Model processing method, outbound method and equipment

Abstract

The invention provides a model processing method and device, electronic equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: inputting user data of a sample user into a first model for first processing to obtain a first estimated response rate and a second estimated response rate; wherein the first processing comprises the steps of encoding user features in the user data to obtain a first feature vector, and encoding time features in the user data to obtain a second feature vector; determining a first estimated response rate of the sample user in a plurality of first time periods based on the first feature vector; selecting a first time period from the plurality of first time periods, and determining a second estimated response rate of the sample user in the selected first time period based on the first feature vector and the second feature vector; and adjusting model parameters of the first model based on the first estimated response rate and the second estimated response rate. According to the invention, accurate prediction of the response rate of the user in different time periods can be realized.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a model processing method, an outbound calling method, a device, an electronic device, a computer-readable storage medium, and a computer program product. Background Art

[0002] In outbound call scenarios, it is often necessary to consider the timing of making outbound calls to users (i.e., the user's preferred timing) in order to improve the user's answer rate and the effectiveness of the outbound calls. In addition, by accurately matching the user's timing preference and making outbound calls to the user at the appropriate time period, an opportunity to communicate with the user can be obtained, thereby achieving the purpose of the outbound call. Therefore, the corresponding prediction model can be trained to achieve accurate prediction of the user's outbound call timing. However, the relevant technology finds it difficult to explicitly consider the causal relationship between the dialing time period and the answer rate during the prediction model training process, and the prediction effect of the trained prediction model is poor. Summary of the invention

[0003] The embodiments of the present application provide a model processing method, an outbound calling method, an apparatus, an electronic device, a computer-readable storage medium, and a computer program product, which can improve the incremental prediction effect of the time period on the willingness to connect.

[0004] The technical solution of the embodiment of the present application is implemented as follows:

[0005] The present application embodiment provides a model processing method, including:

[0006] Inputting user data of the sample user into the first model for first processing to obtain a first estimated response rate and a second estimated response rate; wherein the first processing includes: encoding user features in the user data to obtain a first feature vector, and encoding time features in the user data to obtain a second feature vector, wherein the time feature includes a plurality of first time periods; determining a first estimated response rate of the sample user in the plurality of first time periods based on the first feature vector; selecting a first time period from the plurality of first time periods, and determining a second estimated response rate of the sample user in the selected first time period based on the first feature vector and the second feature vector;

[0007] Model parameters of the first model are adjusted based on the first predicted response rate and the second predicted response rate.

[0008] The embodiment of the present application also provides an outbound calling method, including:

[0009] Setting N second time periods for making outbound calls to the user, where N is a positive integer greater than 1;

[0010] Acquire a user feature from the user data of the user, encode the user feature to obtain a third feature vector, and encode the N second time periods respectively to obtain N fourth feature vectors;

[0011] Based on the third eigenvector and the N fourth eigenvectors, performing prediction processing through a pre-trained first model to obtain an estimated response rate of the user in each of the second time periods;

[0012] Initiate an outbound call to the user during the second time period when the estimated answer rate is the largest.

[0013] The present application provides a model processing device, including:

[0014] A processing module, configured to input user data of a sample user into a first model for first processing to obtain a first estimated response rate and a second estimated response rate; wherein the first processing includes: encoding user features in the user data to obtain a first feature vector, and encoding time features in the user data to obtain a second feature vector, wherein the time feature includes a plurality of first time periods; determining a first estimated response rate of the sample user in the plurality of first time periods based on the first feature vector; selecting a first time period from the plurality of first time periods, and determining a second estimated response rate of the sample user in the selected first time period based on the first feature vector and the second feature vector;

[0015] An adjustment module is used to adjust a model parameter of the first model based on the first estimated response rate and the second estimated response rate.

[0016] The embodiment of the present application also provides an outbound calling device, including:

[0017] A generating module, used to set N second time periods for making outbound calls to the user, where N is a positive integer greater than 1;

[0018] an encoding module, configured to obtain user features from the user data of the user, encode the user features to obtain a third feature vector, and encode the N second time periods respectively to obtain N fourth feature vectors;

[0019] A prediction module, configured to perform prediction processing based on the third eigenvector and the N fourth eigenvectors through a pre-trained first model to obtain an estimated response rate of the user in each of the second time periods;

[0020] An outbound calling module is used to initiate an outbound call to the user during the second time period when the estimated answering rate is the largest.

[0021] An embodiment of the present application provides an electronic device, including:

[0022] A memory for storing executable instructions;

[0023] The processor is used to implement the model processing method or outbound calling method provided in the embodiment of the present application when executing the executable instructions stored in the memory.

