Call prediction model training method, call prediction method and device

Abstract

The invention relates to a call prediction model training method, and a call prediction method and device. The call prediction model training method comprises the steps of determining call association characteristics of a sample object based on sample object information, call record information and a call result label in sample data, determining a predicted call result of the call record information based on a to-be-trained call prediction model and the call association characteristics, and training the predicted call result. And training a to-be-trained call prediction model according to the predicted call result and the call result label to obtain a target call prediction model capable of being used for actual prediction. According to the method and the device, the call prediction model can learn the relationship between the call time and the call result of different sample objects, so that the call prediction model can accurately predict the probability that the user object gets through the call request within the preset time period, and the call effect is improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a call prediction model training method, a call prediction method and a device. Background Art

[0002] With the development of Internet technology and the advancement of communication equipment, telephone sales are becoming more and more popular. Telemarketing is a sales method that recommends products to users by phone. The probability of users answering the phone varies when calling users at different times. In order to increase the probability of telephone sales, it is crucial to determine the best time to call users. Summary of the invention

[0003] Based on this, it is necessary to provide a call prediction model training method, a call prediction method and a device that can accurately predict the best time for users to make calls in response to the above technical problems.

[0004] In a first aspect, the present application provides a call prediction model training method, the method comprising:

[0005] Determine call correlation features of the sample object based on sample object information, call record information, and call result labels in the sample data, wherein the call correlation features include a correlation relationship between call time and call result;

[0006] Determining a predicted call result of the call record information based on the call prediction model to be trained and the call association features;

[0007] The call prediction model to be trained is trained according to the predicted call result and the call result label to obtain a target call prediction model.

[0008] In a second aspect, the present application provides a call prediction method, the method comprising:

[0009] Obtaining a call prediction request for a target object, wherein the call prediction request carries object information and call record information of the target object, wherein the call record information includes a call time and a call result indicating whether the call is connected;

[0010] Determining a call correlation feature of the target object based on the object information and the call record information, wherein the call correlation feature includes a correlation relationship between call time and call result;

[0011] Based on the target call prediction model and the call-related features, the probability that the target object answers the call request within a preset time period is determined, and the target call prediction model is trained using the method described in the first aspect above.

[0012] In a third aspect, the present application provides a call prediction model training device, the device comprising:

[0013] A feature extraction module, used to determine the call-related features of the sample object based on the sample object information, call record information and call result labels in the sample data, wherein the call-related features include the correlation between the call time and the call result;

[0014] A call result prediction module, used to determine the predicted call result of the call record information based on the call prediction model to be trained and the call association features;

[0015] The model training module is used to train the call prediction model to be trained according to the predicted call result and the call result label to obtain a target call prediction model.

[0016] In a fourth aspect, the present application provides a call prediction device, the device comprising:

[0017] A request acquisition module, used to acquire a call prediction request for a target object, wherein the call prediction request carries object information and call record information of the target object, wherein the call record information includes a call time and a call result indicating whether the call is connected;

[0018] A feature extraction module, configured to determine a call-related feature of the target object based on the object information and the call record information, wherein the call-related feature includes a correlation between a call time and a call result;

[0019] A call result prediction module is used to determine the probability of the target object answering the call request within a preset time period based on a target call prediction model and the call-related features, wherein the target call prediction model is trained using the method described in the first aspect above.

[0020] In a fifth aspect, the present application provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method described in the first aspect or the second aspect when executing the computer program.

[0021] In a sixth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method described in the first aspect or the second aspect are implemented.

[0022] In a seventh aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the method described in the first aspect or the second aspect above.

[0023] The above-mentioned call prediction model training method, call prediction method and device determine the call association features of the sample object based on the sample object information, call record information and call result label in the sample data, determine the predicted call result of the call record information based on the call prediction model to be trained and the call association features, and train the call prediction model to be trained according to the predicted call result and the call result label to obtain a target call prediction model that can be used for actual prediction. It determines the call association features used to characterize the association relationship between the call time and the call result of the sample object from the sample object information, call record information and call result label of the sample object, and trains the model based on the predicted call result and the call result label determined by the call association features, so that the model can learn the relationship between the call time and the call result of different sample objects, thereby improving the prediction accuracy of the model. Then, the target call prediction model obtained after training can accurately predict the probability of any user object connecting the call request within a preset time period, so as to determine the best dialing time for the user object, that is, determine the time period with the highest probability of the user object connecting the call request, thereby realizing the prediction of the best dialing time for the user object, which is conducive to improving the call effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 A flowchart of a call prediction model training method in one embodiment;

[0026] Figure 2 A schematic diagram of a flow chart of a model training step in an embodiment;

[0027] Figure 3 is a flow chart of a feature extraction step in one embodiment;

[0028] Figure 4 A flowchart of a call prediction model training method in another embodiment;

[0029] Figure 5 A flowchart of a call prediction method in one embodiment;

[0030] Figure 6 It is a structural block diagram of a call prediction model training device in one embodiment;

[0031] Figure 7is a structural block diagram of a call prediction device in one embodiment;

[0032] Figure 8 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0034] In related technologies, by analyzing user historical behaviors, we can know that some users have basically the same probability of answering calls at any time. This group of people does not have a dialing opportunity (i.e., a time when the probability of answering calls is higher than other time periods), and they are not sensitive to the dialing time. Some users also have dialing opportunities, and the sensitivity of dialing opportunities (i.e., the degree of increase in the probability of answering calls corresponding to dialing opportunities compared to non-dial opportunities) also varies.

[0035] In order to increase the probability of users answering the call, time is usually used as a feature to predict whether the call will be connected, and then the call is made to the user based on the prediction result. There are two commonly used methods in traditional technology: one is to use the MLP (Multilayer Perceptron) model to make predictions on the time-related features as other features. However, since a large number of users do not have the opportunity to make a call, this method will lead to the underestimation of the importance of time for users who have the opportunity to make a call.

