A method, apparatus, equipment, medium, and product for predicting number portability.

Abstract

This disclosure presents a method, apparatus, device, medium, and product for predicting number portability. The method includes: acquiring multi-source heterogeneous data of a target user; wherein the multi-source heterogeneous data indicates the user characteristics of the target user; processing the multi-source heterogeneous data to obtain a heterogeneous graph corresponding to the target user; wherein the heterogeneous graph includes target nodes and their associated edge weights, the associated edge weights indicating the importance between target nodes of different node types related to the target user; performing aggregation processing on the heterogeneous graph to obtain an embedding vector of the target user; wherein the embedding vector indicates the feature vector resulting from the combination of the target node and the associated edge weights; and inputting the embedding vector into a trained multilayer perceptron for processing to obtain a number portability prediction result for the target user.

Description

Technical Field

[0001] This disclosure generally relates to the field of data processing technology, and in particular to a prediction method, apparatus, equipment, medium and product for number portability. Background Technology

[0002] Currently, with the global telecommunications market becoming increasingly saturated, users are switching networks more and more frequently, which poses a significant challenge to operators' user retention and market strategies.

[0003] In related technologies, analysis of historical behavioral data is typically employed, lacking consideration for users' social influences and personalized needs. Furthermore, these methods are inadequate when handling heterogeneous data, failing to fully uncover complex correlations and dynamic patterns among data points. Therefore, inaccurate predictions of users' switching behavior may arise. Summary of the Invention

[0004] This disclosure addresses some deficiencies mentioned in the background art by providing a method, apparatus, device, medium, and product for predicting number portability.

[0005] In a first aspect, embodiments of this disclosure provide a method for predicting number portability, comprising: Acquire multi-source heterogeneous data of the target user; wherein the multi-source heterogeneous data is used to indicate the user characteristics of the target user; Based on the processing of the multi-source heterogeneous data, a heterogeneous graph corresponding to the target user is obtained; wherein, the heterogeneous graph includes target nodes and the weights of the associated edges of the target nodes, and the weights of the associated edges are used to indicate the importance between target nodes of different node types related to the target user; Based on the heterogeneous graph, an aggregation process is performed to obtain the embedding vector of the target user; wherein, the embedding vector is used to indicate the feature vector after combining the target node and the associated edge weights; The embedded vector is input into a trained multilayer perceptron for processing to obtain the number portability prediction result for the target user.

[0006] In one embodiment of the first aspect, the step of processing the multi-source heterogeneous data to obtain the heterogeneous graph corresponding to the target user includes: The feature data of each type in the multi-source heterogeneous data are normalized to obtain normalized multi-source heterogeneous data. The feature data of each type in the normalized multi-source heterogeneous data are quantified to obtain the standardized feature data of the target user. Based on the standardized feature data, a heterogeneous graph corresponding to the target user is obtained.

[0007] In one embodiment of the first aspect, the node type includes user nodes, associated user nodes, and other types of nodes, wherein the other types are at least one node type; The process of processing the standardized feature data to obtain the heterogeneous graph corresponding to the target user includes: Determine the sub-feature data corresponding to each node type from the standardized feature data; Feature mapping is performed on the sub-feature data corresponding to each of the node types to obtain the target feature vector corresponding to each target node; The association features between the user node and the associated user node are quantified to obtain the weight of the first sub-association edge. The association features between the user node and each of the other types of nodes are quantified to obtain the weight of the second sub-association edge. The weights of the first and second sub-associated edges are determined as the associated edge weights of the target node.

[0008] In one embodiment of the first aspect, the aggregation process based on the heterogeneous graph to obtain the embedding vector of the target user includes: Based on the target feature vector of the target node, the target node is aggregated to obtain the aggregated feature vector corresponding to the node type of each type of target node; Based on the weights of the associated edges, an aggregation process is performed to obtain the target importance corresponding to each associated edge; The embedding vector of the target user is obtained by processing the aggregated feature vector and the target importance.

[0009] In one embodiment of the first aspect, the node type includes user nodes, associated user nodes, and other types of nodes, wherein the other types are at least one node type; The aggregation process based on the target feature vector of the target node to obtain the aggregated feature vector corresponding to each node type includes: The user node and its associated user nodes are processed using a self-attention mechanism to obtain first importance; the user node and other types of nodes are processed using the same self-attention mechanism to obtain second importance. The first importance and the second importance are normalized respectively to obtain the normalized importance; The normalized importance is aggregated based on the target feature vector of each node type to obtain the aggregated feature vector corresponding to each node type.

