention Network (GAT) model to obtain the feature data of each node in the network relationship attribute graph includes:

[0075] Projecting the initial feature data of each node into a preset representation space based on a weight matrix in the network relationship attribute graph;

[0076] Calculating corresponding attention scores for each pair of neighbor nodes in the network relationship attribute graph; wherein, the two nodes in the pair of neighbor nodes respectively correspond to two users who have communication and social events; the attention score is used to represent the importance of one node to another node;

[0077] Normalizing all attention scores to obtain the normalized attention scores;

[0078] Using the normalized attention scores to perform weighted summation on the feature data of the neighbor nodes of each node, and the new node feature data obtained is the feature data of each node in the network relationship attribute graph.

[0079] Here, the formula for calculating the attention score e ij can be expressed as: e ij = a([Wh i ||Wh j ), j ∈ N i , where the shared parameter W is a learnable weight matrix used to perform a linear transformation on the input feature h i to increase the dimension of the feature of h i . [.||.] is to splice the results after the transformation of nodes i and j, and then through the attention mechanism a function, map the spliced high-dimensional features to a real number, usually a small neural network, which accepts the features of two nodes and outputs a scalar value, and can be implemented by a single-layer feedforward neural network. What is obtained is the calculation of the relationship strength between node i and node j:

[0080]

[0081] where the function a is a linear aggregation function; W is a learnable weight matrix used to linearly transform the input features;

[0082] The above normalization is usually completed using the softmax function (as shown below) to ensure that the sum of all scores is 1:

[0083] Leaky ReLU is a type of activation function; where Ni is the set of neighbor nodes of node i; exp is the exponential function with the natural constant e as the base.

[0084] The above weighted method is implemented by the following formula: h′ i It is the new feature output by GAT for each node i that integrates domain information. is the activation function; the rectangular box in brackets represents concatenate.

[0085] The above process describes a periodic graph feature learning. Due to the time series characteristics of communication data, the period, time distance and frequency of communication can reflect the depth of user relationships. Therefore, the embodiment of the present application proposes a graph network model feature learning based on time series (multiple periods), as follows:

[0086] The communication events are divided into [T0, T1, ..., Tn] periods, and each period can be learned and represented by a graph network model to represent the vector representation of the association relationship between each node.

[0087] The processing method based on the multi-head attention mechanism is to perform n different attention calculations on the same node. This method can capture more complex and rich relationships between nodes. Because communication events between nodes do not necessarily occur in every cycle. If the frequency is high and there are communication events in each cycle, the relationship between nodes is closer. If the last communication occurred n cycles ago, the node relationship will gradually become distant as the cycle changes. Because each head may learn different attention weights in different cycles, it can pay attention to different neighboring nodes, or different characteristics of the same neighboring nodes.

[0088] In the embodiment of the present application, the feature data of each node in more than two cycles is obtained based on the multi-head attention mechanism, including:

[0089] For a single node, assume that a graph attention layer has K independent heads, and perform independent attention calculation on each head to obtain K new node feature sets; among them, all node feature data in each cycle corresponds to a graph attention layer;

[0090] In the first n-1 graph attention layers, the outputs of (n-1)*k heads are integrated by averaging; n represents the number of cycles;

[0091] In the nth graph attention layer, according to the event decay characteristics and combined with the intensity temporal decay function, the outputs of the K heads are spliced in a splicing manner to obtain the final node features of the single node.

[0092] Here, the value of K has nothing to do with the nodes in the network relationship attribute graph, but is related to the complexity of the graph network model. The larger K is, the more complex the model is, the more information is learned, and the more computing power is consumed.

[0093] The number of trainable parameters of multi-head attention is the number of parameters of single-head attention multiplied by the number of attention heads. Specifically, for each attention head, the number of parameters of the linear transformation is the number of parameters of the original linear transformation multiplied by the number of attention heads. For example, if single-head attention has 3 linear transformation parameters (Wq, Wk, Wv), and multi-head attention has 5 attention heads, then the number of linear transformation parameters of multi-head attention is 3 parameters multiplied by 5 attention heads, that is, 15 parameters. Through the multi-head attention mechanism, the model can use different attention heads to learn different features and context information, thereby enhancing the model's representation ability and expressive ability. Each attention head can focus on different aspects of the input features, thus providing a more comprehensive and diverse information representation.