[0036] Another method is to strengthen the expression of time-related features by taking time-related features as wide features (features with a large number of possible values ​​in the field of machine learning and data science) and establishing a wide & deep model (a machine learning model that combines a linear model and a deep neural network). However, this method does not take into account users who are less sensitive to the timing of dialing. Since their connection probability is indeed unrelated to time, it is forcibly assumed that linear correlation will affect the prediction of these users, and ultimately affect the prediction effect of the dialing timing of the overall users.

[0037] On this basis, some people have proposed a two-stage modeling. The first stage: first predict whether there is a time to make a call, that is, divide the users into users who have a time to make a call and users who do not have a time to make a call. The second stage: model the users who have a time to make a call to obtain the connection probability of users at different times. Although this solution can establish an effective timing prediction model and solve the problem of time features being insignificant or too significant. However, how to mark whether the user has a time to make a call becomes a difficult point. Since many users do not have enough historical data, it is impossible to mark them, resulting in a small sample size. In addition, the two-stage modeling requires the deployment of two models, which makes the implementation of the solution relatively complicated.

[0038] Based on this, the embodiment of the present application provides a method for training a call prediction model, which extracts features from the acquired sample data to determine the call-related features of the sample object, and calls the call prediction model to be trained to predict the call record information of the sample object according to the call-related features to obtain the predicted call result, and trains the call prediction model to be trained according to the predicted call result and the call result label in the sample data to obtain a target call prediction model that can be used for actual prediction. It determines the call-related features used to characterize the association between the call time and the call result of the sample object from the sample object information, call record information and call result label of the sample object, and trains the model based on the predicted call result and call result label determined by the call-related features, so that the model can learn the relationship between the call time and the call result of different sample objects, thereby improving the prediction accuracy of the model.

[0039] In one embodiment, the call prediction model training method provided in the present application is executed by an electronic device. The electronic device may be a terminal or a server. Specifically, the terminal may be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The Internet of Things devices may be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server may be implemented as an independent server or a server cluster consisting of multiple servers.

[0040] like Figure 1 As shown, in this embodiment, the call prediction model training method may include the following steps:

[0041] Step 102: Determine the call-related features of the sample object based on the sample object information, call record information, and call result labels in the sample data.

[0042] The sample data is training data for model training. In this embodiment, before model training, the electronic device first needs to obtain the corresponding sample data. Specifically, the sample data may include sample object information, call record information, and call result labels of the sample object. The call result labels can reflect the real behavior response of the sample object and are used to indicate the call result of whether the call record information is connected.

[0043] Among them, the sample object is a user object with call experience, and the sample object information may include portrait information of the sample object, such as the gender, age, occupation, etc. of the sample object; the sample object information may also include other historical behavior information of the sample object, such as the login behavior record and browsing behavior record of the sample object's target APP (Application). The target APP may be an APP with certain specific functions, such as a shopping APP, a credit APP, etc.

[0044] The call record information is used to represent the call records of the sample object, such as all the call record information of the telephone calls with the sample object in the historical period, which at least records the specific time of the call to the sample object. The call result label can be the call result of whether each call record is connected. Specifically, the call time can be the time point of the specific call, such as year, month, day, hour, minute, and second. The call result label is the result of whether the call to the sample object is connected. If it is connected, the corresponding call result label can be set to 1, and if it is not connected, the corresponding call result label can be set to 0. The specific label content can be set according to the actual scenario, and this embodiment does not limit this.

[0045] The call-related features include the association between the call time and the call result of the sample object. Specifically, the call-related features may include features for indicating that the call result of the sample object is related to time, and features for indicating that the call result of the sample object is independent of time. For example, the call record information of the sample object may be statistically related to time based on the preset time period and the call result label, so as to determine the call statistics information of the sample object in the preset time period, that is, to determine the features of the call result of the sample object related to time. The preset time period may be any time period for dividing the dialing time in the call record information, such as year, month, day, hour, minute, and second, based on the time granularity of model training. For example, it may be divided into a morning period and an afternoon period, or into a weekend period and a non-weekend period, etc. The call record information of the sample object may also be statistically independent of time based on the sample object information and the call result label, so as to determine the features of the call result of the sample object that are independent of time.

[0046] In this embodiment, the electronic device can perform feature extraction based on the sample object information, call record information and call result labels in the sample data, so as to determine the call-related features of the sample object, including features of the call results of the sample object that are related to time, and features of the call results of the sample object that are not related to time.

[0047] Step 104: Determine the predicted call result of the call record information based on the call prediction model to be trained and the call-related features.

[0048] The call prediction model to be trained may be a deep neural network that needs to be trained to predict call results. Specifically, the call prediction model to be trained may include an attention network, a gated neural network, or other neural networks with similar functions that can adjust different feature weights, which is not limited in this embodiment.

[0049] The predicted call result is the predicted behavioral response of the sample object to the call record information predicted by the call prediction model to be trained. Specifically, the predicted call result is the probability of the sample object to connect to the call record information predicted by the call prediction model to be trained based on the call-related features of the sample object. Its value is usually between 0 and 1, where 0 means that the call will not be connected and 1 means that the call will be connected.

[0050] In this embodiment, the electronic device can call the call prediction model to be trained based on the call-related features of the sample object, predict the connection probability of the sample object for the call record information, and determine the predicted call result.

[0051] Step 106: Train the call prediction model to be trained according to the predicted call result and the call result label to obtain a target call prediction model.