[0010] In one embodiment of the first aspect, the trained multilayer perceptron is obtained by training in the following manner: Acquire historical multi-source heterogeneous data of historical users; The historical multi-source heterogeneous data is processed using a heterogeneous graph attention network to obtain the embedding vector of the historical user. The actual number portability status of the historical users is determined, and training samples are generated based on the embedding vector and the actual number portability status; wherein, the embedding vector is the feature of the training sample, and the actual number portability status is the label of the training sample; The multilayer perceptron to be trained is trained using the training samples to obtain the trained multilayer perceptron.

[0011] In a second aspect, embodiments of this disclosure provide a number portability prediction device, comprising: An acquisition unit is used to acquire multi-source heterogeneous data of a target user; wherein the multi-source heterogeneous data is used to indicate the user characteristics of the target user; The first processing unit is used to process the multi-source heterogeneous data to obtain a heterogeneous graph corresponding to the target user; wherein, the heterogeneous graph includes target nodes and the weights of the associated edges of the target nodes, and the weights of the associated edges are used to indicate the importance between target nodes of different node types related to the target user; The second processing unit is used to perform aggregation processing based on the heterogeneous graph to obtain the embedding vector of the target user; wherein the embedding vector is used to indicate the feature vector after the target node and the associated edge weights are combined; The third processing unit is used to input the embedding vector into the trained multilayer perceptron for processing to obtain the number portability prediction result of the target user.

[0012] In a third aspect, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of a number portability prediction method.

[0013] In a fourth aspect, a computer-readable storage medium is provided having a computer program / instructions stored thereon, which, when executed by a processor, implements the steps of a number portability prediction method. In a fifth aspect, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement steps of a number portability prediction method.

[0014] As will be described in detail below, a method, apparatus, device, medium, and product for predicting number portability according to embodiments of this disclosure are disclosed. By acquiring multi-source heterogeneous data of target users and constructing a heterogeneous graph, user characteristics are transformed into structured data containing target nodes and associated edge weights. The associated edge weights quantify the importance between different types of nodes and incorporate potential association information such as user social influence, thus solving the deficiency of related technologies that ignore non-behavioral association factors.

[0015] The embedding vector is obtained by aggregation based on heterogeneous graphs, which enables deep integration of target node features and associated edge weights, fully explores the complex correlations between multi-source heterogeneous data, and makes up for the insufficient ability of related technologies to process heterogeneous data. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a method for predicting number portability provided in this embodiment of the disclosure; Figure 2 This is a structural diagram of a number portability prediction system provided in an embodiment of the present disclosure; Figure 3 A schematic diagram of a number portability prediction device provided in an embodiment of this disclosure; Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of the present disclosure. Detailed Implementation

[0017] The present disclosure will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present disclosure and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present disclosure are shown in the drawings, not the entire structure.

[0018] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0019] In this document, the term "and / or" merely describes a relationship, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0020] Research has found that, with the global telecommunications market becoming increasingly saturated, users are switching networks more frequently while keeping their phone numbers, which poses a significant challenge to operators' user retention and market strategies.

[0021] In related technologies, analysis of historical behavioral data is typically employed, lacking consideration for users' social influences and personalized needs. Furthermore, these methods are inadequate when handling heterogeneous data, failing to fully uncover complex correlations and dynamic patterns among data points. Therefore, inaccurate predictions of users' switching behavior may arise.

[0022] Based on the above research, this disclosure provides a prediction method for number portability. By acquiring multi-source heterogeneous data of target users and constructing a heterogeneous graph, user characteristics are transformed into structured data containing target nodes and associated edge weights. The associated edge weights quantify the importance between different types of nodes and incorporate potential association information such as user social influence, thus solving the defect of related technologies that ignore non-behavioral association factors.

[0023] The embedding vector is obtained by aggregation based on heterogeneous graphs, which enables deep integration of target node features and associated edge weights, fully explores the complex correlations between multi-source heterogeneous data, and makes up for the insufficient ability of related technologies to process heterogeneous data.

[0024] To facilitate understanding of this embodiment, a method for predicting number portability disclosed in this disclosure will first be described in detail. The execution entity of this method is generally an electronic device with a certain computing power. In some possible implementations, this method can be implemented by a processor calling computer-readable instructions stored in memory.

[0025] Figure 1 The diagram shows a flowchart of a number portability prediction method provided in this embodiment of the present disclosure. The method includes steps S101 to S104, wherein: S101. Obtain multi-source heterogeneous data of the target user; wherein, the multi-source heterogeneous data is used to indicate the user characteristics of the target user.

[0026] In the embodiments of this disclosure, initial multi-source heterogeneous data can be obtained; wherein, the initial multi-source heterogeneous data includes, but is not limited to: user basic information, geographical location, APP usage behavior, communication records, and social network activities.

