Mobile user off-network prediction method and device and storage medium

Abstract

According to the mobile subscriber off-network prediction method and device and the storage medium, a relational graph is constructed through all SIM number cards in a certain area, feature nodes on the relational graph are formed by using network behaviors of the SIM number cards in a period of time, feature extraction on the relational graph is performed by using a graph convolution method, and the mobile subscriber off-network prediction result is obtained. And further predicting that a certain number card is off-grid in a period of time in the future. According to the method, the SIM number card is added into the social relation of the number card user, and off-network prediction is more accurate and more effective compared with off-network prediction performed by purely using the user network behavior. According to the method, a number card relation graph is constructed by using the feature of user communication, and feature extraction is performed on information between number cards by using a graph neural network, so that model convergence is further promoted by adding user classification. According to the invention, the communication information of the number card user is added to the model, so that the prediction of whether the user is off-network is more accurate; meanwhile, the addition of the number card affiliation cross entropy loss accelerates the model convergence.

Description

Technical Field

[0001] The present invention relates to the technical field of off-network prediction, and in particular to a method, device and storage medium for predicting off-network of a mobile user. Background Art

[0002] Off-network prediction is to accurately predict whether a user will leave the network in the future based on the network behavior of a certain user (call duration this month, traffic used, etc.), which is very important for telecom operators. It can help operators better manage and retain users; develop personalized services for users. Through off-network prediction, operators can provide personalized services based on the behavior and needs of each user. For example, for those users who are predicted to leave the network, operators can provide customized preferential plans or special services to increase user satisfaction and loyalty; it can better help enterprises control costs. Predicting user off-network behavior can help operators better plan their production and resource allocation to achieve more effective cost control. For example, if an operator knows that a user may leave the network, it can reasonably allocate customer service resources to most effectively solve the user's problems and avoid unnecessary cost waste; it can retain customers to the greatest extent possible, and predicting user off-network and taking corresponding measures can effectively retain existing users. When an operator knows that a user may leave the network, it can provide solutions or preferential activities in a timely manner to retain these users. This can reduce user churn and improve user loyalty and long-term benefits.

[0003] At present, in order to predict mobile off-network users, the off-network prediction of the target SIM card account in the network can be made based on the number association data corresponding to each SIM card account, reflecting the degree of user dependence on the SIM card. However, the existing prediction methods usually only use decision trees, which cannot compare the advantages and disadvantages, resulting in low prediction accuracy. However, in daily life, SIM card users have certain social behaviors and will have connections with the outside world. Generally speaking, the more frequently SIM card users contact the outside world, the less likely they are to leave the network, and vice versa.

[0004] Transformer is a deep learning model for processing sequence data, commonly used in natural language processing tasks such as machine translation, text classification and speech recognition. It mainly consists of two parts: encoder and decoder. The encoder converts the input sequence into a set of vectors representing semantic information, while the decoder uses these vectors to generate the output sequence. The Transformer model is characterized by the use of a self-attention mechanism to calculate the weight of each word to other words, allowing the model to better capture local and global information when processing sequence data. In addition, Transformer also uses techniques such as residual connections and normalization to alleviate the gradient disappearance / explosion problem and accelerate model training. In the encoder, each input word is converted into a vector representation, which is used to calculate the self-attention weight. Then, these weights are used to calculate a context vector, which summarizes all the information in the input sequence. In the decoder, this context vector is used to generate each word of the output sequence, and this process is repeated until the complete output sequence is generated. Summary of the invention

[0005] The present invention provides a method, device and storage medium for predicting when a mobile user will leave the network, which can at least solve one of the technical problems in the background technology.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] A method for predicting mobile user churn, comprising the following steps:

[0008] S1. Prepare a number-graph of the relationship between SIM cards in a certain area;

[0009] S2, prepare the degree-graph matrix of the relationship between the number cards, that is, how many communications each number card has in the recent period;