[0052] Since the call result label can reflect the actual behavioral response of the sample object to the call record information, the predicted call result is the predicted behavioral response of the sample object to the call record information predicted by the call prediction model to be trained based on the call association characteristics of the sample object. Therefore, the difference between the call result label of the call record information and the predicted call result can reflect the accuracy of the call prediction model to be trained. By training the call prediction model to be trained through the call result label and the predicted call result of the call record information, the accuracy of the call prediction model to be trained can be improved. When the model is trained to converge, a target call prediction model that can be used for actual prediction can be obtained. It can be understood that the target call prediction model finally obtained can be used to predict the probability of any user object connecting a call request within a preset time period, so as to determine the best time to call for the user object, that is, the time period with the highest probability of the user object connecting the call request.

[0053] In the above-mentioned call prediction model training method, the call association features of the sample object are determined based on the sample object information, call record information and call result labels in the sample data, the predicted call result of the call record information is determined based on the call prediction model to be trained and the call association features, and the call prediction model to be trained is trained according to the predicted call result and the call result label to obtain a target call prediction model that can be used for actual prediction. It determines the call association features used to characterize the association relationship between the call time and the call result of the sample object from the sample object information, call record information and call result label of the sample object, and trains the model based on the predicted call result and the call result label determined by the call association features, so that the model can learn the relationship between the call time and the call result of different sample objects, thereby improving the prediction accuracy of the model. Then, the target call prediction model obtained after training can be used to predict the probability of any user object connecting the call request within a preset time period, so as to determine the best time to call for the user object, that is, to determine the time period with the highest probability of the user object connecting the call request, which is conducive to improving the call effect.

[0054] In an exemplary embodiment, the sample data may also include the dialing timing sensitivity of the sample object, wherein the dialing timing sensitivity is used to characterize the degree to which the sample object is affected by the dialing time. Specifically, the dialing timing sensitivity may be the variance of the connection rate of the sample object based on each preset time period calculated based on the call record information of the sample object and the call result label. The smaller the corresponding value, the less sensitive the sample object is to time, that is, the less affected by the dialing time, and the larger the value, the more sensitive the sample object is to time, that is, the greater the degree to which it is affected by the dialing time. Usually, for sample objects with sufficient historical performance (that is, the number of call records in the call record information is sufficient), the corresponding dialing timing sensitivity can be marked, so that the call prediction task can be performed on the full amount of samples, and the call record information of this part of the sample objects can be used to accurately estimate the dialing timing sensitivity, so that every sample data can be fully utilized to further improve the generalization and accuracy of the model.

[0055] Specifically, the call timing sensitivity can be calculated based on the following method: for example, if the preset time period is divided into the morning period and the afternoon period based on the time granularity, if the call record information includes the call records of the sample object for one month in history, then the electronic device can determine the connection rate of the sample object in the morning period and the connection rate of the afternoon period in the history of one month based on the records, wherein the connection rate of the morning period is the ratio of the number of call records with the call result label "connected" in the call records with the call time in the morning in the call record information to the number of all call records with the call time in the morning, and similarly, the connection rate of the afternoon period is the ratio of the number of call records with the call result label "connected" in the call records with the call time in the afternoon in the call record information to the number of all call records with the call time in the afternoon. Then, the average connection rate can be calculated by the connection rate of the morning period and the connection rate of the afternoon period, and then the connection rate variance of the sample object can be calculated according to the connection rate of the morning period, the connection rate of the afternoon period and the average connection rate, and the connection rate variance is the call timing sensitivity of the sample object.

[0056] In this embodiment, when the sample data also includes the calling timing sensitivity of the sample object, then in step 106, the call prediction model to be trained is trained according to the predicted call results and the call result labels. Before obtaining the target call prediction model, the above-mentioned call prediction model training method may also include: embedding the sample object information to obtain the object characteristics of the sample object information; based on the call prediction model to be trained and the object characteristics, determining the predicted calling timing sensitivity for the sample object.

[0057] Among them, the predicted dialing timing sensitivity can be the call prediction model to be trained based on the object features of the sample object, and the predicted dialing timing sensitivity of the sample object. Specifically, the object features can be obtained by embedding processing based on the sample object information of the sample object. Embedding is essentially the aggregation and decoupling of information. The embedding processing in this embodiment can specifically be the vectorization or vector representation of the sample object information, that is, by vectorizing the sample object information, the corresponding embedding vector is generated, that is, the corresponding object features based on the vector representation are obtained.

[0058] The call prediction model to be trained may also include a first MLP (Multi-Layer Perceptron) and a second MLP, wherein the first MLP is used to perform nonlinear processing on the object features of the sample object, including but not limited to feature intersection, feature activation, and feature fusion processing, etc., to output the first fused features of the object features. The second MLP predicts the predicted dialing timing sensitivity of the sample object based on the first fused features output by the first MLP. Therefore, the electronic device performs nonlinear processing on the object features of the sample object by calling the first MLP in the call prediction model to be trained to output the first fused features of the object features, and then calls the second MLP for prediction based on the first fused features output by the first MLP, and outputs the predicted dialing timing sensitivity of the sample object.

[0059] Then in step 106, the call prediction model to be trained is trained according to the predicted call results and call result labels to obtain a target call prediction model, which may specifically include: according to the predicted call results and call result labels, as well as the calling timing sensitivity and the predicted calling timing sensitivity, the call prediction model to be trained is trained to obtain a target call prediction model.

[0060] Specifically, in an exemplary embodiment, Figure 2 As shown, according to the predicted call result and the call result label, as well as the dialing timing sensitivity and the predicted dialing timing sensitivity, the call prediction model to be trained is trained to obtain the target call prediction model, which may specifically include:

[0061] Step 202, calculating a first loss based on a preset first loss function, a predicted call result, and a call result label.