[0094] Here, each head has its own parameters (such as the weight matrix W of the linear transformation and the attention mechanism a function). The K new sets of node features can be represented as:

[0095] The output integrated by the averaging method is expressed as:

[0096] The final node feature is expressed as:

[0097] Among them, the rectangular box before k(t - s) represents concatenate; the k(t - s) is a kernel function used to simulate the time decay effect of past history on the current event, and it adopts the form of an exponential function, which can meet the fact that the formation of current neighbor nodes may be affected by more intense recent events, while events that occurred in a longer history contribute less to the current target event.

[0098] In the embodiments of the present application, determining the off-net family members of the users of this network based on the feature data of each node within two or more cycles includes:

[0099] Select an off-net user from the users corresponding to each node in the network relationship attribute graph as the center point (the operator can directly identify the off-net user through the number segment);

[0100] Based on the attention scores between the node corresponding to the off-net user and different neighbor nodes, screen out a preset number of close-relationship users with the highest close-relationship with the off-net user from the neighbor users of the off-net user;

[0101] Calculate the similarity of the night-time permanent residence between the off-net user and the close-relationship users based on the permanent residence location information. If the similarity of the night-time permanent residence between the two is high, determine that the close-relationship user is a family member of the off-net user;

[0102] Select a home network user from the family members of the off-net user, and determine that the off-net user is an off-net family member of the home network user.

[0103] The location information of a user mainly includes the daytime permanent residence (which can be regarded as the permanent work place) and the nighttime permanent residence (residence and other mobile locations). In order to screen family users (family members), only the similarity of the nighttime permanent residence of the user needs to be calculated; in this application, the dynamic time warping method is used to calculate the similarity of the nighttime permanent residence of user i and user j; the idea of dynamic time warping (DTW) is to automatically warp two sequences and perform local scaling alignment on the time axis to make their shapes as consistent as possible, so as to obtain the greatest possible similarity. DTW maps the points of two trajectories in a many-to-many manner, thus more efficiently solving the problem of data misalignment.

[0104] Assume that the nighttime permanent residence sequences of user i and user j are: tr i and tr j , and their lengths are n and m respectively: tr i =q1, q2,..., qi,..., qn, where qi is the i-th permanent residence information of user i, and tr j =c1, c2,..., cj,..., cm, where cj is the j-th permanent residence information of user j, and m and n are not necessarily equal;

[0105] Its dynamic programming algorithm is as follows:

[0106]

[0107] Among them, the tri represents the Head(tr) of user i, which represents the first point of this trajectory; Rest(tr) represents the subsequence composed of all points except the first point; d DTW (tr i , Rest(tr j )) is the distance matrix for calculating two sequences using DWT.

[0108] This application believes that there are different numbers of family circles in different user age stages. Therefore, the value of N can be determined through the age attribute of this user, and the TOP-N intimate relationship users of this off-net user can be screened as the family circle of this off-net user.

[0109] In the embodiments of this application, the marketing plan for determining the off-net family members based on the service data of the home network user includes:

[0110] Analyze the service data of the home network user;

[0111] Based on the service data, determine the door-to-door marketing plan for the off-net family members of the home network user;

[0112] Among them, the service data includes one or more of the following:

[0113] Broadband quality data;

[0114] Broadband complaint data;

[0115] Smart home service subscription data;

[0116] Networking service data.

[0117] In order to obtain the opportunity to contact users of other networks for porting-in marketing, it is necessary to obtain door-to-door marketing business opportunities through their family members; therefore, it is necessary to analyze the service data of family member users (users of this network) for business opportunity mining. For example, analyze the home broadband quality, complaint data, smart home service subscriptions, networking services, etc. of family member users, and use broadband free medical services, smart home product contract recommendations, home broadband quality on-site inspections, etc. as opportunities to obtain door-to-door service opportunities, so as to obtain the opportunity to reach family users of other networks and conduct porting-in marketing of users of other networks.