[0010] S3. Use the degree matrix of the relationship graph between number cards to normalize the relationship graph between number cards to obtain the adjacency matrix between number cards

[0011] Darea:

[0012] S4. Construct the feature vector number-feature of each number card's network behavior, and record the number-label of the number card, i.e. whether it is off-network, and the number-user of each number card;

[0013] S5. Construct a fully connected neural network user_fully_connect whose number of input neurons is the number-feature dimension of the feature vector and whose number of output neurons is the total number of users;

[0014] S6. Construct a fully connected neural network off_grid_fully_connect with the number of input neurons equal to the dimension of the feature vector number-feature and the number of output neurons equal to 2;

[0015] S7, construct a 2-layer card graph convolution network and set the card convolution parameter N;

[0016] S8. Build a transformer feature extractor.

[0017] S9, performing inter-number card graph convolution feature extraction on the number card feature vector to obtain a high-dimensional number card feature vector Output-number;

[0018] S10, sending the high-dimensional number card feature vector Output-number to the transformer feature extractor for higher-dimensional feature extraction to obtain number-high;

[0019] S11, send number-high to the fully connected neural network user_fully_connect to obtain the number card prediction attribution label number-user-classifier;

[0020] S12, send number-high to the fully connected neural network off_grid_fully_connect to obtain the prediction label number-off-grid of whether the number card is off the grid;

[0021] S13, inputting the prediction label number-off-grid of whether the number card is off-grid and the real label number-label of whether the number card is off-grid into the loss function L to obtain the cross entropy loss loss-off-grid of whether the number card is off-grid;

[0022] S14, sending the number card predicted belonging label number-user-classifier and the number card real belonging label number-user into the loss function L to obtain the number card belonging cross entropy loss loss-user;

[0023] S15. The cross entropy loss loss-off-grid of whether the number card is off-grid and the cross entropy loss loss-user of the number card are jointly trained to obtain a mobile user off-grid prediction model for predicting the mobile user off-grid situation.

[0024] Furthermore, the adjacency matrix between number cards Darea:

[0025] Darea=(degree-graph)^-0.5*(number-graph)*(degree-graph)^-0.5.

[0026] Further, step S9 includes,

[0027] Output-number-1=(N*item-feature)*Darea

[0028] Output-number=(N*Output-number-1)*Darea.

[0029] Furthermore, the loss function L is,

[0030]

[0031] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0032] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0033] From the above technical solution, it can be seen that the present invention constructs a relationship graph through all SIM cards in a certain area, uses the network behavior of SIM cards in a period of time to form feature nodes on the relationship graph, and uses the graph convolution method to extract features on the relationship graph, thereby predicting that a certain card will be offline in the future. This method adds the social relationship of the card user to the SIM card, which is more accurate and effective than simply using the user's network behavior to predict offline.

[0034] The present invention utilizes the feature of user communication to construct a number card relationship graph, uses a graph neural network to extract features from the information between number cards, and further promotes model convergence by adding user classification.

[0035] The present invention adds the communication information of the number card user to the model, thereby making the prediction of whether the user is off-network more accurate; at the same time, the addition of the number card attribution cross entropy loss accelerates the convergence of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a flowchart of an embodiment of the present invention;

[0037] Figure 2 1 is a diagram showing the relationship between number cards in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0039] like Figure 1 As shown, the method for predicting mobile user network churn according to the embodiment of the present invention comprises the following steps:

[0040] S1. Prepare a number-graph of the relationship between SIM cards in a certain area;

[0041] S2, prepare the degree-graph matrix of the relationship between the number cards, that is, how many communications each number card has in the recent period;

[0042] S3. Use the degree matrix of the relationship graph between number cards to normalize the relationship graph between number cards to obtain the adjacency matrix between number cards

[0043] Darea:

[0044] Darea=(degree-graph)^-0.5*(number-graph)*(degree-graph)^-0.5;

[0045] S4. Construct the feature vector number-feature of each number card's network behavior, and record the number-label of the number card, i.e. whether it is off-network, and the number-user of each number card;

[0046] S5. Construct a fully connected neural network user_fully_connect whose number of input neurons is the number-feature dimension of the feature vector and whose number of output neurons is the total number of users;

[0047] S6. Construct a fully connected neural network off_grid_fully_connect with the number of input neurons equal to the dimension of the feature vector number-feature and the number of output neurons equal to 2;

[0048] S7, construct a 2-layer card graph convolution network and set the card convolution parameter N;

[0049] S8. Build a transformer feature extractor.