[0024] An embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions for implementing the model processing method or outbound calling method provided in the embodiment of the present application when executed by a processor.

[0025] An embodiment of the present application provides a computer program product, including a computer program or computer executable instructions, which are used to implement the model processing method or outbound call method provided in the embodiment of the present application when executed by a processor.

[0026] The embodiments of the present application have the following beneficial effects:

[0027] During the model training process, the first model is called to perform predictive processing on the user data to obtain the first estimated answer rate and the second estimated answer rate. This allows the model to explicitly consider the causal relationship between the calling period and the answer rate, thereby effectively improving the model's prediction effect on the answer rate based on time characteristics; in addition, by combining the first estimated answer rate and the second estimated answer rate, two different estimated answer rates, and adjusting the parameters of the first model, the prediction effect of the trained first model can be effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a schematic diagram of the architecture of the model processing system 100 provided in an embodiment of the present application;

[0029] Figure 2A is a schematic diagram of the structure of an electronic device 500 provided in an embodiment of the present application;

[0030] Figure 2B is a schematic diagram of the structure of an electronic device 600 provided in an embodiment of the present application;

[0031] Figure 3 It is a flow chart of the model processing method provided in the embodiment of the present application;

[0032] Figure 4 It is a flow chart of the model processing method provided in the embodiment of the present application;

[0033] Figure 5 It is a flow chart of the model processing method provided in the embodiment of the present application;

[0034] Figure 6 It is a flow chart of the model processing method provided in the embodiment of the present application;

[0035] Figure 7 It is a flow chart of the model processing method provided in the embodiment of the present application;

[0036] Figure 8 It is a flow chart of the model processing method provided in the embodiment of the present application;

[0037] Fig. 9 It is a flowchart of the outbound calling method provided in the embodiment of the present application;

[0038] Fig.10 It is a schematic diagram of the principle of the model processing method provided in the embodiment of the present application. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0040] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it can be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0041] It is understandable that in the embodiments of the present application, related data such as user information (such as user feature data) is involved. When the embodiments of the present application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards.

[0042] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program with a predetermined function, and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories) or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0043] In the following description, the terms "first\second\..." involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\..." can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0045] Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations.

[0046] 1) Uplift Model: Uplift modeling is a modeling method used in the field of marketing and personalized decision-making to evaluate the impact of interventions (such as coupons, advertising, etc.) on user behavior. Its core is to quantify the incremental impact of interventions on user responses, that is, to evaluate the effectiveness of interventions by comparing the differences in user responses with and without intervention. The purpose of Uplift modeling is to more accurately identify users who have a positive response to specific interventions, thereby optimizing resource allocation and improving the efficiency and effectiveness of marketing activities. In terms of modeling methods, there are two main types of Uplift models: dual models and single models. The dual model models the experimental group users and the control group users separately, and then takes the difference in the predicted results as the estimated uplift value. The single model is suitable for binary classification scenarios, and is modeled by category conversion and reconstruction of the loss function.

[0047] 2) Activation function: Activation function (such as Sigmoid function) is used in the output layer or hidden layer of the neural network. It can map all real numbers to the interval (0,1) to achieve data normalization. In the output layer of regression prediction and binary classification models, Sigmoid function plays an important role.

[0048] 3) Cross-Entropy Loss Function: The cross-entropy loss function is a loss function commonly used in classification problems, especially in neural networks, to measure the difference between the actual output distribution and the model's predicted output distribution. The cross-entropy loss function encourages the model to output a probability distribution rather than a single predicted value, which helps the model learn the correct classification probability.

[0049] 4) Average willingness to connect: Average willingness to connect refers to the answer rate of calls made by users in a historical period. For example, if a user makes N calls in a historical period, the answer rate of these N calls is calculated. If the answer rate is greater than the threshold, the average willingness to connect is 1, otherwise the average willingness to connect is 0.

[0050] In marketing campaign scenarios, it is often necessary to consider personalized marketing timing that suits users (i.e. user preferred timing) in order to improve the effectiveness of the campaign and obtain greater benefit value; accurately hitting the user's timing preferences and doing marketing at the right time can bring opportunities for deeper communication with users, thereby achieving marketing goals.

[0051] During training, related technical solutions generally use time period as one of the features, and combine it with other user features to estimate whether the call is connected. During the estimation, the response rate of the user in different time periods is estimated by inputting different time feature values ​​and other feature values ​​of the user. Then, the response rates in different time periods are subtracted to obtain the increment. The problem of improved response rate brought by the time period is not considered in the model training structure.