[0027] Here, after collecting the initial multi-source heterogeneous data, the initial multi-source heterogeneous data can be cleaned to obtain cleaned data.

[0028] Subsequently, feature extraction was performed on the cleaned data to obtain multi-source heterogeneous data of the target user.

[0029] S102. Process the multi-source heterogeneous data to obtain the heterogeneous graph corresponding to the target user; wherein, the heterogeneous graph includes the target node and the weight of the associated edge of the target node, and the weight of the associated edge is used to indicate the importance between target nodes of different node types related to the target user.

[0030] In the embodiments of this disclosure, the dimensions in multi-source heterogeneous data can be unified, and the multi-source heterogeneous data after unification can be analyzed to obtain the target node and the associated edge weight of the target node in the multi-source heterogeneous data.

[0031] The target nodes include user nodes, app nodes, and geolocation nodes. User nodes are used to indicate the nodes corresponding to the target user and users associated with the target user.

[0032] The APP node is used to indicate the application category (app_category) and usage frequency (app_usage_freq) of the APP used by the target user.

[0033] Among them, the geographic location node is used to indicate the residential and workplace location information of the target user and users who are related to the target user.

[0034] Then, the importance of various relationships (e.g., the target user's communication behavior and social connections) between target nodes of different node types related to the target user can be determined to obtain a heterogeneous graph corresponding to the target user.

[0035] S103. Perform aggregation processing based on heterogeneous graphs to obtain the embedding vector of the target user; wherein, the embedding vector is used to indicate the feature vector after the target node and the associated edge weights are combined.

[0036] In the embodiments of this disclosure, the importance of each target node and the associated edges between them can be determined. After determining the importance, aggregation can be performed based on the importance to obtain the embedding vector of the target user.

[0037] S104. Input the embedded vector into the trained multilayer perceptron for processing to obtain the number portability prediction result for the target user.

[0038] In embodiments of this disclosure, the trained multilayer perceptron can predict the probability of a target user switching mobile networks based on the embedding vector.

[0039] Then, the predicted number portability result for the target user can be determined based on the probability of the target user switching networks.

[0040] Here, if the probability of the target user switching mobile networks is greater than or equal to a preset probability threshold, the prediction result of the target user switching mobile networks can be determined as the target user's prediction result of switching mobile networks; otherwise, the prediction result of the target user not switching mobile networks can be determined as the target user's prediction result of switching mobile networks.

[0041] Those skilled in the art can set preset probability thresholds based on actual needs, which will not be elaborated here.

[0042] In the embodiments of this disclosure, firstly, multi-source heterogeneous data of the target user is acquired; wherein, the multi-source heterogeneous data is used to indicate the user characteristics of the target user; secondly, the multi-source heterogeneous data is processed to obtain a heterogeneous graph corresponding to the target user; wherein, the heterogeneous graph includes target nodes and the weights of the associated edges of the target nodes, and the weights of the associated edges are used to indicate the importance between target nodes of different node types related to the target user; thirdly, the heterogeneous graph is aggregated to obtain the embedding vector of the target user; wherein, the embedding vector is used to indicate the feature vector after the target node and the weights of the associated edges are combined; finally, the embedding vector is input into a trained multilayer perceptron for processing to obtain the number portability prediction result of the target user.

[0043] In the above implementation, by acquiring multi-source heterogeneous data of target users and constructing a heterogeneous graph, user features are transformed into structured data containing target nodes and associated edge weights. The associated edge weights quantify the importance between different types of nodes and incorporate potential association information such as user social influence, thus solving the defect of related technologies that ignore non-behavioral association factors.

[0044] The embedding vector is obtained by aggregation based on heterogeneous graphs, which enables deep integration of target node features and associated edge weights, fully explores the complex correlations between multi-source heterogeneous data, and makes up for the insufficient ability of related technologies to process heterogeneous data.

[0045] In one optional implementation, processing is performed based on multi-source heterogeneous data to obtain a heterogeneous graph corresponding to the target user, specifically including the following steps: First, the feature data of each type in the multi-source heterogeneous data are normalized to obtain normalized multi-source heterogeneous data. Secondly, the feature data of each type in the normalized multi-source heterogeneous data are quantified to obtain the standardized feature data of the target user. Finally, based on the standardized feature data, a heterogeneous graph corresponding to the target user is obtained.

[0046] In the embodiments of this disclosure, the feature data of various types can be normalized in the following manner: X norm = (XX) min ) / (Xmax -X min ); Among them, X morn Let X be the normalized feature data, and X be the feature vector corresponding to any type of feature data. min X is the smallest eigenvalue of this type of feature data. max This is the largest eigenvalue of this type of feature data.