[0118] In actual application, based on the service usage data of the family member users of this network, their consumption levels and consumption preferences can be analyzed, and combined with the home service subscription situation of the family member users, combined recommendations of packages, home services, and primary / secondary SIM card services can be made, and tenant services can be recommended to users, so as to achieve integrated marketing of family members. For example, if there is a decision-making user among the family member users, a shared package can be recommended to this user of other networks to achieve porting-in of other networks at the lowest cost; if there is a broadband service of other networks among the family member users, a service of free broadband can be handled for the porting-in of users of other networks, and by effectively combining the services of family members, the maximum benefit of the service combination of each family member can be achieved.

[0119] To implement the method of the embodiments of the present application, the embodiments of the present application also provide a device for porting users of other networks, as Figure 3 shown. The device includes:

[0120] A first processing unit 301, configured to construct a network relationship attribute graph based on the communication and social event data of users of this network obtained within a single period; the nodes in the network relationship attribute graph include: the node corresponding to the user of this network and the nodes corresponding to other users who have communication and social events with the user of this network; learning the network relationship attribute graph based on the graph attention network GAT model to obtain the feature data of each node in the network relationship attribute graph;

[0121] A second processing unit 302, configured to obtain the feature data of each node within two or more periods based on the multi-head attention mechanism; determine the family members of other networks of users of this network based on the feature data of each node within the two or more periods; and determine the marketing plan for the family members of other networks based on the service data of the users of this network.

[0122] In the embodiments of the present application, the communication and social event data includes: communication and social event data between at least two users; the at least two users include: at least one user of the same network and at least one user of a different network;

[0123] The communication and social event data further includes: a pre-configured association weight between the user of the same network and the other users.

[0124] In the embodiments of the present application, the feature data of each node at least includes:

[0125] The association relationship between each node and other nodes; the association relationship between each node and other nodes includes: the attention score between each node and other nodes; where

[0126] A communication and social event occurs between the user corresponding to each node and the user corresponding to other nodes.

[0127] In the embodiments of the present application, the first processing unit 301 learns the network relationship attribute graph based on the graph attention network GAT model to obtain the feature data of each node in the network relationship attribute graph, including:

[0128] Project the initial feature data of each node into a preset representation space based on a weight matrix in the network relationship attribute graph;

[0129] Calculate the corresponding attention score for each pair of neighbor nodes in the network relationship attribute graph; where the two nodes in a pair of neighbor nodes respectively correspond to two users who have a communication and social event; the attention score is used to represent the importance of one node to another node;

[0130] Normalize all attention scores to obtain the normalized attention scores;

[0131] Use the normalized attention scores to perform weighted summation on the feature data of the neighbor nodes of each node, and the new node feature data obtained is the feature data of each node in the network relationship attribute graph.

[0132] In the embodiments of the present application, the second processing unit 302 obtains the feature data of each node within more than two cycles based on the multi-head attention mechanism, including:

[0133] For a single node, assuming that a graph attention layer has K independent heads, perform independent attention calculations for each head to obtain K new node feature sets; where all the node feature data within each cycle corresponds to a graph attention layer;

[0134] In the first n - 1 graph attention layers, the outputs of (n - 1)*k heads are integrated using the average method; where n represents the number of periods.

[0135] In the nth graph attention layer, according to the event attenuation characteristic and in combination with the intensity time - series attenuation function, the outputs of K heads are concatenated in a concatenation manner to obtain the final node feature of the single node.

[0136] In the embodiment of the present application, the second processing unit 302 determines the off - network family members of the users of this network based on the feature data of each node within the two or more periods, including:

[0137] Select an off - network user from the users corresponding to each node in the network relationship attribute graph as the center point;

[0138] Based on the attention scores between the node corresponding to the off - network user and different neighbor nodes, screen out a preset number of close - relationship users with the highest close - relationship with the off - network user from the neighbor users of the off - network user;

[0139] Calculate the similarity of the night resident locations between the off - network user and the close - relationship users based on the resident location information. If the similarity of their night resident locations is high, determine the close - relationship users as the family members of the off - network user;

[0140] Select the users of this network from the family members of the off - network user, and determine the off - network user as the off - network family member of the users of this network.