[0052] Compared with the related art, the embodiment of the present application expresses the answer rate of a call as the sum of the average willingness to connect and the time period connection increment, which can reflect the incremental effect of the time period on the answer rate and is directly related to the connection increment brought by the system decision variable time period. At the same time, the embodiment of the present application also proposes a method for calculating the average willingness to connect, which can make the average willingness to connect estimate more accurate through an auxiliary loss function.

[0053] That is to say, the embodiments of the present application provide a model processing method, an outbound calling method, an apparatus, an electronic device, a computer-readable storage medium, and a computer program product, which can improve the incremental prediction effect of the time period on the willingness to connect. The electronic device provided in the embodiments of the present application can be implemented as a server, or implemented by a server and a terminal in collaboration. The following is an example of the model processing method provided by the embodiment of the present application being implemented in collaboration by a server and a terminal.

[0054] For example, see Figure 1 , Figure 1 is a schematic diagram of the architecture of the model processing system 100 provided in the embodiment of the present application, in order to realize supporting a model processing application, such as Figure 1 As shown, the model processing system 100 includes: a server 200, a network 300, and a terminal 400. The terminal 400 is connected to the server 200 via the network 300, wherein the network 300 can be a local area network or a wide area network, or a combination of the two; the terminal 400 is a terminal associated with the user, and a client 410 runs on the terminal 400. The client 410 can be various types of clients, such as a dedicated model training client, an online marketing client, and a browser.

[0055] In some embodiments, the server 200 may input the user data of the sample user into a first model (e.g., a time period estimation model) for a first processing to obtain a first estimated response rate and a second estimated response rate, wherein the first processing refers to encoding the user features in the user data to obtain a first feature vector (e.g., a user feature vector of the sample user), and encoding the time features in the user data to obtain a second feature vector (e.g., a time period feature vector), the time features including multiple first time periods, determining the first estimated response rate of the sample user in the multiple first time periods based on the first feature vector, and selecting a first time period from the multiple first time periods, and determining the second estimated response rate of the sample user in the selected first time period based on the first feature vector and the second feature vector; subsequently, the server 200 adjusts the model parameters of the first model based on the first estimated response rate and the second estimated response rate, thereby obtaining the trained first model; finally, the server 200 may send the trained first model to the terminal 400 via the network 300, i.e., the trained first model may be deployed on the client 410 of the terminal 400.

[0056] It should be noted that the technical solution provided by the embodiment of the present application can be applied to various application scenarios, such as network marketing, online advertising, personalized recommendations, etc. For example, the trained first model can be deployed to the marketing client to predict the response rate of customers in different time periods, and automatically make marketing recommendations to users in the time period corresponding to the customer's maximum response rate, thereby achieving the marketing purpose; for example, the trained first model can be applied to the online advertising scenario, and advertising recommendations can be made to customers for the time period preferred by the user, thereby achieving the marketing purpose.

[0057] In addition, it should be noted that Figure 1 The server 200 in the example may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms. The terminal 400 may be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, a car terminal, etc., but is not limited thereto. The terminal 400 and the server 200 may be directly or indirectly connected via wired or wireless communication, which is not limited in the embodiments of the present application.

[0058] The following is a description of the structure of an electronic device for implementing a model processing method provided in an embodiment of the present application. Figure 2A , Figure 2A is a schematic diagram of the structure of an electronic device 500 provided in an embodiment of the present application, Figure 2A The electronic device 500 shown includes: at least one processor 510, a memory 540, and at least one network interface 520. The various components in the electronic device 500 are coupled together via a bus system 530. It is understood that the bus system 530 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 530 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, the bus system 530 is not described in detail. Figure 2A Various buses are labeled as bus system 530 .

[0059] The processor 510 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0060] The memory 540 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard drives, optical drives, etc. The memory 540 may optionally include one or more storage devices that are physically remote from the processor 510.

[0061] The memory 540 includes a volatile memory or a non-volatile memory, and may also include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 540 described in the embodiments of the present application is intended to include any suitable type of memory.

[0062] In some embodiments, memory 540 is capable of storing data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplarily described below.

[0063] Operating system 541, including system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks;

[0064] A network communication module 542, for reaching other computing devices via one or more (wired or wireless) network interfaces 520, exemplary network interfaces 520 include: Bluetooth, Wireless Authentication (WiFi), and Universal Serial Bus (USB);

[0065] In some embodiments, the device provided in the embodiments of the present application can be implemented in software. Figure 2A A model processing device 543 stored in the memory 540 is shown, which can be software in the form of a program and a plug-in, including the following software modules: a processing module 5431 and an adjustment module 5432. These modules are logical and can therefore be arbitrarily combined or further split according to the functions implemented.

