CN115146717A - Recommendation method, device, electronic device and storage medium based on multi-task model - Google Patents

Abstract

The embodiment of the invention provides a recommendation method, a recommendation device, electronic equipment and a storage medium based on a multitask model, which comprise the following steps: acquiring user behavior data from the log data according to the mobile phone number of the user; screening feature data from the user behavior data, wherein the feature data comprises basic features, interactive features and time slice features; adopting the characteristic data to carry out a prediction task aiming at voice use and a prediction task aiming at flow use through a multi-task model to obtain a prediction result; and determining the voice package and the flow package recommended for the user according to the prediction result. In the embodiment of the invention, the two tasks of the flow and the voice can be respectively predicted by using the multi-task model, so that the prediction efficiency is improved, the package of the most suitable user can be more accurately matched, the service is furthest realized, and the user is kept.

Description

Recommendation method, device, electronic device and storage medium based on multi-task model

technical field

The present invention relates to the technical field of online marketing, and in particular, to a recommendation method based on a multi-task model, a recommendation device based on the multi-task model, an electronic device and a computer-readable storage medium.

Background technique

Because the packages selected by users when they access the network often do not match their subsequent usage habits, user stickiness is not high, and users are lost. In addition, with the advent of the 5G era, more and more users need to change their mobile phone cards from 4G to 5G. Therefore, in order to retain more users when users port numbers to the Internet, how to recommend more suitable ones to users? Packages to increase users' trust in products and user stickiness have become an urgent problem to be solved.

SUMMARY OF THE INVENTION

In view of the above problems, the embodiments of the present invention are proposed to provide a multi-task model-based recommendation method, a multi-task model-based recommendation device, an electronic device and A computer-readable storage medium.

In order to solve the above problem, an embodiment of the present invention discloses a recommendation method based on a multi-task model, and the method includes:

Obtain user behavior data from log data according to the user's mobile phone number;

Screening feature data from the user behavior data, where the feature data includes basic features, interaction features, and time slice features;

Using the feature data, a multi-task model is used to perform a prediction task for voice usage and a prediction task for traffic usage to obtain a prediction result;

According to the prediction result, a voice package and a data package recommended for the user are determined.

Optionally, determining the voice package and data package recommended for the user according to the prediction result, including:

If the predicted result is traffic and voice duration, then map the traffic and voice duration to voice packages and traffic packages according to a preset mapping relationship;

According to the voice package and the data package, a basic package is combined to be recommended to the user.

Optionally, the multi-task model includes: two gate networks, multiple expert networks and two Tower networks, the two gate networks are in one-to-one correspondence with the two Tower networks; the use of the feature data , perform prediction tasks for speech usage and prediction tasks for traffic usage through a multi-task model, and obtain prediction results, including:

Process the feature data through each of the gate networks to obtain the weights of multiple expert networks corresponding to each of the gate networks;

Perform feature extraction on the feature data through a plurality of expert networks corresponding to each of the gate networks to obtain a plurality of first features;

According to the weights of multiple expert networks corresponding to each gate network, weighted summation is performed on multiple first features corresponding to each gate network, and the prediction result is obtained through the Tower network corresponding to each gate network.

Optionally, the feature data is processed through each gate network to obtain the weights of multiple expert networks corresponding to each gate network, including:

Perform feature extraction on the feature data through each of the gate networks to obtain the probability that a plurality of the expert networks are selected by each of the gate networks;

Normalize the probability that the plurality of expert networks are selected by each of the gate networks to obtain the weights of the multiple expert networks corresponding to each of the gate networks.

Optionally, the multi-task model is trained as follows:

acquiring feature sample data, where the feature sample data includes basic feature samples, interactive feature samples, and time slice feature samples;

using the feature sample data as the input of the multi-task model;

In the multi-task model, the feature sample data is used to perform a prediction task for voice usage and a prediction task for traffic usage to obtain a prediction result;

According to the prediction result, the multi-task model is trained.