[0047] After normalizing all types of feature data in the multi-source heterogeneous data using the above method, normalized multi-source heterogeneous data is obtained.

[0048] Here, the types of multi-source heterogeneous data feature data include, but are not limited to: social influence, geographical location, user behavior patterns, etc.

[0049] Here, the feature data of each type of multi-source heterogeneous data can be quantified to obtain the standardized feature data of the target user.

[0050] Here, the social influence of a target user can be determined using degree centrality (i.e., quantifying social influence), for example: degree centrality=Number of connections / Total possible connections; Here, Number of connections represents the number of objects that the target user actually interacts with effectively in the social network, and Total possible connections represents the maximum number of objects that the target user can theoretically interact with.

[0051] Here, the stability of geographic location can be quantified by analyzing the target user's residence and workplace. For example, the frequency of change of the user's residence and workplace within a preset observation period can be calculated. After quantification, the location stability meets the following conditions: Location Stability=1-Number of location changes / Total monthsobserved; Here, "Number of location changes" refers to the number of times the target user's residence or workplace changed within the preset observation period, and "Total months observed" is the preset observation period. Those skilled in the art can set the preset observation period according to actual needs, for example, 6 months.

[0052] Here, user behavior patterns can be quantified using cluster analysis or other machine learning algorithms to obtain a quantified user behavior model (Cluster assignment): Cluster assignment=K-means(X app usage ); Where K-means() is the clustering function, X app usage This is the feature vector corresponding to the user behavior pattern.

[0053] After determining the standardized feature data of the target users, a feature quantification table for the target users can be constructed based on this standardized feature data. That is, the quantified standardized feature data is integrated into a feature table to obtain the feature quantification table. Table 1 shows an example of a feature quantization table.

[0054] Table 1

[0055] Here, after determining the standardized feature data, user nodes and associated edges can be determined based on the standardized feature data, thereby generating a heterogeneous graph corresponding to the target user based on the target node and associated edges.

[0056] In one alternative implementation, the node types include user nodes, associated user nodes, and other types of nodes, where the other types are at least one node type.

[0057] The heterogeneous graph corresponding to the target user is obtained by processing standardized feature data, specifically including the following steps: First, determine the sub-feature data corresponding to each node type in the standardized feature data; Secondly, feature mapping is performed on the sub-feature data corresponding to each node type to obtain the target feature vector corresponding to each target node; Secondly, the association features between user nodes and associated user nodes are quantified to obtain the weight of the first sub-association edge; Secondly, the association features between user nodes and other types of nodes are quantified to obtain the weights of the second sub-association edges. Finally, the weights of the first and second sub-associated edges are determined as the associated edge weights of the target node.

[0058] In the embodiments of this disclosure, different transformation matrices can be set for different types of target nodes, and the sub-feature data of each target node can be mapped to a unified feature space through feature mapping.

[0059] Here, feature mapping of the target node can be performed in the following way: h'i =M i ×h i ; h i M represents the standardized feature data corresponding to the i-th target node. i h' is the transformation matrix corresponding to the type of the i-th target node. i Let be the target feature vector obtained after the transformation of the i-th target node.

[0060] Here, the association features between user nodes and associated user nodes are the features corresponding to the association edges between user nodes and associated user nodes.

[0061] Here, when the communication frequency between the target user and users associated with the target user exceeds a preset frequency threshold, an association edge is established between the target user and users associated with the target user.

[0062] Here, the aforementioned association features can be processed based on the communication behavior between the target user and users who are related to the target user to obtain the weight of the first sub-association edge.

[0063] Communication behaviors include, but are not limited to, communication frequency and SMS exchange frequency.

[0064] Here, the weight w of the first sub-association edge ij The following conditions must be met: w ij = Communication frequency i to j / Maximum number of communications; Wherein, the communication frequency i to j is the communication frequency between user node i and associated user node j, and the maximum number of communication is the maximum value of the communication frequency between the target user and users associated with the target user.

[0065] Here, the association features between user nodes and application nodes can be quantified based on the frequency or duration of application use by the target user to obtain the weight of the second sub-association edge.

[0066] Here, we can also perform quantification based on the frequency of the target user's geographic location to obtain the weight of the second sub-association edge corresponding to the user node and the geographic location node.

[0067] Here, the weight w of the second sub-association edge corresponding to the user node and the application node. ik The following conditions must be met: w ik =Usage duration i to k / Maximum usage duration; Wherein, usage duration i to k is the usage duration of user node i to application node k, and maximum usage duration is the maximum usage duration of user node i relative to application node k.