[0141] In the embodiment of the present application, the second processing unit 302 determines the marketing plan for the off - network family members based on the service data of the users of this network, including:

[0142] Analyze the service data of the users of this network;

[0143] Based on the service data, determine the on - site marketing plan for the off - network family members of the users of this network;

[0144] Among them, the service data includes one or more of the following:

[0145] Broadband quality data;

[0146] Broadband complaint data;

[0147] Smart home service subscription data;

[0148] Networking service data.

[0149] In actual application, the first processing unit 301 and the second processing unit 302 can be implemented by the processor in the off - network user porting device.

[0150] It should be noted that: when the off-net user porting device provided in the above embodiments communicates, only the division of the above program modules is used for illustration. In actual applications, the above processing can be allocated to different program modules according to needs, that is, the internal structure of the device is divided into different program modules to complete all or part of the above-described processing. In addition, the device provided in the above embodiments and the method embodiments belong to the same concept. For the specific implementation process, please refer to the method embodiments, which will not be elaborated here.

[0151] Based on the hardware implementation of the above program modules, and in order to implement the method of the embodiments of the present application, the embodiments of the present application further provide an off-net user porting device, as Figure 4 shown. The off-net user porting device 400 includes:

[0152] A first communication interface 401, capable of interacting with the terminal and / or other nodes on the network side;

[0153] A first processor 402, connected to the first communication interface 401 to implement information interaction with the terminal and / or other nodes on the network side, and used to execute the methods provided by one or more technical solutions on the off-net user porting device side when running a computer program;

[0154] A first memory 403, on which the computer program is stored.

[0155] Specifically, the first processor 402 is used to construct a network relationship attribute graph based on the communication and social event data of the on-net users obtained within a single cycle; the nodes in the network relationship attribute graph include: the nodes corresponding to the on-net users and the nodes corresponding to other users who have communication and social events with the on-net users; based on the graph attention network GAT model, learn the network relationship attribute graph to obtain the feature data of each node in the network relationship attribute graph;

[0156] Obtain the feature data of each node within two or more cycles based on the multi-head attention mechanism; determine the off-net family members of the on-net users based on the feature data of each node within the two or more cycles; determine the marketing plan for the off-net family members based on the service data of the on-net users.

[0157] In the embodiments of the present application, the communication and social event data includes: communication and social event data between at least two users; the at least two users include: at least one on-net user and at least one off-net user;

[0158] The communication and social event data further includes: a pre-configured association weight between the on-net user and the other users.

[0159] In the embodiments of the present application, the feature data of each node at least includes:

[0160] The association relationship between each node and other nodes; the association relationship between each node and other nodes includes: the attention score between each node and other nodes; wherein,

[0161] A communication and social event occurs between the user corresponding to each node and the user corresponding to other nodes.

[0162] In the embodiments of the present application, the first processor 402 learns the network relationship attribute graph based on the graph attention network GAT model to obtain the feature data of each node in the network relationship attribute graph, including:

[0163] Project the initial feature data of each node into a preset representation space based on a weight matrix in the network relationship attribute graph;

[0164] Calculate the corresponding attention score for each pair of neighbor nodes in the network relationship attribute graph; wherein, the two nodes in a pair of neighbor nodes respectively correspond to two users who have a communication and social event; the attention score is used to represent the importance of one node to another node;

[0165] Normalize all attention scores to obtain the normalized attention scores;

[0166] Use the normalized attention scores to perform weighted summation on the feature data of the neighbor nodes of each node, and the new node feature data obtained is the feature data of each node in the network relationship attribute graph.

[0167] In the embodiments of the present application, the first processor 402 obtains the feature data of each node in more than two cycles based on the multi-head attention mechanism, including:

[0168] For a single node, assume that a graph attention layer has K independent heads, and perform independent attention calculations for each head to obtain K new node feature sets; wherein, all the node feature data in each cycle corresponds to a graph attention layer;

[0169] In the first n - 1 graph attention layers, use the averaging method to integrate the outputs of (n - 1)*k heads; the n represents the number of cycles;

[0170] In the nth graph attention layer, according to the event decay characteristic and combined with the intensity time decay function, splice the outputs of K heads in a splicing manner to obtain the final node feature of the single node.