Optionally, the multitasking model includes: two gate networks, multiple expert networks, and two Tower networks, the two gate networks are in one-to-one correspondence with the two Tower networks; In the model, the feature sample data is used to perform prediction tasks for voice usage and prediction tasks for traffic usage to obtain prediction results, including:

Process the feature sample data through each of the gate networks to obtain the weights of a plurality of the expert networks corresponding to each of the gate networks;

Perform feature extraction on the feature sample data through a plurality of expert networks corresponding to each of the gate networks to obtain a plurality of first feature samples;

According to the weights of multiple expert networks corresponding to each gate network, weighted summation is performed on multiple first feature samples corresponding to each gate network, and the Tower network corresponding to each gate network is used to obtain Predicted results of traffic and voice duration.

Optionally, the processing of the feature sample data through each of the gate networks to obtain the weights of the multiple expert networks corresponding to each of the gate networks, including:

Perform feature extraction on the feature sample data through each of the gate networks to obtain the probability that a plurality of the expert networks are selected by each of the gate networks;

Normalize the probability that the plurality of expert networks are selected by each of the gate networks to obtain the weights of the multiple expert networks corresponding to each of the gate networks.

The embodiment of the present invention also discloses a recommendation device based on a multi-task model, the device comprising:

The data acquisition module is used to acquire user behavior data from the log data according to the user's mobile phone number;

a screening feature module, configured to screen feature data from the user behavior data, where the feature data includes basic features, interaction features, and time slice features;

A feature prediction module, configured to use the feature data to perform a prediction task for voice usage and a prediction task for traffic usage through a multi-task model to obtain a prediction result;

A package determination module, configured to determine a voice package and a traffic package recommended for the user according to the prediction result.

Optionally, the package determination module includes:

a package mapping submodule, configured to map the traffic and the voice duration to a voice package and a traffic package according to a preset mapping relationship if the predicted result is traffic and voice duration;

The package recommendation submodule is configured to combine the voice package and the traffic package into a basic package to recommend to the user.

Optionally, the multi-task model includes: two gate networks, multiple expert networks and two Tower networks, the two gate networks are in one-to-one correspondence with the two Tower networks; the feature prediction module includes :

A weight acquisition sub-module, configured to process the feature data through each of the gate networks to obtain the weights of multiple expert networks corresponding to each of the gate networks;

A feature extraction sub-module, configured to perform feature extraction on the feature data through a plurality of expert networks corresponding to each of the gate networks to obtain a plurality of first features;

The prediction result sub-module is used to perform weighted summation of multiple first features corresponding to each gate network according to the weights of multiple expert networks corresponding to each gate network, and pass the Tower corresponding to each gate network. The network obtains the predicted result.

Optionally, the weight acquisition sub-module includes:

obtaining a probability unit for performing feature extraction on the feature data through each of the gate networks to obtain the probability that a plurality of the expert networks are selected by each of the gate networks;

A weight unit is obtained, which is used for normalizing the probability that multiple expert networks are selected by each of the gate networks to obtain the weights of multiple expert networks corresponding to each of the gate networks.

Optionally, the multi-task model is trained as follows:

a sample acquisition module, configured to acquire feature sample data, where the feature sample data includes basic feature samples, interactive feature samples, and time slice feature samples;

a sample input module, configured to use the feature sample data as the input of the multi-task model;

A sample prediction module, configured to use the feature sample data to perform a prediction task for voice usage and a prediction task for traffic usage in the multi-task model to obtain a prediction result;

A model training module, for training the multi-task model according to the prediction result.

Optionally, the multi-task model includes: two gate networks, multiple expert networks and two Tower networks, the two gate networks are in one-to-one correspondence with the two Tower networks; the sample prediction module includes:

a sample weight submodule, configured to process the feature sample data through each of the gate networks to obtain the weights of a plurality of the expert networks corresponding to each of the gate networks;

A sample extraction sub-module, configured to perform feature extraction on the feature sample data respectively through a plurality of expert networks corresponding to each of the gate networks, so as to obtain a plurality of first feature samples;

The sample prediction sub-module is used to perform weighted summation on multiple first feature samples corresponding to each gate network according to the weights of the multiple expert networks corresponding to each gate network, and pass through each gate network Corresponding Tower network to get the prediction results of traffic and voice duration.

Optionally, the sample weight submodule includes:

a sample probability unit, configured to perform feature extraction on the feature sample data through each of the gate networks, so as to obtain the probability that a plurality of the expert networks are selected by each of the gate networks;

The sample weight unit is used for normalizing the probability that the plurality of expert networks are selected by each of the gate networks to obtain the weights of the plurality of expert networks corresponding to each of the gate networks.