[0062] Among them, the first loss is the difference between the predicted call result of the call record information calculated based on the first loss function and the call result label. Since the call result label can reflect the actual behavioral response of the sample object to the call record information, and the predicted call result is the predicted behavioral response of the sample object to the call record information predicted by the call prediction model to be trained, the difference between the predicted call result of the call record information and the call result label can reflect the accuracy of the call prediction model to be trained in predicting the call result. Based on this, the electronic device can calculate the first loss based on the preset first loss function, the predicted call result and the call result label. The lower the first loss, the closer the predicted call result of the model is to the call result label, that is, the higher the accuracy of the model.

[0063] Specifically, the first loss function can adopt the binary cross-entropy loss function (also known as the logarithmic loss function). The binary cross-entropy loss function measures the difference between the probability distribution predicted by the model and the probability distribution of the true label. Specifically, it calculates the logarithmic difference between the predicted probability and the true label. The greater the difference, the higher the loss value.

[0064] Step 204 , calculating a second loss based on a preset second loss function, the dialing timing sensitivity, and the predicted dialing timing sensitivity.

[0065] Among them, the second loss is the difference between the predicted dialing timing sensitivity and the dialing timing sensitivity calculated based on the second loss function. Since the dialing timing sensitivity is the label information calculated based on the call record information of the sample object, it can reflect the real dialing timing sensitivity of the sample object, and the predicted dialing timing sensitivity is predicted based on the call record information of the sample object. Therefore, the difference between the predicted dialing timing sensitivity and the dialing timing sensitivity can reflect the accuracy of the call prediction model to be trained in predicting the dialing timing sensitivity. Based on this, the electronic device can calculate the second loss based on the preset second loss function as well as the dialing timing sensitivity and the predicted dialing timing sensitivity. Specifically, the second loss function can adopt the mean square error loss function, and the second loss is the corresponding mean square error loss, which is used to measure the mean square error between the dialing timing sensitivity predicted by the model and the real dialing timing sensitivity. The smaller its value is, the smaller the difference between the predicted value and the true value of the model is, and the better the performance of the model is.

[0066] Step 206, calculating the total loss according to the first loss and the second loss.

[0067] The total loss is the overall loss calculated by the electronic device based on the first loss and the second loss for training the call prediction model to be trained. Specifically, if the first loss is loss1 and the second loss is loss2, the overall loss L=loss1+w×loss2, where w is a hyperparameter that determines the weight of the second loss loss2 in the overall loss. Its specific value needs to be tried during training based on the training effect to determine the best value.

[0068] Step 208: Train the call prediction model to be trained according to the total loss to obtain a target call prediction model.

[0069] Specifically, the electronic device trains the call prediction model to be trained by performing back propagation in the call prediction model to be trained using the total loss until the total loss converges, indicating that the model converges, i.e., the model training is stable, thereby obtaining a trained target call prediction model that can be used for actual prediction.

[0070] In the above embodiment, when the sample data also includes the dialing timing sensitivity of the sample object, the first loss is calculated based on the preset first loss function, the predicted call result and the call result label, and the second loss is calculated based on the preset second loss function, the dialing timing sensitivity and the predicted dialing timing sensitivity. The total loss is determined based on the first loss and the second loss, and the call prediction model to be trained is trained based on the total loss to obtain the target call prediction model. The total loss is calculated through the predicted dialing timing sensitivity and the annotated dialing timing sensitivity of the sample object, combined with the predicted call result and the call result label of the call record information, so as to train the model based on the total loss. Since the dialing timing sensitivity is used to characterize the degree to which the sample object is affected by the dialing time, the model is trained by combining the loss of the dialing timing sensitivity so that the model can learn to recognize the sensitivity to the dialing timing, so as to further improve the accuracy of the model prediction.

[0071] It is understandable that when there is no call timing sensitivity label for a sample object in a certain sample data, that is, for sample data of a sample object without sufficient historical performance (i.e., the number of call records in the call record information is small), the first loss is calculated based on the preset first loss function and the call result label and predicted call result of the call record information, and the first loss is used for back propagation to train the call prediction model to be trained until the first loss converges, completing the training process. This is to achieve the purpose of fully utilizing the full amount of sample data to train the model and improve the generalization ability of the model. In an exemplary embodiment, Figure 3 As shown, in step 102, based on the sample object information, call record information and call result label in the sample data, the call correlation feature of the sample object is determined, which may specifically include:

[0072] Step 302: extract time-independent features from the sample object information, call record information, and call result labels to obtain non-time-related call features.

[0073] The non-time-related call feature is used to include features that the call result of the sample object is independent of the call time. Specifically, the electronic device extracts time-independent features from the sample object information, call record information, and call result labels to obtain the corresponding non-time-related call feature.

[0074] Exemplarily, the electronic device may perform embedding processing on the sample object information of the sample object to obtain the object features of the sample object, and perform call feature extraction on the call record information and call result label of the sample object to obtain the call record features of the sample object, and then concatenate the call record features and the object features to obtain non-time-related call features.

[0075] Specifically, the object feature can be an embedding vector of the sample object obtained by embedding the sample object information of the sample object. For example, if the sample object information of the sample object includes the sample object's gender (such as male), age (such as 30 years old), and occupation information (such as a foreign company employee), then by vectorizing it through embedding, the corresponding embedding vector can be generated, thereby obtaining the object feature of the sample object.

[0076] The call record feature can be a feature re-expression of the call record information based on the call result tag. Specifically, the electronic device can perform time-independent statistics based on the call result tag of the call record information of the sample object, such as the total number of call records in the call record information (i.e., the total number of calls made to the sample object recorded in the call record information), the number of connections, the connection probability, etc., and by embedding the statistical information that is independent of time, the historical call features of the corresponding call record information are obtained.

[0077] Since the object feature is an embedding vector obtained after embedding based on the sample object information and is used to represent the sample object information, the historical call feature is used to represent the time-independent embedding vector of the corresponding call record information, and the sample object information is the portrait information of the sample object, and the call record information is the historical call record of the sample object, both of which are time-independent. Therefore, the electronic device splices the embedding vectors of these time-independent information (i.e., the object feature of the sample object information and the historical call feature of the call record information) to obtain a spliced ​​fusion vector, which is the non-time-related call feature of the sample object.