[0068] Here, the weight w of the second sub-association edge corresponding to the user node and the geographic location node. il The following conditions must be met: w il = Frequency of occurrence i to l / Total frequency of occurrence; Wherein, the frequency of occurrence i to l is the frequency of occurrence of user node i in geographical location node l, and the total frequency of occurrence is the total frequency of occurrence of user node in all geographical location nodes.

[0069] After determining the weights of the first and second sub-associative edges, aggregation can be performed based on the weights of the first and second sub-associative edges and each target node to construct a heterogeneous graph of the target user.

[0070] Here, target nodes and associated edges can be integrated into a unified graph structure based on the weights of the associated edges, resulting in a heterogeneous graph of the target user. Each target node and each associated edge corresponds to a feature and weight of the response.

[0071] In a heterogeneous graph, target nodes represent different entities, associated edges represent relationships between entities, and the weights of associated edges reflect the strength or importance of these relationships.

[0072] For example, a heterogeneous graph G can be represented in matrix form, and the feature matrix H of the target node and the weight matrix W of the associated edges can be used for subsequent training of the graph attention network: The node feature matrix H contains the feature vectors of all target nodes.

[0073] H={h' i , i=1……N}.

[0074] Edge weight matrix W: Contains the weights of all associated edges, used to adjust the importance of information transfer between target nodes in the attention mechanism.

[0075] W={w ij ;w ik ;w il , i=1……N, j=1……M, k=1……P}; where j takes values ​​from 1 to M, M is the number of associated user nodes, and k takes values ​​from 1 to P, P is the number of geographic location nodes.

[0076] In the above implementation, through such a construction and quantization process, heterogeneous graph attention networks can effectively process and learn complex information from different data sources and types, thereby improving the model's predictive performance and accuracy.

[0077] In an optional implementation, aggregation processing is performed based on the heterogeneous graph to obtain the embedding vector of the target user, specifically including the following steps: First, the target nodes are aggregated based on their target feature vectors to obtain aggregated feature vectors for each node type. Secondly, aggregation is performed based on the weights of the associated edges to obtain the target importance corresponding to the associated edges of each type of target node; Finally, based on the aggregated feature vector and the importance of the target, the embedding vector of the target user is obtained.

[0078] In the embodiments of this disclosure, the target feature vector of the target node can be aggregated by a self-attention mechanism to obtain the aggregated feature vector corresponding to each node type.

[0079] Here, the weights of the first and second sub-associative edges can be concatenated to obtain the concatenated weights.

[0080] Then, linear transformation and normalization can be performed based on the concatenated weights to obtain the target importance corresponding to each type of associated edge.

[0081] Here, the target importance Φij corresponding to the user node and the associated user node (i.e., the weight of the first sub-associated edge) satisfies the following condition: Φij=exp(W ij Φ ) / exp(W Φ ); Among them, W ij Φ The first sub-association edge weight is the linearized weight, exp() is the exponential function, and W Φ The weights are the linearized weights after concatenation.

[0082] Here, after determining the target importance and the aggregated feature vector, the aggregated feature vector of the same type and the target importance can be multiplied to obtain the multiplied vector of that type.

[0083] Then, the multiplication vectors of each type are summed to obtain the embedding vector of the target user.

[0084] In one optional implementation, the node types include user nodes, associated user nodes, and other types of nodes, where the other types are at least one node type; The target nodes are aggregated based on their target feature vectors to obtain aggregated feature vectors for each node type. The specific steps include the following: First, user nodes and their associated user nodes are processed using a self-attention mechanism to determine their primary importance; then, user nodes are processed using a self-attention mechanism to determine their secondary importance compared to other types of nodes. Secondly, the first importance and the second importance are normalized respectively to obtain the normalized importance; Finally, the normalized importance is aggregated based on the target feature vector of each node type to obtain the aggregated feature vector corresponding to each node type.

[0085] In the embodiments of this disclosure, a linear transformation is performed on the weight of the first sub-associated edge to obtain a first linear weight; a linear transformation is performed on the weight of the second sub-associated edge to obtain a second linear weight.

[0086] Here, the user node and its associated user nodes can be processed based on the first linear weight using a self-attention mechanism to obtain the first importance.

[0087] Among these, the second importance can be obtained by processing user nodes and other types of nodes based on a second linear weight through a self-attention mechanism.

[0088] Among them, the first importance e Φij The following conditions must be met: e Φij =attnode(h' i ,h' j ;Φij); Where attnode() is the neural network function for node-level attention, h' i h' is the feature vector of the user node. j This is the feature vector of the associated user node.