The embodiments of the present invention include the following advantages:

In the embodiment of the present invention, user behavior data is obtained from log data according to the user's mobile phone number; feature data is filtered from user behavior data, and the feature data is used to perform prediction tasks for voice usage and prediction tasks for traffic usage through a multi-task model. Predict the task and get the prediction result; according to the prediction result, determine the voice package and data package recommended for the user. In the embodiment of the present invention, by using a multi-task model, the two tasks of traffic and voice can be predicted respectively, which improves the prediction efficiency, and can more accurately match the most suitable user package, so as to achieve maximum service and retain users. .

Description of drawings

1 is a flowchart of steps of a multitask model-based recommendation method provided by an embodiment of the present invention;

2 is a structural diagram of a multi-task model provided by an embodiment of the present invention;

3 is a flowchart of a training method for a multi-task model provided by an embodiment of the present invention;

FIG. 4 is a structural block diagram of a recommendation device based on a multi-task model provided by an embodiment of the present invention.

Detailed ways

In order to make the above objects, features and advantages of the present invention more clearly understood, the present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.

With the advent of the 5G era, more and more users need to replace their mobile phone cards with 5G. In order to maximize service and improve the conversion rate of users' number portability, it is necessary to accurately predict the most suitable package for users, but at present It is not accurate to predict the most suitable package for a user, and the prediction efficiency is low. Therefore, in order to more accurately predict the most suitable package for a user and improve the prediction efficiency, the user can improve the service and improve the conversion rate of the user's number portability.

One of the core concepts of the embodiments of the present invention is to perform data processing through the user behavior data when the user uses the package, and input the user behavior data after data processing into the trained, based on the two tasks of traffic and voice for prediction The multi-task model is used for processing, so as to obtain the prediction result. While improving the prediction efficiency, it can more accurately match the most suitable package for the user, further improve the user experience, and achieve the maximum service and retain the user.

Referring to FIG. 1, it shows a flowchart of steps of a recommendation method based on a multi-task model provided by an embodiment of the present invention. The method may specifically include the following steps:

Step 101, obtaining user behavior data from log data according to the user's mobile phone number;

Exemplarily, log data is stored in the log table, and the relevant behavior information of the user is recorded in the log table. User behavior data can be obtained from the log table according to the user's mobile phone number, and the user behavior data can include user basic information, user behavior data. information and user terminal information, etc. User basic information can include user age, user gender, current province and current package type, etc.; user behavior information can include the number of times the user has purchased gas packs, the type of gas packs purchased, and the user's click behavior, the amount of billing in the past N months, the recent N-month traffic usage, etc.; user terminal information may include the number of terminals used by the user, the terminal brand, and the like.

Step 102, screening feature data from the user behavior data, where the feature data includes basic features, interaction features, and time slice features;

Exemplarily, the specific operation of screening the feature data may be, firstly performing preliminary selection on the user behavior data, and then performing data cleaning on the preliminarily selected user behavior data.

Preliminary selection of user behavior data can be performed by visual analysis of user behavior data, such as preliminary selection of user behavior data through box plots, heat maps, etc., to remove user behaviors with strong correlation, missing values, and serious outliers. Data; utilizes random forests to examine user behavior data to filter out irrelevant data.

Perform data cleaning on the preliminarily selected user behavior data, for example, convert the format of the data, such as the unified conversion of the amount label into yuan; for the data with less serious missing values, use the median and mode according to the actual situation. Complement value processing; use k-means clustering algorithm to reassign outliers; perform id processing for continuous features such as age; cross important features such as age and gender to construct cross features, etc. Specifically, how to filter out the characteristic data can be determined by those skilled in the art according to the actual situation, which is not limited in the present invention.

In this embodiment of the present invention, the feature data may include basic features, interactive features, and time slice features; wherein, the basic features may include user age, gender, etc.; the interactive features may include user behavior on packages or traffic packages, package popularity, etc.; The time slice features may include the previous day, the last three days, the beginning of the month, the middle of the month, and the end of the month, the last month, the last three months, and so on.