[0066] The structure of the electronic device for implementing the outbound calling method provided in the embodiment of the present application is further described below. Taking the electronic device 600 as a server as an example, see Figure 2B , Figure 2B is a schematic diagram of the structure of an electronic device 600 provided in an embodiment of the present application, Figure 2B The electronic device 600 shown includes an outbound call device 643, which can be software in the form of programs and plug-ins, including the following software modules: a generation module 6431, an encoding module 6432, a prediction module 6433 and an outbound call module 6434. The functions of each module will be explained below.

[0067] It should be noted that Figure 2B The electronic device 600 shown includes a processor 610, a network interface 620, a bus system 630, a memory 640, an operating system 641, and a network communication module 642. Figure 2A The corresponding modules included in have the same structure and the same function, and the embodiments of the present application will not be repeated here.

[0068] The model processing method provided in the embodiment of the present application will be specifically described below in combination with the exemplary application and implementation of the server provided in the embodiment of the present application.

[0069] See also Figure 3 , Figure 3 is a flow chart of the model processing method provided in the embodiment of the present application, which will be combined with Figure 3 The steps shown are explained.

[0070] In step 101, user data of sample users is input into a first model for first processing to obtain a first estimated response rate and a second estimated response rate.

[0071] In some embodiments, see Figure 4 , Figure 4is a flow chart of the model processing method provided in the embodiment of the present application, such as Figure 4 As shown, Figure 3 Step 101 shown may be performed by Figure 4 Steps 1011 to 1013 shown are implemented by combining Figure 4 The steps shown are explained.

[0072] In step 1011, the user features in the user data are encoded to obtain a first feature vector, and the time features in the user data are encoded to obtain a second feature vector.

[0073] Here, the first feature vector may refer to a user feature vector of a sample user, and the second feature vector may refer to a time period feature vector.

[0074] It should be noted that the user characteristics and time characteristics in the user data can be collected through the collection system. For example, a certain period of time (such as 2 hours) can be used as the time feature, and the user's age, gender and other characteristics can be used as user characteristics. The type and content of the characteristics are determined according to needs and are not specifically limited here.

[0075] In some embodiments, encoding user features and time features may refer to converting non-numeric data into a numerical form that can be processed by a machine, wherein the encoding method may be common encoding methods such as one-hot encoding, label encoding, binary encoding, frequency encoding, feature hashing, etc.

[0076] For example, taking feature hash coding as an example, first determine the number of hash functions M and the dimension N of the feature space, usually M is much smaller than N; then, define M hash functions for user features and time features respectively, each hash function maps the input feature to an integer between 1 and M; then, for each sample (such as user name and time period), use the defined hash function to hash the user name and time period to obtain the corresponding hash code values ​​respectively; finally, concatenate the obtained hash code values ​​to obtain the corresponding feature vector. For example, if the user name is "user123" and the time feature is "afternoon", the user name is mapped to a hash value of 3 through the hash function, and the time feature is mapped to a hash value of 2. The feature vector obtained by concatenating the user feature and the time feature is [3,2].

[0077] In step 1012, a first estimated response rate of the sample user in a plurality of first time periods is determined based on the first feature vector.

[0078] In some embodiments, different user feature vectors and the first true response rates corresponding to the different user feature vectors are first obtained; then, the correspondence between the different user feature vectors and the first true response rate is determined, which can be obtained through training with a deep learning regression model (such as a convolutional neural network, etc.) or through a simple regression model (such as a linear regression model, etc.), which is not limited here; finally, based on the obtained correspondence, the user feature vector is predicted to obtain the first estimated response rate corresponding to the user feature vector.

[0079] In some embodiments, see Figure 5 , Figure 5 is a flow chart of the model processing method provided in the embodiment of the present application, such as Figure 5 As shown, Figure 4 Step 1012 may be performed by Figure 5 Steps 10121 to 10123 shown are implemented by combining Figure 5 The steps shown are explained.

[0080] In step 10121, a linear transformation is performed on the first eigenvector to obtain a first linear transformation result.

[0081] In some embodiments, a weight matrix is ​​first defined, where each row of the weight matrix represents the weight of the model for a specific feature; then, the weight matrix is ​​initialized, where the weights can be randomly initialized or the weight matrix can be trained using some optimization algorithms (such as gradient descent); finally, the weight matrix is ​​dot-producted with the user feature vector to obtain a first linear conversion result.

[0082] For example, suppose the first eigenvector is [x 1 ,x 2 ,x 3 ], the weight matrix is ​​[[1,2,3][4,5,6]], the first eigenvector is dot-producted with the weight matrix, and the first linear transformation result is [x 1 +2x 2 +3x 3 ,4x 1 +5x 2 +6x 3 ].