[0050] S9. Perform inter-card graph convolution feature extraction on the card feature vector to obtain the high-dimensional card feature vector Output-number:

[0051] Output-number-1=(N*item-feature)*Darea

[0052] Output-number=(N*Output-number-1)*Darea

[0053] S10, sending the high-dimensional number card feature vector Output-number to the transformer feature extractor for higher-dimensional feature extraction to obtain number-high;

[0054] S11, send number-high to the fully connected neural network user_fully_connect to obtain the number card prediction attribution label number-user-classifier;

[0055] S12, send number-high to the fully connected neural network off_grid_fully_connect to obtain the prediction label number-off-grid of whether the number card is off the grid;

[0056] S13, send the number-off-grid prediction label and the number-label real label to Get the cross entropy loss loss-off-grid whether the number card is off-grid;

[0057] S14, send the number card predicted belonging label number-user-classifier and the number card real belonging label number-user into Get the card-attribution cross entropy loss loss-user;

[0058] S15. The cross entropy loss loss-off-grid of whether the number card is off-grid and the cross entropy loss loss-user of the number card are jointly trained to obtain a mobile user off-grid prediction model for predicting the mobile user off-grid situation.

[0059] The following examples illustrate:

[0060] like Figure 1 and Figure 2 The relationship diagram between the number cards shown in the figure, the method for predicting mobile user network churn described in this embodiment includes the following steps:

[0061] 1. Prepare a relationship graph between SIM cards in a certain area (if two SIM cards have communicated, it is represented as 1, if there is no communication, it is represented as 0) number-graph

[0062]

[0063]

[0064] 2. Prepare the degree-graph matrix of the relationship between the number cards (i.e. how many communications each number card has had in the recent period of time)

[0065]

[0066] 3. Use the degree matrix of the relationship graph between number cards to normalize the relationship graph between number cards to obtain the adjacency matrix Darea between number cards:

[0067] Darea=(degree-graph)^-0.5*(number-graph)*(degree-graph)^-0.5

[0068] 4. Construct the feature vector number-feature of each number card's network behavior (the feature vector of each number card can be constructed based on the call duration in the past three months, the number of recharges in the past three months, the network duration, the number of shutdowns in the past three months, the account balance in the past three months, etc.) and record the number card label (whether it is off the network) number-label, and the belonging label of each number card (which user each number card belongs to) number-user

[0069] 5. Construct a fully connected neural network user_fully_connect whose number of input neurons is the number-feature dimension of the feature vector and whose number of output neurons is the total number of users

[0070] 6. Construct a fully connected neural network off_grid_fully_connect with the number of input neurons being the number-feature dimension of the feature vector and the number of output neurons being 2 (whether it is off-grid).

[0071] 7. Build a 2-layer card graph convolution network and set the card convolution parameter N

[0072] 8. Build a layer of transformer feature extractor

[0073] 9. Perform inter-card graph convolution feature extraction on the card feature vector to obtain the high-dimensional card feature vector Output-number:

[0074] Output-number-1=(N*item-feature)*Darea

[0075] Output-number=(N*Output-number-1)*Darea

[0076] 10. Send the high-dimensional number card feature vector Output-number to the transformer feature extractor for higher-dimensional feature extraction to obtain number-high

[0077] 11. Send number-high to user_fully_connect to get the number card prediction attribute label number-user-classifier

[0078] 12. Send number-high to off_grid_fully_connect to get the prediction label number-off-grid for whether the number card is off-grid

[0079] 13. Send the number-off-grid prediction label and the number-label real label to Get whether the number card is off-grid cross entropy loss loss-off-grid

[0080] 14. Send the number card predicted belonging label number-user-classifier and the number card real belonging label number-user to Get the card number belonging to the cross entropy loss loss-user

[0081] 15. Combine the cross entropy loss loss-off-grid of whether the number card is off the grid and the cross entropy loss loss-user of the number card for joint training.