Step 103, using the feature data, perform a prediction task for voice usage and a prediction task for traffic usage through a multi-task model, and obtain a prediction result;

In the embodiment of the present invention, the feature data screened by the user behavior data is input into the trained multi-task model to obtain the corresponding prediction result. Among them, the trained multi-task model performs predictive modeling based on speech tasks and traffic tasks respectively. Specifically, which method to use to train the multi-task model can be determined according to the actual situation, which is not limited in the present invention.

Referring to FIG. 2 , it is a structural diagram of a multi-task model provided by an embodiment of the present invention. The multi-task model can include two gate networks (Gate A and Gate B), multiple expert networks (Expert), and a Tower network corresponding to the gate network one-to-one; wherein, each task corresponds to a Gate (gate network) configuration. ), the number of expert networks may be multiple, and specifically, the data of expert networks is determined according to the actual situation, which is not limited in the present invention.

The gate network can process the input feature data to obtain the probability that the expert network corresponding to the gate network is selected, and the feature data is weighted and summed based on multiple expert networks, and output to the corresponding Tower network; the Tower network is used to represent The final output, the expression looks like this:

Among them, g ^k (x) represents the output of Gate, which is a multi-layer perceptron model, which is implemented as a simple linear transformation plus a softmax (flexible maximum transfer function) layer.

Softmax is used to assign a probability value to the outcome of each output classification, representing the likelihood of belonging to each class.

Loss function: Calculate the RMSE (Root Mean Square Error: root mean square error) of the two tasks separately, and then sum it up as the loss function of the entire task.

In an optional embodiment, the multi-task model includes: two gate networks, multiple expert networks, and two Tower networks, the two gate networks are in one-to-one correspondence with the two Tower networks; the The step 103 may include the following sub-steps S11-S13:

Sub-step S11, processing the feature data through each of the gate networks to obtain the weights of multiple expert networks corresponding to each of the gate networks;

In this embodiment of the present invention, the trained multi-task model may include two gate networks, multiple expert networks, and a Tower network corresponding to the gate network one-to-one; in specific implementation, Gate A may be used as a voice task, and Gate B As a traffic task; Gate A can process the input feature data to obtain the weights of multiple expert networks corresponding to the voice task; Gate B can process the input feature data to obtain the weights of multiple expert networks corresponding to the traffic task. Weights. Specifically, Gate A can also be used as a traffic task, and Gate B can also be used as a voice task, and the gate network corresponding to each task can be set according to the actual situation, which is not limited in the present invention.

Sub-step S12, through a plurality of expert networks corresponding to each of the gate networks, respectively perform feature extraction on the feature data to obtain a plurality of first features;

Sub-step S13, according to the weights of the multiple expert networks corresponding to each gate network, perform a weighted summation on the multiple first features corresponding to each gate network, and obtain the obtained data through the Tower network corresponding to each gate network. the predicted results.

In the embodiment of the present invention, each gate network may correspond to multiple expert networks, and multiple expert networks respectively extract the input feature data, thereby obtaining multiple corresponding first features; based on the multiple expert networks corresponding to each gate network The weight of each expert network is weighted and summed for multiple first features corresponding to each gate network, and the prediction result can be obtained through the Tower network corresponding to each gate network. Among them, the number of first features is the same as the number of expert networks.

In an optional embodiment, the sub-step S11 may include: performing feature extraction on the feature data through each of the gate networks, so as to obtain a plurality of the expert networks selected by each of the gate networks The probability that multiple expert networks are selected by each gate network is normalized to obtain the weights of multiple expert networks corresponding to each gate network.

In the embodiment of the present invention, each gate network can perform feature extraction on the input feature data to obtain the second feature of the input feature data; perform softmax processing on the second feature through the gate network to obtain multiple expert networks The probability of being selected by the gate network is normalized to the probability of multiple expert networks corresponding to each gate network, so as to obtain the weight of multiple expert networks corresponding to each gate network. Among them, the size of the probability can be 0.

Step 104: Determine, according to the prediction result, a voice package and a data package recommended for the user.

Exemplarily, different Tower networks are set based on different task prediction requirements. For example, if it is to predict the click-through rate of users, a Tower network for predicting the click-through rate of users can be set, and the clicks of users can be obtained through the Tower network. probability.