[0078] Step 304 , based on the preset time period and the call result label, extract the call time feature of the call record information of the sample object to obtain the time-related call feature of the sample object.

[0079] Among them, the call time feature refers to the feature information related to the call time, and the time-related call feature refers to all the features extracted to characterize the call results of the sample object and related to the call time, that is, the re-expression of the time-related feature information of the sample object. Specifically, based on the preset time period and call result label, the time-related call feature information can be extracted from the call record information of the sample object, and embedded based on the extracted time-related call feature information to obtain the corresponding time-related call feature.

[0080] Specifically, the call feature information related to time can be determined based on the call records corresponding to the preset time period in the call record information. For example, if the preset time period includes the morning time period and the afternoon time period, the call records with the call time of the morning time period and the call records with the call time of the afternoon time period can be extracted from the call record information respectively based on the dialing time of the call records in the call record information, and the corresponding call time feature information corresponding to the morning time period, such as the number of dialing, the number of connection, and the connection rate, can be counted based on the call records in the morning time period, and the corresponding call time feature information corresponding to the afternoon time period, such as the number of dialing, the number of connection, and the connection rate, can be counted based on the call records in the afternoon time period. Then, the call time feature information corresponding to the morning time period is embedded to obtain the time-related call feature corresponding to the morning time period. The call time feature information corresponding to the afternoon time period is embedded to obtain the time-related call feature corresponding to the afternoon time period.

[0081] It can be understood that the target call prediction model obtained after training can make corresponding predictions based on the preset time period when extracting call time features. For example, if the preset time period when extracting call time features includes the morning period and the afternoon period, the target call prediction model obtained after training can predict the connection probability of the corresponding morning period and the connection probability of the corresponding afternoon period; if the preset time period when extracting call time features includes the weekend period and the non-weekend period, the target call prediction model obtained after training can predict the connection probability of the corresponding weekend period and the connection probability of the corresponding non-weekend period.

[0082] Step 306 , concatenate the non-time-related call features and the time-related call features of the sample object to obtain the full features of the sample object.

[0083] The full feature is used to represent the fusion vector of all information of the sample object, that is, the vector expression of all information of the sample object. Therefore, the electronic device can obtain a spliced ​​fusion vector by splicing the time-related call features and the non-time-related call features of the sample object, and the fusion vector is the full feature of the sample object.

[0084] Step 308: determine the non-time-related call features, the time-related call features, and the full features as the call-related features corresponding to the sample object.

[0085] Specifically, the non-time-related call features, time-related call features and full features extracted above can be used together as call-related features corresponding to the sample object.

[0086] In the above embodiment, the electronic device extracts time-independent features from the sample object information, call record information, and call result labels to obtain non-time-related call features, extracts call time features from the call record information of the sample object based on the preset time period and call result labels to obtain time-related call features of the sample object, concatenates the non-time-related call features and time-related call features of the sample object to obtain the full features of the sample object, and determines the non-time-related call features, time-related call features, and full features as call-related features corresponding to the sample object. It extracts non-time-related call features, time-related call features, and full features from the input sample data, and then trains the model, so that the model can learn the importance of different features to calls of different users, so as to further improve the accuracy of model prediction.

[0087] In an exemplary embodiment, the call prediction model to be trained includes an attention network. In step 104, based on the call prediction model to be trained and the call-related features, the predicted call result of the call record information is determined, which may specifically include: mapping the object features of the sample object to the query vector of the attention network; mapping the non-time-related call features, the full features and the time-related call features to the key vector and value vector of the attention network; determining the weights between the query vector and the non-time-related call features, the full features and the time-related call features in the key vector; performing weighted summation according to the weights and the corresponding value vectors to obtain the predicted call result of the call record information.

[0088] Among them, the attention network, or Attention, is a deep learning architecture that allows the model to focus on certain parts of the input data when processing data. The core idea of ​​Attention is that when generating output, the model can dynamically select the part of the input sequence that is most relevant to the current task. This selection is achieved by calculating the correlation between each element in the input sequence and the current output element.

[0089] In the attention mechanism, query, key and value are the core of the attention mechanism, and they play a key role in calculating the attention weight. Among them, query (Query, referred to as Q) is the element that the model is currently processing. For example, in this embodiment, it is the object feature of the sample object to be predicted. Key (Key, referred to as K) is the element in the input sequence, which is compared with the query to determine the attention weight. Value (Value, referred to as V) is the information associated with the key, which will be weighted and summed according to the attention weight to generate the final output. In sequence-to-sequence tasks, the value can be the same sequence as the key.

[0090] In this embodiment, by mapping the object features of the sample object to the query vector of the attention network, the non-time-related call features, full features and time-related call features of the sample object are mapped to the key vector and value vector of the attention network, that is, the value vector is the same as the key vector. This allows Attention to assign different weights to each feature, which determines the degree of attention the model pays to the input, so as to achieve adaptive selection of weights based on the input, thereby improving the accuracy of the output.

[0091] Specifically, the query is determined by calculating the similarity between the query vector and all key vectors. First, the query Q is dot-producted with all keys K (including non-time-related call features, full features, and time-related call features) to obtain a similarity score, which can be implemented by matrix multiplication. In order to prevent the dot product from being too large, the similarity score is usually divided by a scaling factor, which can be the square root of the dimension of the key vector. The similarity score is then normalized by the softmax function to obtain the attention weight of each key, and the sum of these weights is 1. The obtained attention weight is used to perform a weighted summation on the value V to obtain the final output, that is, the predicted call result of the call record information, which is usually a probability value.