[0171] In the embodiment of the present application, the first processor 402 determines the off-net family members of the users on this network based on the feature data of each node within the above two or more cycles, including:

[0172] Select an off-net user from the users corresponding to each node in the network relationship attribute graph as the center point;

[0173] Based on the attention scores between the node corresponding to the off-net user and different neighbor nodes, screen out a preset number of users with the highest intimacy relationship with the off-net user from the neighbor users of the off-net user;

[0174] Calculate the similarity of the night resident locations between the off-net user and the users with intimacy relationship. If the similarity of the night resident locations between the two is high, determine the users with intimacy relationship as the family members of the off-net user;

[0175] Select the users on this network from the family members of the off-net user, and determine the off-net user as the off-net family member of the users on this network.

[0176] In the embodiment of the present application, the first processor 402 determines the marketing plan for the off-net family members based on the service data of the users on this network, including:

[0177] Analyze the service data of the users on this network;

[0178] Based on the service data, determine the on-site marketing plan for the off-net family members of the users on this network;

[0179] Among them, the service data includes one or more of the following:

[0180] Broadband quality data;

[0181] Broadband complaint data;

[0182] Smart home service subscription data;

[0183] Networking service data.

[0184] It should be noted that: the specific processing processes of the first communication interface 401 and the first processor 402 can be understood with reference to the above method, and will not be elaborated here.

[0185] Of course, in actual application, each component in the off-net user's porting device 400 is coupled together through the bus system 404. It can be understood that the bus system 404 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 404 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clear description, in Figure 4 all kinds of buses are labeled as the bus system 404.

[0186] The first memory 403 in the embodiments of the present application is used to store various types of data to support the operation of the cross-network user porting device 400. Examples of such data include: any computer programs for operating on the cross-network user porting device 400.

[0187] The methods disclosed in the embodiments of the present application described above can be applied to the first processor 402 or implemented by the first processor 402. The first processor 402 may be an integrated circuit chip with signal processing capabilities. During implementation, the steps of the above methods can be completed through the integrated logic circuit of the hardware in the first processor 402 or instructions in software form. The above first processor 402 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The first processor 402 can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor may be a microprocessor or any conventional processor, etc. Combining the steps of the methods disclosed in the embodiments of the present application, it can be directly embodied as being executed and completed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium, and this storage medium is located in the first memory 403. The first processor 402 reads the information in the first memory 403 and combines its hardware to complete the steps of the foregoing methods.

[0188] In an exemplary embodiment, the cross-network user porting device 400 can be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontroller units (MCUs), microprocessors, or other electronic components for executing the foregoing methods.

[0189] In an exemplary embodiment, the embodiment of the present application further provides a storage medium, specifically a computer-readable storage medium, for example, including a first memory 403 storing a computer program, and the above computer program can be executed by a first processor 402 of the off-net user porting device 400 to complete the steps of the foregoing method on the off-net user porting device side. The computer-readable storage medium may be a FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disc, or CD-ROM, etc.

[0190] It should be noted that: "first", "second", etc. are used to distinguish similar objects and do not necessarily have to be used to describe a specific order or sequence.

[0191] In addition, the technical solutions described in the embodiments of the present application can be arbitrarily combined without conflict.

[0192] The above is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application.

Claims

1. A method for transferring a user between different networks, characterized in that, The method includes: Constructing a network relationship attribute graph based on the communication and social event data of the users on the local network obtained within a single period; the nodes in the network relationship attribute graph include: the nodes corresponding to the users on the local network and the nodes corresponding to other users who have communication and social events with the users on the local network; Learning the network relationship attribute graph based on the Graph Attention Network (GAT) model to obtain the feature data of each node in the network relationship attribute graph; Obtaining the feature data of each node within more than two periods based on the multi-head attention mechanism; Determining the off-network family members of the users on the local network based on the feature data of each node within more than two periods; Determining the marketing plan for the off-network family members based on the service data of the users on the local network.

2. The method according to claim 1, characterized in that The communication and social event data includes: the communication and social event data between at least two users; the at least two users include: at least one user on the local network and at least one off-network user; The communication and social event data further includes: the pre-configured association weight between the users on the local network and the other users.

3. The method according to claim 1, wherein The feature data of each node at least includes: The association relationship between each node and other nodes; the association relationship between each node and other nodes includes: the attention score between each node and other nodes; wherein, A communication and social event occurs between the user corresponding to each node and the users corresponding to other nodes.