[0083] In step 10122, activation processing is performed on the first linear transformation result to obtain a first activation result.

[0084] Here, the first linear transformation result is activated by an activation function, and the activation function may be a common type such as Sigmoid, ReLU, Tanh, Swish, etc., which may be determined according to different application scenarios and is not specifically limited here.

[0085] In some embodiments, a corresponding activation function is determined, and activation processing is performed on the first linear conversion result according to the determined activation function to obtain a first activation result.

[0086] For example, assuming that the Sigmoid activation function is used, after the first linear transformation result is activated, the first activation result obtained will be between 0 and 1; assuming that the ReLU activation function is used, after the first linear transformation result is activated, the first activation result obtained will be a non-negative number. The specific result will vary depending on the activation function.

[0087] In step 10123, the first activation result is used as a first estimated response rate of the sample user in multiple first time periods.

[0088] It should be noted that the first activation result obtained by activating the first linear conversion result through the activation function is the first estimated response rate of the sample user in the multiple first time periods.

[0089] Continuing with the above example, assuming that the Sigmoid activation function is used, the first estimated response rate of the sample users in multiple first time periods is obtained will be between 0 and 1; assuming that the ReLU activation function is used, the first estimated response rate of the sample user in multiple first time periods is Is a non-negative number, and the specific result will vary depending on the activation function.

[0090] In step 1013, a first time period is selected from the plurality of first time periods, and a second estimated response rate of the sample users in the selected first time period is determined based on the first feature vector and the second feature vector.

[0091] Here, a first time period may be randomly selected from a plurality of first time periods, or may be selected from a plurality of first time periods in sequence, and the embodiment of the present application does not make any specific limitation thereto.

[0092] In some embodiments, see Figure 6 , Figure 6 is a flow chart of the model processing method provided in the embodiment of the present application, such as Figure 6 As shown, Figure 4 Step 1013 shown may be performed by Figure 6 Steps 10131 to 10134 shown are implemented by combining Figure 6 The steps shown are explained.

[0093] In step 10131, a linear transformation is performed on the first eigenvector to obtain a first linear transformation result.

[0094] It should be noted that the implementation of step 10131 in the embodiment of the present application is similar to the implementation of the above-mentioned step 10121. For details, please refer to the implementation of the above-mentioned step 10121, which will not be repeated here.

[0095] In step 10132, a linear transformation is performed on the sub-feature vector in the second feature vector corresponding to the selected first time period to obtain a second linear transformation result.

[0096] It should be noted that the implementation of step 10132 in the embodiment of the present application is similar to the implementation of the above-mentioned step 10121. For details, please refer to the implementation of the above-mentioned step 10121, which will not be repeated here.

[0097] In step 10133, activation processing is performed on the first linear transformation result and the second linear transformation result to obtain a second activation result.

[0098] In some embodiments, a corresponding activation function is determined, and activation processing is performed on the first linear transformation result and the second linear transformation result according to the determined activation function to obtain a corresponding second activation result.

[0099] For example, taking the ReLU activation function as an example, assuming that the first linear transformation result is activated by the ReLU activation function to obtain result A1, and the second linear transformation result is activated by the ReLU activation function to obtain result A2, and the results A1 and A2 are combined to obtain the corresponding second activation result (assuming it is denoted as p).

[0100] In step 10134, based on the second activation result and the first estimated response rates of the sample users in the plurality of first time periods, a second estimated response rate of the sample users in the selected first time period is determined.

[0101] It should be noted that the second estimated response rate for the selected first time period is mainly composed of the first estimated response rate and the estimated response rate increment for the selected first time period. By adding the first estimated response rate to the estimated response rate increments for different time periods, the second estimated response rate for the current time period is determined, thereby achieving accurate prediction of the user response rate.

[0102] In some embodiments, see Figure 7 , Figure 7 is a flow chart of the model processing method provided in the embodiment of the present application, such as Figure 7 As shown, Figure 6 Step 10134 shown can be performed by Figure 7 Steps 101341 to 101342 shown are implemented by combining Figure 7 The steps shown are explained.

[0103] In step 101341, the second activation result is converted to obtain a conversion result, and the conversion result is used as the estimated response rate increment of the sample user in the selected first time period.

[0104] It should be noted that, by converting the first eigenvector and the second eigenvector, the influence of different time periods on the user response rate is determined, and then the corresponding relationship between the time period and the response rate increment is established.

[0105] In some embodiments, the above-mentioned step 101341 can be implemented in the following manner: multiply the activation result by a constant to obtain a multiplication result; divide the constant by a set value to obtain a division result; subtract the multiplication result from the division result to obtain the estimated response rate increment of the sample user in the selected first time period.