[0089] Then, the first and second importance can be normalized using the softmax() function to obtain the normalized importance.

[0090] Then, based on the normalized importance and the target feature vector of the node to be aggregated (the target feature vector of the aggregated node and the target feature vectors of the surrounding nodes of the aggregated node), the aggregation process is performed to obtain the aggregated feature vector corresponding to each node type.

[0091] In an optional implementation, the trained multilayer perceptron is trained as follows: First, acquire historical multi-source heterogeneous data of historical users; Secondly, the historical multi-source heterogeneous data is processed based on the heterogeneous graph attention network to obtain the embedding vectors of historical users; Secondly, determine the actual number portability status of historical users, and generate training samples based on the embedding vectors and the actual number portability status; where the embedding vectors are the features of the training samples, and the actual number portability status is the label of the training samples. Finally, the multilayer perceptron to be trained is trained using training samples to obtain the trained multilayer perceptron.

[0092] In the embodiments of this disclosure, the data corresponding to users within the target area is determined as historical multi-source heterogeneous data D. Specifically, if a user switches carriers within one year while keeping their number, the user's tag is determined to be 1; otherwise, the user's tag is determined to be 0.

[0093] Here, feature processing can be performed using the heterogeneous network HATN based on the above data to obtain the historical user embedding vectors that aggregate the heterogeneous data extracted from the above users and weighted by the attention mechanism. That is, Z=HATN(D).

[0094] Then, the embedding vectors and labels of the aforementioned historical users can be used as training samples for the target region.<Z,y> , where y is the label corresponding to Z.

[0095] The training samples can then be input into the multilayer perceptron to be trained to obtain the trained multilayer perceptron.

[0096] In this case, the cross-entropy loss function L can be used as the loss function for the multilayer perceptron to be trained, for example: L=∑ （h∈D） [α(y h log(y') h ) + (1-y h log(1-y') h )); Among them, y' h Let y be the prediction result of the multilayer perceptron to be trained for the h-th sample in the training samples. h Let be the label of the h-th sample in the training samples, and α be the weight for imbalance adjustment.

[0097] In one alternative implementation, real-time data can be continuously received from various data sources, such as user behavior data, transaction records, social network data, etc.

[0098] Here, real-time updated data will be used to continuously adjust and optimize the model, ensuring that the model reflects current user behavior and market dynamics.

[0099] One approach is to update the parameters of the multilayer perceptron after training using small-batch data updates, enabling the multilayer perceptron to frequently receive and learn new data while retaining its memory of historical learning.

[0100] Each time new data is received, the parameters in the trained multilayer perceptron can be updated again using the loss function.

[0101] Here, since the multilayer perceptron is a binary classification model after training, the F1 score can be used to measure the performance of the binary classification model under the condition of imbalanced data classes.

[0102] The F1 function satisfies the following conditions: F1=2×[Precision×Recall / (Precision+Recall)]; Among them, Precision is the proportion of users predicted to be number portability users in the training data who are actually number portability users, out of the total number of users predicted to be number portability users; Recall is the proportion of users predicted to be number portability users who are actually number portability users out of the total number of users who are actually number portability users.

[0103] Here, if the result of the F1 function exceeds a preset score threshold, prediction is performed on the training samples or the samples to be predicted. Those skilled in the art can set the preset score threshold based on actual needs, for example, 0.8.

[0104] In the above implementation, the parameters of the trained multilayer perceptron can be continuously updated and adjusted based on real-time data, so that the prediction results are more in line with the current market environment and user behavior, and effectively cope with the rapid evolution of 5G technology and the diversification of user needs.

[0105] Based on the same inventive concept, this disclosure also provides a number portability prediction system corresponding to the number portability prediction method. Since the principle of the system in this disclosure for solving the problem is similar to the number portability prediction method described above in this disclosure, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0106] Reference Figure 2 The diagram shown is a structural diagram of a number portability prediction system provided in this embodiment of the present disclosure, including: a feature extraction module 21, a feature fusion module 22, and a prediction module 23; wherein: The feature extraction module 21 extracts the geographical location, basic user feature information, and APP usage of the user to be detected, and extracts features from the above information using K-means to obtain multi-source heterogeneous data of the user to be detected.

[0107] The specific content, example variables, and data sources of multi-source heterogeneous data are shown in Table 2 below: Table 2

[0108] The feature fusion module 22 processes the multi-source heterogeneous data of the user to be detected to obtain a heterogeneous graph corresponding to the user; it aggregates the nodes based on the feature vectors of the nodes in the heterogeneous graph to obtain aggregated feature vectors corresponding to the node types of each type (i.e., multi-source node fusion); it aggregates the weights of the associated edges in the heterogeneous graph to obtain the specified importance of each associated edge (i.e., multi-source link fusion); and it integrates the aggregated feature vectors and the specified importance to obtain the embedding vector of the user to be detected.