In the embodiment of the present invention, a prediction task for voice usage and a prediction task for traffic usage can be performed through the trained multi-task model, and the prediction result can be obtained through the set Tower network; wherein, the prediction result can be performed by setting the Tower network. Adjustment.

In an optional embodiment, the step 104 may include the following sub-steps S21-S22:

Sub-step S21, if the predicted result is traffic and voice duration, then according to the preset mapping relationship, the traffic and voice duration are mapped to a voice package and a traffic package;

Sub-step S22, combining the voice package and the traffic package into a basic package to recommend to the user.

In the embodiment of the present invention, the prediction result may be traffic and voice duration; according to the preset voice package and traffic package, the traffic and voice duration are mapped to the traffic package and the voice package, and a basic package is formed and recommended to the user. For example, the data packages are A1, A2, and A3, and the corresponding traffic is 5G, 10G, and 30G; the voice packages are B1, B2, and B3, and the corresponding voice durations are 30 minutes, 80 minutes, and 150 minutes, respectively. When the predicted traffic is 9G and the voice duration is 70 minutes, the traffic package A2 and the voice package B2 are combined into a basic package and recommended to the corresponding users.

Exemplarily, the method of recommending to the user may be through a text message, a phone call, or a pop-up window of an APP or the like. Specifically, how to recommend the package to the user can be determined by those skilled in the art according to the actual situation, which is not limited in the present invention.

In the embodiment of the present invention, user behavior data is obtained from log data according to the user's mobile phone number; feature data is filtered from user behavior data, and the feature data is used to perform prediction tasks for voice usage and prediction tasks for traffic usage through a multi-task model. Predict the task and get the prediction result; according to the prediction result, determine the voice package and data package recommended for the user. By using a multi-task model, the two tasks of traffic and voice are predicted respectively, which improves the prediction efficiency, and can more accurately match the most suitable user package, achieve maximum service and retain users.

Referring to FIG. 3, a flowchart of a training method of a multi-task model in an embodiment of the present invention is shown, and the training method of a multi-task model includes:

Step 301, acquiring feature sample data, the feature sample data includes basic feature samples, interactive feature samples, and time slice feature samples;

Exemplarily, the feature sample data may include basic feature samples, interactive feature samples, and time slice feature samples, etc. The acquisition of the feature sample data may be filtered through the user's historical package usage data, specifically, how to obtain the feature sample data, Those skilled in the art can decide according to the actual situation, and the present invention is not limited here.

Step 302, using the feature sample data as the input of the multi-task model;

Step 303, in the multi-task model, using the feature sample data to perform a prediction task for voice usage and a prediction task for traffic usage to obtain a prediction result;

In the embodiment of the present invention, the acquired feature sample data may be used as the input of the multi-task model, so as to obtain the prediction results for speech and traffic through the multi-task model. Among them, the multi-task model is modeled based on voice tasks and traffic tasks respectively; the corresponding prediction results can be voice duration and traffic. Specifically, the Tower network can be set based on the predicted requirements of different tasks, so as to obtain and demand corresponding prediction results.

In an optional embodiment, the multi-task model includes: two gate networks, multiple expert networks, and two Tower networks, the two gate networks are in one-to-one correspondence with the two Tower networks; the The step 303 may include the following sub-steps S31-S33:

Sub-step S31, processing the feature sample data through each of the gate networks to obtain the weights of a plurality of the expert networks corresponding to each of the gate networks;

In an optional embodiment, the sub-step S31 may include:

Perform feature extraction on the feature sample data through each of the gate networks to obtain the probability that a plurality of the expert networks are selected by each of the gate networks; The selected probability is normalized to obtain the weights of multiple expert networks corresponding to each gate network.

In this embodiment of the present invention, each gate network can be used to perform feature extraction on the input feature sample data to obtain a second feature sample of the input feature data; perform softmax processing on the second feature sample through the gate network, to obtain the probability that multiple expert networks are selected by each gate network; wherein, the size of the probability can be 0.

The weights of the multiple expert networks corresponding to each gate network are obtained by normalizing the probabilities of the multiple expert networks corresponding to each gate network.