[0106] For example, assuming that the activation result p is obtained and the constant is a, the activation result p is multiplied by the constant a to obtain the multiplication result a*p. Here, the constant a can be set differently according to different application scenarios. Assuming that the setting value is 2, the constant a is divided by the setting value 2 to obtain the division result a / 2. Finally, the multiplication result a*p is subtracted from the division result a / 2 to obtain the estimated response rate increment of the sample user in the first selected period. is a*pa / 2, where the value range of the estimated response rate increment of the sample users in the first selected time period is (-a / 2, a / 2).

[0107] In step 101342, based on the estimated response rate increment of the sample user in the selected first time period and the first estimated response rates of the sample user in multiple first time periods, a second estimated response rate of the sample user in the selected first time period is determined.

[0108] In some embodiments, the above-mentioned step 101342 can be implemented by adding the estimated response rate increment of the sample user in the selected first time period and the first estimated response rate of the sample user in multiple first time periods to obtain the second estimated response rate of the sample user in the selected first time period.

[0109] For example, the estimated response rate increment of the sample users in the first selected period is and the first estimated response rate of sample users in multiple first time periods Add them together to get the second estimated response rate of the sample users in the selected first period for

[0110] In step 102, model parameters of the first model are adjusted based on the first estimated response rate and the second estimated response rate.

[0111] In some embodiments, a loss function is first defined, and this loss function can comprehensively consider the difference between the first estimated response rate and the first true response rate, as well as the difference between the second estimated response rate and the second true response rate; then, according to the different importance of the first estimated response rate and the second estimated response rate to the final result, a weighted loss function can be considered, so that more weight can be given to the first estimated response rate, or more weight can be given to the second estimated response rate; then, the above loss functions are combined to construct a total loss function, which can be achieved by simply adding or subtracting the two loss functions. If a weighted loss function is used, it is also necessary to consider how to assign weights; finally, after constructing the loss function, an optimization algorithm (such as gradient descent, stochastic gradient descent, etc.) can be used to minimize the total loss function. This helps to adjust the parameters of the model, thereby improving the accuracy of the model prediction.

[0112] In other embodiments, the parameters of the first model are first initialized; then, the training sample is input into the first model, and the first model correspondingly outputs a first estimated response rate and a second estimated response rate. A loss function is used to calculate the difference between the predicted value and the true value, and the corresponding loss value is calculated; then, through back propagation, the gradient of the loss with respect to the model parameters is calculated, and the parameters of the model are updated through an optimizer (such as an Adam optimizer, etc.) to minimize the loss.

[0113] In some embodiments, see Figure 8 , Figure 8 is a flow chart of the model processing method provided in the embodiment of the present application, such as Figure 8 As shown, Figure 3 Step 102 shown may be performed by Figure 8 Steps 1021 to 1024 shown are implemented by combining Figure 8 The steps shown are explained.

[0114] In step 1021, the first estimated response rate and the first true response rate of the sample users in the first time periods are substituted into the first loss function to obtain a first error.

[0115] In some embodiments, the corresponding target loss function is first determined according to the specific task and data (for example, for classification problems, the cross entropy loss function can be used; for regression problems, the mean square error (MSE) or other types of loss functions can be used); then, the first loss is constructed for the data.

[0116] For example, assuming that the first loss function is a cross entropy loss function, the first loss L1 can be constructed using formula (1).

[0117]

[0118] Among them, y i is the first true response rate, is the first estimated response rate, and N is the sample size.

[0119] Assuming that the first loss function is Hinge Loss, the first loss L1 can be constructed using formula (2).

[0120]

[0121] Among them, y i is the first true response rate, is the first estimated response rate.

[0122] In step 1022, the second estimated response rate and the second true response rate of the sample users in the selected first time period are substituted into the second loss function to obtain a second error.

[0123] Here, the type of the second loss function may be the same as or different from the type of the first loss function, and is not specifically limited here.

[0124] It should be noted that the implementation of step 1022 in the embodiment of the present application is similar to the implementation of step 1021 described above. For details, please refer to the implementation of step 1021 described above, which will not be repeated here.

[0125] For example, assuming that the second loss is a cross entropy loss function, the second loss function L2 can be constructed using formula (3).

[0126]

[0127] in, is the second true response rate, is the second estimated response rate, and N is the sample size.

[0128] Assuming that the second loss is Hinge Loss, the second loss function L2 can be constructed using formula (4).

[0129]

[0130] in, is the second true response rate, is the second estimated response rate.

[0131] In step 1023 , a total error is determined based on the first error and the second error.