[0109] The prediction module 23 is used to predict the embedding vector of the user to be detected by the trained multilayer perceptron (MLP) to obtain the prediction result of the user to be detected; and to update the parameters of the trained multilayer perceptron by real-time data.

[0110] In actual implementation, this embodiment has the following effects: (1) Enhanced quantification and integration of multi-source heterogeneous user characteristics: By integrating multi-source data such as user's geographical location, APP usage behavior, and social network structure, a more comprehensive and accurate user profile was constructed. This comprehensive data integration improved the prediction accuracy and adaptability of the multilayer perceptron for individual user behavior, enabling the multilayer perceptron to provide more personalized user insights in practical applications.

[0111] (2) Effective fusion and processing of multi-source heterogeneous data: It can more effectively explore the intrinsic connections and complex behavioral patterns between data, and significantly improve the explanatory power and prediction accuracy of multilayer perceptron.

[0112] (3) Enhanced the dynamic adaptive capability of the prediction model: The parameters can be continuously updated and adjusted according to real-time data, so that the prediction results are more in line with the current market environment and user behavior, and effectively cope with the rapid evolution of 5G technology and the diversification of user needs.

[0113] Based on the same inventive concept, this disclosure also provides a number portability prediction device corresponding to the number portability prediction method. Since the principle of the device in this disclosure is similar to the number portability prediction method described above, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0114] Reference Figure 3 The diagram shown is a schematic of a number portability prediction device provided in an embodiment of this disclosure, including: an acquisition unit 31, a first processing unit 32, a second processing unit 33, and a third processing unit 34; wherein: Acquisition unit 31 is used to acquire multi-source heterogeneous data of the target user; wherein, the multi-source heterogeneous data is used to indicate the user characteristics of the target user; The first processing unit 32 is used to process the multi-source heterogeneous data to obtain a heterogeneous graph corresponding to the target user; wherein, the heterogeneous graph includes target nodes and the weights of the associated edges of the target nodes, and the weights of the associated edges are used to indicate the importance between target nodes of different node types related to the target user; The second processing unit 33 is used to perform aggregation processing based on the heterogeneous graph to obtain the embedding vector of the target user; wherein, the embedding vector is used to indicate the feature vector after the target node and the associated edge weights are combined; The third processing unit 34 is used to input the embedding vector into the trained multilayer perceptron for processing to obtain the number portability prediction result of the target user.

[0115] This embodiment acquires multi-source heterogeneous data of target users and constructs a heterogeneous graph, transforming user features into structured data containing target nodes and associated edge weights. The associated edge weights quantify the importance between different types of nodes and incorporate potential association information such as user social influence, thus solving the defect of related technologies that ignore non-behavioral association factors.

[0116] The embedding vector is obtained by aggregation based on heterogeneous graphs, which enables deep integration of target node features and associated edge weights, fully explores the complex correlations between multi-source heterogeneous data, and makes up for the insufficient ability of related technologies to process heterogeneous data.

[0117] Corresponding to Figure 1 In addition to the method for predicting number portability in mobile networks, this disclosure also provides an electronic device 400, such as... Figure 4 The diagram shown is a structural schematic of an electronic device 400 provided in an embodiment of this disclosure, including: The system includes a processor 41, a memory 42, and a bus 43. The memory 42 stores execution instructions and includes main memory 421 and external memory 422. The main memory 421, also called internal memory, temporarily stores the computational data in the processor 41, as well as data exchanged with external memory such as a hard disk. The processor 41 exchanges data with the external memory 422 through the main memory 421. When the electronic device 400 is running, the processor 41 communicates with the memory 42 through the bus 43, causing the processor 41 to execute the following instructions: Acquire multi-source heterogeneous data of the target user; wherein the multi-source heterogeneous data is used to indicate the user characteristics of the target user; Based on the processing of the multi-source heterogeneous data, a heterogeneous graph corresponding to the target user is obtained; wherein, the heterogeneous graph includes target nodes and the weights of the associated edges of the target nodes, and the weights of the associated edges are used to indicate the importance between target nodes of different node types related to the target user; Based on the heterogeneous graph, an aggregation process is performed to obtain the embedding vector of the target user; wherein, the embedding vector is used to indicate the feature vector after combining the target node and the associated edge weights; The embedded vector is input into a trained multilayer perceptron for processing to obtain the number portability prediction result for the target user.