4. The method according to claim 1, characterized in that, The learning of the network relationship attribute graph based on the Graph Attention Network (GAT) model to obtain the feature data of each node in the network relationship attribute graph includes: Projecting the initial feature data of each node into a preset representation space based on a weight matrix in the network relationship attribute graph; Calculating the corresponding attention score for each pair of neighbor nodes in the network relationship attribute graph; wherein, the two nodes in a pair of neighbor nodes respectively correspond to two users who have a communication and social event; the attention score is used to represent the importance of one node to another node; Normalizing all the attention scores to obtain the normalized attention scores; Using the normalized attention scores to perform weighted summation on the feature data of the neighbor nodes of each node, and the new node feature data obtained is the feature data of each node in the network relationship attribute graph.

5. The method according to claim 1, wherein The obtaining of the feature data of each node within more than two periods based on the multi-head attention mechanism includes: For a single node, assuming that a graph attention layer has K independent heads, performing independent attention calculations for each head to obtain K new node feature sets; wherein, all the node feature data within each period corresponds to a graph attention layer; Adopting the averaging method to integrate the outputs of (n - 1)*K heads in the first n - 1 graph attention layers; the n represents the number of periods; In the nth graph attention layer, according to the event attenuation characteristic and in combination with the intensity time series attenuation function, splicing the outputs of the K heads in a splicing manner to obtain the final node feature of the single node.

6. The method according to claim 4, characterized in that, The determining of the off-network family members of the users on the local network based on the feature data of each node within more than two periods includes: Select a cross-network user from the users corresponding to each node in the network relationship attribute graph as the center point; Based on the attention scores between the node corresponding to the cross-network user and different neighbor nodes, screen out a preset number of intimate relationship users with the highest intimate relationship with the cross-network user from the neighbor users of the cross-network user; Calculate the night resident similarity between the cross-network user and the intimate relationship user based on the permanent residence location information. If the night resident similarity between the two is high, determine that the intimate relationship user is a family member of the cross-network user; Select an in-network user from the family members of the cross-network user, and determine that the cross-network user is an out-of-network family member of the in-network user.

7. The method according to claim 6, wherein The marketing plan for determining the out-of-network family member based on the service data of the in-network user includes: Analyze the service data of the in-network user; Based on the service data, determine the door-to-door marketing plan for the out-of-network family member of the in-network user; Among them, the service data includes one or more of the following: Broadband quality data; Broadband complaint data; Smart home service subscription data; Networking service data.

8. A device for transferring a user between different networks, characterized in that The device includes: A first processing unit, configured to construct a network relationship attribute graph based on the communication and social event data of in-network users obtained within a single period; the nodes in the network relationship attribute graph include: the node corresponding to the in-network user and the nodes corresponding to other users who have communication and social events with the in-network user; learn the network relationship attribute graph based on the Graph Attention Network (GAT) model to obtain the feature data of each node in the network relationship attribute graph; A second processing unit, configured to obtain the feature data of each node within two or more periods based on the multi-head attention mechanism; determine the out-of-network family members of the in-network user based on the feature data of each node within the two or more periods; determine the marketing plan for the out-of-network family members based on the service data of the in-network user.

9. A device for a different network user to carry and transfer, characterized in that, Include: A first communication interface and a first processor; wherein, The first processor is configured to construct a network relationship attribute graph based on the communication and social event data of in-network users obtained within a single period; the nodes in the network relationship attribute graph include: the node corresponding to the in-network user and the nodes corresponding to other users who have communication and social events with the in-network user; learn the network relationship attribute graph based on the Graph Attention Network (GAT) model to obtain the feature data of each node in the network relationship attribute graph; Obtain the feature data of each node within two or more periods based on the multi-head attention mechanism; determine the out-of-network family members of the in-network user based on the feature data of each node within the two or more periods; determine the marketing plan for the out-of-network family members based on the service data of the in-network user.

10. A cross-network user porting device, characterized in that, Include: A first processor and a first memory for storing a computer program that can run on the processor, Wherein, when the first processor is used to run the computer program, it executes the steps of the method according to any one of claims 1 to 7.

11. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 7.

Images