[0132] In some embodiments, first, a combination strategy of the first loss and the second loss is determined. A simple combination strategy can be used here, such as adding or subtracting the first loss and the second loss, using a weighted sum of weights, or some complex combination strategies can be used, such as minimizing the maximum value, average value, etc., which are not specifically limited here; then, the total loss is constructed according to the combination strategy.

[0133] For example, assuming that the average method is adopted, the total loss can be constructed by formula (5) to obtain the total loss L.

[0134]

[0135] Among them, L1 is the first loss and L2 is the second loss.

[0136] In some embodiments, the above step 1023 can be implemented in the following manner: performing a weighted summation on the first error and the second error, and taking the weighted summation result as the total error.

[0137] It should be noted that you can obtain only the weighting coefficient of the first loss, or only the weighting coefficient of the second loss, or you can obtain the weighting coefficients corresponding to the first loss and the second loss at the same time. It can be determined according to the specific application scenario and is not specifically limited here.

[0138] For example, when only the weighted coefficient w of the first loss is obtained 1 When , the total loss L can be calculated by formula (6).

[0139] L=w 1 *L1+L2 (6)

[0140] Among them, L1 is the first loss and L2 is the second loss.

[0141] For example, when only the weighted coefficient w of the second loss is obtained 2 When , the total loss L can be calculated by formula (7).

[0142] L=L1+w 2 *L2 (7)

[0143] Among them, L1 is the first loss and L2 is the second loss.

[0144] When the weighted coefficient w of the first loss is obtained at the same time 1 and the weighted coefficient w of the second loss 2 When , the total loss L can be calculated by formula (8).

[0145] L=w 1 *L1+w 2 *L2 (8)

[0146] Among them, L1 is the first loss and L2 is the second loss.

[0147] In step 1024 , model parameters of the first model are adjusted based on the total error.

[0148] In some embodiments, the above step 1024 can be implemented in the following manner: performing back propagation based on the total error, and updating the model parameters of the first model layer by layer during the back propagation process.

[0149] It should be noted that this process may require multiple iterations, and hyperparameters (such as learning rate, batch size, number of iterations, etc.) may need to be adjusted to optimize model performance; in addition, the computational efficiency and accuracy of backpropagation can be improved by using various techniques and optimization methods, such as gradient checking, batch normalization, weight initialization strategies, etc.

[0150] In some embodiments, after calculating the error of the first model, starting from the output layer, each layer is traversed backwards to calculate the gradient of the loss function with respect to each parameter. Here, the chain rule can be used to calculate the gradient, and the error is transferred by multiplying the gradient of the activation value of each intermediate layer; then, after calculating the gradients of all parameters, gradient descent or other optimization algorithms are used to update each parameter; then, the above process can be continuously iterated until the prediction performance of the model reaches a preset stopping condition (for example, the accuracy reaches a certain value or the loss function no longer decreases significantly).

[0151] Taking the loss function (7) of the first model in the embodiment of the present application as an example, the server determines the error based on the loss function, back-propagates the error from the output layer of the regressor in the first model, and back-propagates the error layer by layer in the regressor. When the error reaches each layer, the gradient (that is, the partial derivative of the loss function with respect to the parameters of each layer in the regressor) is solved in combination with the conducted error, and the parameters of each layer in the regressor are updated with the corresponding gradient values.

[0152] For example, assuming the second estimated response rate is The second true response rate is y i The first estimated response rate is The first true response rate is The first loss function and the second loss function are both cross entropy loss functions. and the second true response rate is y i Substituting into the first loss function, we get the first loss function L1 as The first estimated response rate is and the first true response rate is Substituting into the second loss function, the second loss function L2 is obtained as Get the weighted coefficient of the second loss function as w, substitute the first loss function and the second loss function and the weighted coefficient w into the total loss function, and the error is

[0153] The following is a description of the outbound calling method provided in the embodiment of the present application.

[0154] See also Fig. 9 , Fig. 9 is a flowchart of the outbound calling method provided in the embodiment of the present application, which will be combined with Fig. 9 The steps shown are explained.

[0155] In step 201, N second time periods for making outbound calls to a user are set.

[0156] Here, N is a positive integer greater than 1, and the second time period is the candidate time period.

[0157] In some embodiments, N second time periods for making outbound calls to the user are set according to the time period labels corresponding to the user characteristic data.

[0158] It should be noted that when collecting corresponding data of users from the database, a random sampling method can be used to ensure the overall diversity of the sample data.

[0159] For example, N candidate time periods for making outbound calls to a user are set, and the time characteristics of each candidate time period are j=0, ..., N-1.