[0092] Specifically, Figure 4 As shown in the figure, the call prediction model to be trained can be implemented based on an end-to-end model, including a first MLP, a second MLP, a third MLP, a fourth MLP and an attention network, wherein the attention network is implemented based on Attention. In the model training process, call-related features of the sample object can be first obtained based on the sample data, including non-time-related call features, time-related call features and full features. The specific feature acquisition process can be referred to above Figure 3 It is understandable that if the sample object has sufficient historical performance (ie, the number of call record information in the sample data of the sample object is sufficient), the call timing sensitivity corresponding to the sample object can also be marked based on the sample data.

[0093] Then, the object features of the sample object can be subjected to nonlinear fusion processing by the first MLP. Specifically, the object features of the sample object are taken as the input of the first MLP, and the first MLP performs feature crossover, feature activation, and feature fusion processing, thereby obtaining fusion features of the object features corresponding to the sample object, which are referred to as first fusion features in this embodiment.

[0094] The non-time-related call features are subjected to non-linear fusion processing by the third MLP. Specifically, the non-time-related call features are used as the input of the third MLP, and the third MLP performs feature crossover, feature activation, and feature fusion processing, thereby obtaining fusion features of the non-time-related call features, which are referred to as second fusion features in this embodiment.

[0095] The fourth MLP performs nonlinear fusion processing on all features. Specifically, the fourth MLP takes all features of the sample object as input, and performs feature crossover, feature activation, and feature fusion processing, thereby obtaining fusion features of all features, which are referred to as third fusion features in this embodiment.

[0096] Then, the first fused feature of the object feature is mapped to the query vector of the attention network, that is, the first fused feature is used as the Query of Attention, and the second fused feature, the third fused feature and the time-related call feature are mapped to the key vector (i.e., the Key of Attention) and the value vector (i.e., the Value of Attention) of the attention network. In this embodiment, Key=Value=the feature sequence composed of the second fused feature, the third fused feature and the time-related call feature.

[0097] And determine the first weight between the query vector and the second fused feature in the key vector, determine the second weight between the query vector and the third fused feature, and determine the third weight between the query vector and the time-related call feature in the key vector. Among them, the first weight, the second weight and the third weight can be determined based on the corresponding similarity. Specifically, by calculating the similarity between the Query and the Key at each position (which can be calculated based on dot product, concatenation, perceptron, etc.) to obtain the weight of the Query and the Key at each position, that is, the weight of the Key at each position is obtained, such as obtaining the first weight of the second fused feature based on the non-time-related call feature, obtaining the second weight of the third fused feature based on the full feature, and obtaining the third weight based on the time-related call feature.

[0098] Then, based on the query vector, the first weight, the second weight, the third weight and the corresponding value vector are weighted and summed to obtain the predicted call result of the call record information. Specifically, the weight and Value at the corresponding position are weighted and summed to obtain the attention value, and finally the attention value is normalized to obtain the predicted call result of the call record information, that is, the predicted probability of the sample object to connect to the call record information, which is a value between 0 and 1.

[0099] It is understandable that the model can be trained one by one for each specific call record in the call record information, and when the model is trained based on a specific call record, the corresponding call record information is the call record before the call record, that is, the dialing time recorded in the call record information is earlier than the dialing time of the specific call record currently being trained. The predicted call result is the predicted result of the specific call record currently being trained.

[0100] Furthermore, the first loss, i.e., loss1, can be calculated based on the call result label and predicted call result of the call record information. And the second MLP can predict the predicted dialing timing sensitivity of the sample object based on the output of the first MLP. Then, the second loss, i.e., loss2, is calculated based on the corresponding labeled dialing timing sensitivity and predicted dialing timing sensitivity. Furthermore, the total loss can be calculated based on loss1 and loss2, and finally the total loss is used for back propagation to train the call prediction model to be trained until the total loss converges to obtain the target call prediction model.

[0101] The above model training method makes full use of all samples and some samples that can be labeled with dialing timing sensitivity, making the training data samples more sufficient, thereby improving the generalization ability of the model. At the same time, the end-to-end model is also easy to maintain and update, taking up less time and resources. The attention mechanism realizes the adaptive weight selection of time-related call features, non-time-related call features and all features, reflecting the difference in the importance of time-related call features for users with different dialing timing sensitivities, making the overall estimation more accurate. The user's dialing timing sensitivity is learned through additional tasks, and the attention query is strengthened to recognize the user's dialing timing sensitivity according to the object characteristics, so that the adaptive selection of attention is more accurate than when there is no additional task, making the overall prediction effect of the model more accurate.

[0102] In an exemplary embodiment, Figure 4 The attention network in the method can also be replaced by a gated neural network (Gate), and the weights of the second fusion feature, the third fusion feature and the time-related call feature are generated by the gated neural network, thereby achieving the diversification of the call prediction model to be trained.

[0103] Specifically, the weight of the time-related call feature can be generated based on the gated neural network, and the weight of the second fusion feature corresponding to the non-time-related call feature output by the third MLP can be generated, and the weight of the third fusion feature corresponding to the full feature output by the fourth MLP can be generated. The sum of the three weights is 1. For example, if the weights include w1, w2, and w3, then w1+w2+w3=1. Then, w1×time-related call feature+w2×second fusion feature+w3×third fusion feature is used to obtain a vector after weight selection, and finally normalization processing (such as normalization based on sigmoid) is performed to obtain the predicted call result of the call record information, which is a value between 0 and 1.

[0104] In an exemplary embodiment, Figure 5 As shown, the present application also provides a call prediction method, which is executed by an electronic device and may specifically include:

[0105] Step 502: Obtain a call prediction request for the target object.

[0106] The call prediction request carries the target object's object information and call record information. The call record information includes the specific time of calling the target object and the call result of whether the call is connected. Specifically, the call prediction request is an instruction or command for predicting the probability of the target object answering the call request in a preset time period. The target object is any user object that needs to be predicted at present. The object information may include the target object's portrait information, such as the target object's gender, age, occupation, etc.; the object information may also include other historical behaviors of the target object, such as the target object's login behavior record and browsing behavior record of the APP.