Sub-step S32, through a plurality of expert networks corresponding to each of the gate networks, respectively perform feature extraction on the feature sample data to obtain a plurality of first feature samples;

Sub-step S33, according to the weights of the multiple expert networks corresponding to each of the gate networks, perform a weighted sum of the multiple first feature samples corresponding to each gate network, and pass the Tower corresponding to each of the gate networks. network to get traffic and voice duration predictions;

In the embodiment of the present invention, each gate network may correspond to multiple expert networks, and the expert network is used to perform feature extraction on the input feature sample data to obtain a plurality of corresponding first feature samples. Among them, the number of first feature samples is the same as the number of expert networks.

According to the weights of multiple expert networks corresponding to each gate network, the weighted summation of multiple first feature samples is performed, and the prediction results of traffic and voice duration can be obtained through the Tower network corresponding to each gate network.

Step 304: Train the multi-task model according to the prediction result.

In the embodiment of the present invention, the multi-task model can be trained according to the corresponding prediction result; for example, the trained multi-task model can be screened by using a ten-fold cross-validation method or a five-fold cross-validation method.

In the embodiment of the present invention, in the multi-task model, the multi-task model is trained by using the feature sample data for the prediction task of speech usage and the prediction task for traffic usage, which can improve the robustness of the model, and use different gate networks to Predicting voice tasks and traffic tasks at the same time not only improves the accuracy of prediction, but also improves the efficiency of prediction.

It should be noted that, for the sake of simple description, the method embodiments are described as a series of action combinations, but those skilled in the art should know that the embodiments of the present invention are not limited by the described action sequences, because According to embodiments of the present invention, certain steps may be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

Referring to FIG. 4, it shows a structural block diagram of a recommendation device based on a multi-task model provided by an embodiment of the present invention, which may specifically include the following modules:

A data acquisition module 401, configured to acquire user behavior data from log data according to the user's mobile phone number;

A screening feature module 402, configured to screen feature data from the user behavior data, where the feature data includes basic features, interaction features, and time slice features;

The feature prediction module 403 is configured to use the feature data to perform a prediction task for voice usage and a prediction task for traffic usage through a multi-task model to obtain a prediction result;

The package determination module 404 is configured to determine, according to the prediction result, a voice package and a data package recommended for the user.

In one embodiment, the package determination module includes:

a package mapping submodule, configured to map the traffic and the voice duration to a voice package and a traffic package according to a preset mapping relationship if the predicted result is traffic and voice duration;

The package recommendation submodule is configured to combine the voice package and the traffic package into a basic package to recommend to the user.

In an embodiment, the multi-task model includes: two gate networks, multiple expert networks, and two Tower networks, the two gate networks are in one-to-one correspondence with the two Tower networks; the feature prediction modules, including:

A weight acquisition sub-module, configured to process the feature data through each of the gate networks to obtain the weights of multiple expert networks corresponding to each of the gate networks;

A feature extraction sub-module, configured to perform feature extraction on the feature data through a plurality of expert networks corresponding to each of the gate networks to obtain a plurality of first features;

The prediction result sub-module is used to perform weighted summation of multiple first features corresponding to each gate network according to the weights of multiple expert networks corresponding to each gate network, and pass the Tower corresponding to each gate network. The network obtains the predicted result.

In one embodiment, the weight acquisition sub-module includes:

obtaining a probability unit for performing feature extraction on the feature data through each of the gate networks to obtain the probability that a plurality of the expert networks are selected by each of the gate networks;

A weight unit is obtained, which is used for normalizing the probability that multiple expert networks are selected by each of the gate networks to obtain the weights of multiple expert networks corresponding to each of the gate networks.

In one embodiment, the multi-task model is trained by:

a sample acquisition module, configured to acquire feature sample data, where the feature sample data includes basic feature samples, interactive feature samples, and time slice feature samples;

a sample input module, configured to use the feature sample data as the input of the multi-task model;

A sample prediction module, configured to use the feature sample data to perform a prediction task for voice usage and a prediction task for traffic usage in the multi-task model to obtain a prediction result;

A model training module, for training the multi-task model according to the prediction result.