[0118] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0119] The block diagrams of devices, apparatuses, devices, and systems disclosed herein are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0120] Additionally, as used herein, the "or" used in a list of items beginning with "at least one" indicates a separate list, such that a list of, for example, "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not imply that the described example is preferred or better than other examples.

[0121] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0122] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0123] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0124] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. A method for predicting number portability, characterized in that, include: Acquire multi-source heterogeneous data of the target user; wherein the multi-source heterogeneous data is used to indicate the user characteristics of the target user; Based on the processing of the multi-source heterogeneous data, a heterogeneous graph corresponding to the target user is obtained; wherein, the heterogeneous graph includes target nodes and the weights of the associated edges of the target nodes, and the weights of the associated edges are used to indicate the importance between target nodes of different node types related to the target user; Based on the heterogeneous graph, an aggregation process is performed to obtain the embedding vector of the target user; wherein, the embedding vector is used to indicate the feature vector after combining the target node and the associated edge weights; The embedded vector is input into a trained multilayer perceptron for processing to obtain the number portability prediction result for the target user.

2. The method according to claim 1, characterized in that, The process of processing the multi-source heterogeneous data to obtain the heterogeneous graph corresponding to the target user includes: The feature data of each type in the multi-source heterogeneous data are normalized to obtain normalized multi-source heterogeneous data. The feature data of each type in the normalized multi-source heterogeneous data are quantified to obtain the standardized feature data of the target user. Based on the standardized feature data, a heterogeneous graph corresponding to the target user is obtained.

3. The method according to claim 2, characterized in that, The node types include user nodes, associated user nodes, and other types of nodes, wherein the other types are at least one node type. The process of processing the standardized feature data to obtain the heterogeneous graph corresponding to the target user includes: Determine the sub-feature data corresponding to each node type from the standardized feature data; Feature mapping is performed on the sub-feature data corresponding to each of the node types to obtain the target feature vector corresponding to each target node; The association features between the user node and the associated user node are quantified to obtain the weight of the first sub-association edge. The association features between the user node and each of the other types of nodes are quantified to obtain the weight of the second sub-association edge. The weights of the first and second sub-associated edges are determined as the associated edge weights of the target node.

4. The method according to claim 1, characterized in that, The aggregation process based on the heterogeneous graph to obtain the embedding vector of the target user includes: Based on the target feature vector of the target node, the target node is aggregated to obtain the aggregated feature vector corresponding to the node type of each type of target node; Based on the weights of the associated edges, an aggregation process is performed to obtain the target importance corresponding to each associated edge; The embedding vector of the target user is obtained by processing the aggregated feature vector and the target importance.

5. The method according to claim 4, characterized in that, The node types include user nodes, associated user nodes, and other types of nodes, wherein the other types are at least one node type. The aggregation process based on the target feature vector of the target node to obtain the aggregated feature vector corresponding to each node type includes: The user node and its associated user nodes are processed using a self-attention mechanism to obtain first importance; the user node and other types of nodes are processed using the same self-attention mechanism to obtain second importance. The first importance and the second importance are normalized respectively to obtain the normalized importance; The normalized importance is aggregated based on the target feature vector of each node type to obtain the aggregated feature vector corresponding to each node type.

6. The method according to claim 1, characterized in that, The trained multilayer perceptron was obtained by training it in the following manner: Acquire historical multi-source heterogeneous data of historical users; The historical multi-source heterogeneous data is processed using a heterogeneous graph attention network to obtain the embedding vector of the historical user. The actual number portability status of the historical users is determined, and training samples are generated based on the embedding vector and the actual number portability status; wherein, the embedding vector is the feature of the training sample, and the actual number portability status is the label of the training sample; The multilayer perceptron to be trained is trained using the training samples to obtain the trained multilayer perceptron.

7. A predictive device for number portability, characterized in that, include: An acquisition unit is used to acquire multi-source heterogeneous data of a target user; wherein the multi-source heterogeneous data is used to indicate the user characteristics of the target user; The first processing unit is used to process the multi-source heterogeneous data to obtain a heterogeneous graph corresponding to the target user; wherein, the heterogeneous graph includes target nodes and the weights of the associated edges of the target nodes, and the weights of the associated edges are used to indicate the importance between target nodes of different node types related to the target user; The second processing unit is used to perform aggregation processing based on the heterogeneous graph to obtain the embedding vector of the target user; wherein the embedding vector is used to indicate the feature vector after the target node and the associated edge weights are combined; The third processing unit is used to input the embedding vector into the trained multilayer perceptron for processing to obtain the number portability prediction result of the target user.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method as described in any one of claims 1 to 6.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method as described in any one of claims 1 to 6.

Images