[0107] The call record information includes the historical call records of the target object, such as all the call record information of the telephone calls made with the target object in the historical period. For example, the call record information includes but is not limited to the time when the staff called the target object each time in the historical period, the call results of whether the target object answered each call, etc.

[0108] In this embodiment, when it is necessary to predict the probability of a certain user object answering a call request, the electronic device first needs to obtain a call prediction request including the object information and call record information of the user object.

[0109] Step 504: Determine the call-related features of the target object based on the object information and the call record information.

[0110] Among them, the call-related features include the correlation between the call time and the call result of the target object. Specifically, the call-related features may include features that the call result of the target object is related to time, and features that the call result of the target object is not related to time. For example, by performing statistics on the call record information of the target object based on a preset time period, the call statistics information of the target object in the preset time period can be determined, that is, the features that the call result of the target object is related to time can be determined. The preset time period can be any time period that divides the dialing time in the call record information based on a certain time granularity, such as year, month, day, hour, minute, and second, such as divided into a morning period and an afternoon period, or divided into a weekend period and a non-weekend period, etc. For example, statistics that are not related to time can also be performed based on the object information and call record information of the target object, so as to determine the features that the call result of the target object is not related to time.

[0111] In this embodiment, the electronic device can determine the call correlation feature corresponding to the target object based on the object information and call record information of the target object. The specific process of obtaining the call correlation feature can refer to the following. Figure 3 The embodiment shown is not described in detail in this embodiment.

[0112] Step 506, based on the target call prediction model and the call-related features, determine the probability that the target object will answer the call request within a preset time period.

[0113] Among them, the target call prediction model can be adopted as above Figures 1 to 4 In this embodiment, the target call prediction model can predict the probability of the target object answering the call request within a preset time period based on the call-related features of the target object.

[0114] In the above call information prediction method, by obtaining a call prediction request for the target object, the call-related features of the target object are determined based on the object information and call record information, and the probability of the target object answering the call request within a preset time period is determined based on the target call prediction model and the call-related features, thereby predicting the best time to call the target object, that is, determining the time period with the highest probability of the target object answering the call request, which is conducive to improving the call effect. In addition, since the target call prediction model is trained based on the above method, the overall prediction of the model is more accurate.

[0115] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0116] Based on the same inventive concept, the embodiment of the present application also provides a call prediction model training device for implementing the call prediction model training method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more call prediction model training device embodiments provided below can refer to the limitations of the call prediction model training method above, and will not be repeated here.

[0117] In an exemplary embodiment, Figure 6 As shown, a call prediction model training device is provided, including: a feature extraction module 602, a call result prediction module 604 and a model training module 606, wherein:

[0118] A feature extraction module 602, configured to determine a call-related feature of the sample object based on the sample object information, call record information, and call result label in the sample data, wherein the call-related feature includes a correlation between a call time and a call result;

[0119] A call result prediction module 604, configured to determine a predicted call result of the call record information based on the call prediction model to be trained and the call association features;

[0120] The model training module 606 is used to train the call prediction model to be trained according to the predicted call result and the call result label to obtain a target call prediction model.

[0121] In an exemplary embodiment, the sample data also includes the dialing timing sensitivity of the sample object; before the call prediction model to be trained is trained according to the predicted call result and the call result label to obtain the target call prediction model, the method also includes: embedding the sample object information to obtain the object characteristics of the sample object information; based on the call prediction model to be trained and the object characteristics, determining the predicted dialing timing sensitivity for the sample object; the call prediction model to be trained according to the predicted call result and the call result label to obtain the target call prediction model includes: training the call prediction model to be trained according to the predicted call result and the call result label, as well as the dialing timing sensitivity and the predicted dialing timing sensitivity to obtain the target call prediction model.

[0122] In an exemplary embodiment, the call prediction model to be trained is trained according to the predicted call result and the call result label, as well as the dialing timing sensitivity and the predicted dialing timing sensitivity to obtain a target call prediction model, including: calculating a first loss based on a preset first loss function, the predicted call result and the call result label; calculating a second loss based on a preset second loss function, the dialing timing sensitivity and the predicted dialing timing sensitivity; determining a total loss based on the first loss and the second loss, and training the call prediction model to be trained according to the total loss to obtain a target call prediction model.

[0123] In an exemplary embodiment, the call-related features of the sample object are determined based on the sample object information, call record information and call result labels in the sample data, including: performing time-independent feature extraction on the sample object information, the call record information and the call result labels to obtain non-time-related call features; performing call time feature extraction on the call record information of the sample object based on a preset time period and the call result labels to obtain time-related call features of the sample object; concatenating the non-time-related call features and the time-related call features of the sample object to obtain the full features of the sample object; and determining the non-time-related call features, the time-related call features and the full features as the call-related features corresponding to the sample object.

[0124] In an exemplary embodiment, the extracting time-independent features of the sample object information, the call record information and the call result label to obtain the non-time-related call features includes: embedding the sample object information to obtain the object features of the sample object; extracting call features of the call record information and the call result label to obtain the call record features of the sample object; and concatenating the call record features and the object features to obtain the non-time-related call features.

[0125] In an exemplary embodiment, the call prediction model to be trained includes an attention network; the method of determining the predicted call result of the call record information based on the call prediction model to be trained and the call-related features includes: mapping the object features of the sample object into a query vector of the attention network; mapping the non-time-related call features, the full features and the time-related call features into a key vector and a value vector of the attention network; determining the weights between the query vector and the non-time-related call features, the full features and the time-related call features in the key vector respectively; and performing weighted summation according to the weights and the corresponding value vectors to obtain the predicted call result of the call record information.