In an embodiment, the multi-task model includes: two gate networks, multiple expert networks and two Tower networks, the two gate networks are in one-to-one correspondence with the two Tower networks; the sample prediction Modules include:

a sample weight submodule, configured to process the feature sample data through each of the gate networks to obtain the weights of a plurality of the expert networks corresponding to each of the gate networks;

A sample extraction sub-module, configured to perform feature extraction on the feature sample data respectively through a plurality of expert networks corresponding to each of the gate networks, so as to obtain a plurality of first feature samples;

The sample prediction sub-module is used to perform weighted summation on multiple first feature samples corresponding to each gate network according to the weights of the multiple expert networks corresponding to each gate network, and pass through each gate network Corresponding Tower network to get the prediction results of traffic and voice duration.

In one embodiment, the sample weight sub-module includes:

a sample probability unit, configured to perform feature extraction on the feature sample data through each of the gate networks, so as to obtain the probability that a plurality of the expert networks are selected by each of the gate networks;

The sample weight unit is used for normalizing the probability that the plurality of expert networks are selected by each of the gate networks to obtain the weights of the plurality of expert networks corresponding to each of the gate networks.

To sum up, in the embodiment of the present invention, user behavior data is obtained from log data according to the user's mobile phone number; feature data is screened from user behavior data, and feature data is used to perform prediction tasks for voice usage and target data through a multi-task model. According to the prediction task of traffic usage, the prediction result is obtained; according to the prediction result, the voice package and data package recommended for the user are determined. By using a multi-task model, the two tasks of traffic and voice are predicted respectively, which improves the prediction efficiency, and can more accurately match the most suitable user package, achieve maximum service and retain users.

As for the apparatus embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and reference may be made to the partial description of the method embodiment for related parts.

The embodiment of the present invention also provides an electronic device, including:

It includes a processor, a memory, and a computer program stored on the memory and capable of running on the processor, and when the computer program is executed by the processor, each process of the above-mentioned embodiment of the recommendation method based on a multitasking model is implemented, And can achieve the same technical effect, in order to avoid repetition, it is not repeated here.

Embodiments of the present invention further provide a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, each process of the above-mentioned embodiment of the multitask model-based recommendation method is implemented, and The same technical effect can be achieved, and in order to avoid repetition, details are not repeated here.

The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the various embodiments may be referred to each other.

It should be understood by those skilled in the art that the embodiments of the embodiments of the present invention may be provided as a method, an apparatus, or a computer program product. Accordingly, embodiments of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media having computer-usable program code embodied therein, including but not limited to disk storage, CD-ROM, optical storage, and the like.

Embodiments of the present invention are described with reference to flowcharts and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the present invention. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing terminal equipment to produce a machine that causes the instructions to be executed by the processor of the computer or other programmable data processing terminal equipment Means are created for implementing the functions specified in the flow or flows of the flowcharts and/or the blocks or blocks of the block diagrams.

These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing terminal equipment to operate in a particular manner, such that the instructions stored in the computer readable memory result in an article of manufacture comprising instruction means, the The instruction means implement the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing terminal equipment, so that a series of operational steps are performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby executing on the computer or other programmable terminal equipment The instructions executed on the above provide steps for implementing the functions specified in the flowchart or blocks and/or the block or blocks of the block diagrams.

Although preferred embodiments of the embodiments of the present invention have been described, additional changes and modifications to these embodiments may be made by those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the embodiments of the present invention.

Finally, it should also be noted that in this document, relational terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply these entities or that there is any such actual relationship or sequence between operations. Moreover, the terms "comprising", "comprising" or any other variation thereof are intended to encompass non-exclusive inclusion, such that a process, method, article or terminal device comprising a list of elements includes not only those elements, but also a non-exclusive list of elements. other elements, or also include elements inherent to such a process, method, article or terminal equipment. Without further limitation, an element defined by the phrase "comprises a..." does not preclude the presence of additional identical elements in the process, method, article or terminal device comprising said element.

A method for recommending based on a multi-task model and a device for recommending based on a multi-task model provided by the present invention have been described in detail above. Specific examples are used in this paper to illustrate the principles and implementations of the present invention. The description of the embodiment is only used to help understand the method of the present invention and its core idea; meanwhile, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in specific embodiments and application scope. As mentioned above, the contents of this specification should not be construed as limiting the present invention.