[0082] In summary, the embodiment of the present invention constructs a relationship graph through all SIM cards in a certain area, uses the network behavior of SIM cards over a period of time to form feature nodes on the relationship graph, and uses the graph convolution method to extract features on the relationship graph, thereby predicting that a certain SIM card will be offline in the future. This method adds the social relationship of the SIM card user to the SIM card, which is more accurate and effective than simply using the user's network behavior to predict offline.

[0083] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0084] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0085] In another embodiment provided in the present application, a computer program product including instructions is also provided, which, when executed on a computer, enables the computer to execute any of the mobile user network churn prediction methods in the above embodiments.

[0086] It is understandable that the system, device and storage medium provided in the embodiments of the present invention correspond to the method provided in the embodiments of the present invention, and the explanation, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the above methods.

[0087] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.

[0088] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0089] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0090] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting mobile user churn, characterized in that: The following steps are included: S1. Prepare a number-graph of the relationship between SIM cards in a certain area; S2, prepare the degree-graph matrix of the relationship between the number cards, that is, how many communications each number card has in the recent period; S3. Use the degree matrix of the relationship graph between number cards to normalize the relationship graph between number cards and obtain the adjacency matrix Darea between number cards: S4. Construct the feature vector number-feature of each number card's network behavior, and record the number-label of the number card, i.e. whether it is off-network, and the number-user of each number card; S5. Construct a fully connected neural network user_fully_connect whose number of input neurons is the number-feature dimension of the feature vector and whose number of output neurons is the total number of users; S6. Construct a fully connected neural network off_grid_fully_connect with the number of input neurons equal to the dimension of the feature vector number-feature and the number of output neurons equal to 2; S7, construct a 2-layer card graph convolution network and set the card convolution parameter N; S8. Build a transformer feature extractor. S9, performing inter-number card graph convolution feature extraction on the number card feature vector to obtain a high-dimensional number card feature vector Output-number; S10, sending the high-dimensional number card feature vector Output-number to the transformer feature extractor for higher-dimensional feature extraction to obtain number-high; S11, send number-high to the fully connected neural network user_fully_connect to obtain the number card prediction attribution label number-user-classifier; S12, send number-high to the fully connected neural network off_grid_fully_connect to obtain the prediction label number-off-grid of whether the number card is off the grid; S13, inputting the prediction label number-off-grid of whether the number card is off-grid and the real label number-label of whether the number card is off-grid into the loss function L to obtain the cross entropy loss loss-off-grid of whether the number card is off-grid; S14, sending the number card predicted belonging label number-user-classifier and the number card real belonging label number-user into the loss function L to obtain the number card belonging cross entropy loss loss-user; S15. The cross entropy loss loss-off-grid of whether the number card is off-grid and the cross entropy loss loss-user of the number card are jointly trained to obtain a mobile user off-grid prediction model for predicting the mobile user off-grid situation.

2. The method for predicting mobile user churn according to claim 1, characterized in that: The adjacency matrix between number cards Darea: Darea=(degree-graph)^-0.5*(number-graph)*(degree-graph)^-0.

5.

3. The method for predicting mobile user churn according to claim 1, characterized in that: Step S9 includes, Output-number-1=(N*item-feature)*Darea Output-number=(N*Output-number-1)*Darea.

4. The method for predicting mobile user churn according to claim 3, characterized in that: The loss function L is, 5. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 4.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the computer program is executed by the processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 4.

Images