[0126] Each module in the above-mentioned call prediction model training device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0127] In an exemplary embodiment, Figure 7 As shown, a call prediction device is provided, including: a request acquisition module 702, a feature extraction module 704 and a call result prediction module 706, wherein:

[0128] The request acquisition module 702 is used to acquire a call prediction request for a target object, wherein the call prediction request carries object information and call record information of the target object, wherein the call record information includes a call time and a call result of whether the call is connected;

[0129] A feature extraction module 704, configured to determine a call-related feature of the target object based on the object information and the call record information, wherein the call-related feature includes a correlation between a call time and a call result;

[0130] The call result prediction module 706 is used to determine the probability of the target object answering the call request within a preset time period based on the target call prediction model and the call-related features. The target call prediction model adopts the following method: Figures 1 to 4 The method described is trained.

[0131] Each module in the above-mentioned call prediction device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module above.

[0132] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 8 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store sample data or model data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a prediction model training method or a telemarketing information prediction method is implemented.

[0133] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0134] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps of the above method when executing the computer program.

[0135] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0136] In one embodiment, a computer program product is provided, comprising a computer program, which implements the steps of the above method when executed by a processor.

[0137] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0138] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0139] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0140] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A call prediction model training method, characterized in that: The method comprises: Determine call correlation features of the sample object based on sample object information, call record information, and call result labels in the sample data, wherein the call correlation features include a correlation relationship between call time and call result; Determining a predicted call result of the call record information based on the call prediction model to be trained and the call association features; The call prediction model to be trained is trained according to the predicted call result and the call result label to obtain a target call prediction model.

2. The method according to claim 1, characterized in that The sample data also includes the dialing timing sensitivity of the sample object; before the call prediction model to be trained is trained according to the predicted call result and the call result label to obtain the target call prediction model, the method also includes: Performing embedding processing on the sample object information to obtain object features of the sample object information; Determining the predicted dialing timing sensitivity for the sample object based on the call prediction model to be trained and the object characteristics; The step of training the call prediction model to be trained according to the predicted call result and the call result label to obtain a target call prediction model includes: The call prediction model to be trained is trained according to the predicted call result and the call result label, as well as the dialing timing sensitivity and the predicted dialing timing sensitivity to obtain a target call prediction model.

3. The method according to claim 2, characterized in that The step of training the call prediction model to be trained according to the predicted call result and the call result label, as well as the dialing timing sensitivity and the predicted dialing timing sensitivity, to obtain a target call prediction model includes: Calculating a first loss based on a preset first loss function, the predicted call result, and the call result label; Calculating a second loss based on a preset second loss function, the dialing timing sensitivity, and the predicted dialing timing sensitivity; The total loss is determined according to the first loss and the second loss, and the call prediction model to be trained is trained according to the total loss to obtain a target call prediction model.

4. The method according to any one of claims 1 to 3, characterized in that: The determining of the call-related features of the sample object based on the sample object information, call record information and call result label in the sample data includes: Performing time-independent feature extraction on the sample object information, the call record information, and the call result label to obtain non-time-related call features; Based on the preset time period and the call result label, extracting the call time feature of the call record information of the sample object to obtain the time-related call feature of the sample object; splicing the non-time-related call features and the time-related call features of the sample object to obtain the full features of the sample object; The non-time-related call features, the time-related call features and the total features are determined as call-related features corresponding to the sample object.

5. The method according to claim 4, characterized in that The extracting time-independent features from the sample object information, the call record information, and the call result label to obtain non-time-related call features includes: Embedding the sample object information to obtain object features of the sample object; Extracting call features from the call record information and the call result label to obtain call record features of the sample object; The call record feature and the object feature are concatenated to obtain the non-time-related call feature.

6. The method according to claim 5, characterized in that The call prediction model to be trained includes an attention network; and determining the predicted call result of the call record information based on the call prediction model to be trained and the call association feature includes: Mapping the object features of the sample object into a query vector of the attention network; Mapping the non-time-related call features, the full features, and the time-related call features into a key vector and a value vector of the attention network; Determining weights between the query vector and the non-time-related call feature, the full feature, and the time-related call feature in the key vector respectively; A weighted sum is performed according to the weight and the corresponding value vector to obtain a predicted call result of the call record information.

7. A call prediction method, characterized in that: The method comprises: Obtaining a call prediction request for a target object, wherein the call prediction request carries object information and call record information of the target object, wherein the call record information includes a call time and a call result indicating whether the call is connected; Determining a call correlation feature of the target object based on the object information and the call record information, wherein the call correlation feature includes a correlation relationship between call time and call result; Based on the target call prediction model and the call-related features, the probability of the target object answering the call request within a preset time period is determined, and the target call prediction model is trained using the method described in any one of claims 1 to 6.

8. A call prediction model training device, characterized in that: The device comprises: A feature extraction module, used to determine the call-related features of the sample object based on the sample object information, call record information and call result labels in the sample data, wherein the call-related features include the correlation between the call time and the call result; A call result prediction module, used to determine the predicted call result of the call record information based on the call prediction model to be trained and the call association features; The model training module is used to train the call prediction model to be trained according to the predicted call result and the call result label to obtain a target call prediction model.

9. A call prediction device, characterized in that: The device comprises: A request acquisition module, used to acquire a call prediction request for a target object, wherein the call prediction request carries object information and call record information of the target object, wherein the call record information includes a call time and a call result indicating whether the call is connected; A feature extraction module, configured to determine a call-related feature of the target object based on the object information and the call record information, wherein the call-related feature includes a correlation between a call time and a call result; A call result prediction module is used to determine the probability of a target object answering a call request within a preset time period based on a target call prediction model and the call-related features, wherein the target call prediction model is trained using the method described in any one of claims 1 to 6.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

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

12. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images