Claims (10)

1. A recommendation method based on a multi-task model, wherein the method comprises: Obtain user behavior data from log data according to the user's mobile phone number; Screening feature data from the user behavior data, where the feature data includes basic features, interaction features, and time slice features; Using the feature data, a multi-task model is used to perform a prediction task for voice usage and a prediction task for traffic usage to obtain a prediction result; According to the prediction result, a voice package and a data package recommended for the user are determined. 2. The method according to claim 1, wherein, according to the prediction result, determining the voice package and the traffic package recommended for the user, comprising: If the predicted result is traffic and voice duration, then map the traffic and voice duration to voice packages and traffic packages according to a preset mapping relationship; According to the voice package and the data package, a basic package is combined to be recommended to the user. 3. The method according to claim 1, wherein the multi-task model comprises: two gate networks, multiple expert networks and two Tower networks, the two gate networks and the two Tower networks One-to-one correspondence; the feature data is used to perform the prediction task for voice usage and the prediction task for traffic usage through a multi-task model to obtain prediction results, including: Process the feature data through each of the gate networks to obtain the weights of multiple expert networks corresponding to each of the gate networks; Perform feature extraction on the feature data through a plurality of expert networks corresponding to each of the gate networks to obtain a plurality of first features; According to the weights of multiple expert networks corresponding to each gate network, weighted summation is performed on multiple first features corresponding to each gate network, and the prediction result is obtained through the Tower network corresponding to each gate network. 4. The method according to claim 3, wherein the feature data is processed through each gate network to obtain the weights of multiple expert networks corresponding to each gate network, comprising: Perform feature extraction on the feature data through each of the gate networks to obtain the probability that a plurality of the expert networks are selected by each of the gate networks; Normalize the probability that the plurality of expert networks are selected by each of the gate networks to obtain the weights of the multiple expert networks corresponding to each of the gate networks. 5. The method of claim 1, wherein the multitasking model is trained in the following manner: acquiring feature sample data, where the feature sample data includes basic feature samples, interactive feature samples, and time slice feature samples; using the feature sample data as the input of the multi-task model; In the multi-task model, the feature sample data is used to perform a prediction task for voice usage and a prediction task for traffic usage to obtain a prediction result; According to the prediction result, the multi-task model is trained. 6. The method according to claim 5, wherein the multi-task model comprises: two gate networks, multiple expert networks and two Tower networks, the two gate networks and the two Tower networks One-to-one correspondence; in the multi-task model, the feature sample data is used to perform a prediction task for voice usage and a prediction task for traffic usage to obtain a prediction result, including: Process the feature sample data through each of the gate networks to obtain the weights of a plurality of the expert networks corresponding to each of the gate networks; Perform feature extraction on the feature sample data through a plurality of expert networks corresponding to each of the gate networks to obtain a plurality of first feature samples; According to the weights of multiple expert networks corresponding to each gate network, weighted summation is performed on multiple first feature samples corresponding to each gate network, and the Tower network corresponding to each gate network is used to obtain Predicted results of traffic and voice duration. 7 . The method according to claim 6 , wherein the feature sample data is processed through each of the gate networks to obtain the data of a plurality of the expert networks corresponding to each of the gate networks. 8 . weights, including: Perform feature extraction on the feature sample data through each of the gate networks to obtain the probability that a plurality of the expert networks are selected by each of the gate networks; Normalize the probability that the plurality of expert networks are selected by each of the gate networks to obtain the weights of the multiple expert networks corresponding to each of the gate networks. 8. A recommendation device based on a multi-task model, wherein the device comprises: The data acquisition module is used to acquire user behavior data from the log data according to the user's mobile phone number; a screening feature module, configured to screen feature data from the user behavior data, where the feature data includes basic features, interaction features, and time slice features; A feature prediction module, configured to use the feature data to perform a prediction task for voice usage and a prediction task for traffic usage through a multi-task model to obtain a prediction result; A package determination module, configured to determine a voice package and a traffic package recommended for the user according to the prediction result. 9. An electronic device, comprising: a processor, a memory, and a computer program stored on the memory and capable of running on the processor, the computer program being executed by the processor to achieve the following: The steps of the multi-task model-based recommendation method according to any one of claims 1-7. 10. A computer-readable storage medium, characterized in that, a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the method according to any one of claims 1-7 is implemented. Steps of a recommended approach for a multi-task model.

Images