[0160] In step 202, user features are obtained from user data of the user, and the user features are encoded to obtain a third feature vector, and the N second time periods are respectively encoded to obtain N fourth feature vectors.

[0161] Here, the third feature vector refers to a user feature vector of a user who needs to make an outbound call, and the fourth feature vector refers to a time period feature vector of a candidate time period.

[0162] In some embodiments, user characteristics of the user (such as the user's age, gender, purchase history, points of interest, etc.) are first determined; then, based on the determined user characteristics, corresponding data of the user is collected from a database; then, the user characteristics are encoded to obtain a third feature vector.

[0163] It should be noted that the implementation method of encoding the user features to obtain the third feature vector is similar to the implementation method of the above step 1011. For details, please refer to the specific implementation process of step 1011, which will not be repeated here.

[0164] In step 203, based on the third eigenvector and the N fourth eigenvectors, a prediction process is performed by using the pre-trained first model to obtain an estimated response rate of the user in each second time period.

[0165] Here, the first model is based on Figure 3 The model processing method shown is obtained by training.

[0166] In some embodiments, the third feature vector and N fourth feature vectors of the user who needs to make an outbound call are input into a pre-trained first model. The first model predicts the input data and outputs the user's estimated answer rate in each second time period, that is, the user's estimated answer rate in different candidate time periods.

[0167] For example, the user feature vector and four time period feature vectors (i.e., the time period feature vectors corresponding to the four candidate time periods) of the user can be input into the pre-trained first model. Assume that the first model predicts the input data and obtains an estimated response rate of 0.8 for the user in candidate time period 1 (e.g., 8:00-10:00), an estimated response rate of 0.7 for candidate time period 2 (e.g., 10:00-12:00), an estimated response rate of 0.2 for candidate time period 3 (e.g., 12:00-14:00), and an estimated response rate of 0.6 for candidate time period 4 (e.g., 14:00-16:00).

[0168] In step 204, an outbound call is initiated to the user during the second time period when the estimated answer rate is the highest.

[0169] In some embodiments, after obtaining the estimated answer rates of the user corresponding to different candidate time periods, the candidate time period corresponding to the maximum estimated answer rate may be selected, and an outbound call may be initiated to the user during the candidate time period.

[0170] Continuing with the above example, assuming that the user's estimated response rate in candidate time period 1 (for example, 8:00-10:00) is 0.8, the estimated response rate in candidate time period 2 (for example, 10:00-12:00) is 0.7, the estimated response rate in candidate time period 3 (for example, 12:00-14:00) is 0.2, and the estimated response rate in candidate time period 4 (for example, 14:00-16:00) is 0.6, that is, the user's estimated response rate in candidate time period 1 (8:00-10:00) is the highest, then you can choose to initiate an outbound call to the user in candidate time period 1 to get the opportunity to communicate more deeply with the user, thereby achieving marketing goals.

[0171] Next, an exemplary application of the embodiment of the present application in a practical application scenario will be described. The exemplary application describes the specific implementation process of the model processing method and the outbound calling method in a marketing scenario.

[0172] This application aims at predicting the best time for marketing activities. It adopts the idea of ​​Uplift to take the time period as the experimental group, and the average willingness to make calls in all time periods as the control group. The response rate is taken as the target to establish a corresponding relationship between the time period and the response rate increment. In view of the characteristics of the application scenario, since the actual average willingness data is difficult to collect, it is designed to collect average willingness data for users with more covered time periods to assist in the prediction accuracy of the average willingness.

[0173] In some embodiments, the present application is mainly divided into two parts: time period prediction model training and application.

[0174] In the training phase of the time period prediction model (ie, the first model), stage 1: defining the training data set.

[0175] In some embodiments, historical call record data of a historical period (such as one month) in the system is collected, and the time period j (the granularity can be customized, such as 2 hours as a time period) corresponding to the call time is used as the time feature, where j = 0, 1, ... T-1, T is the total number of time periods; then, the user's age, gender, and the user's call connection status in each historical time period are collected as user characteristics. When the final state of the call record is connected, it is recorded as Y = 1, otherwise it is recorded as Y = 0; when the user corresponding to the call record has N calls in a historical period (such as one month), involving M time periods, N and M can be defined according to the business situation, then the answer rate of the user's N calls is calculated. If the answer rate>R, then Y avg =1, otherwise Y avg =0, where R is the response rate threshold, which can be defined according to the business situation.

[0176] Phase 2: Building the model network structure.

[0177] In some embodiments, see Fig.10 , Fig.10 This is a schematic diagram of the principle of the model processing method provided in the embodiment of the present application. Fig.10 The network structure of the time period prediction model is described.

[0178] First, the user features and time features are extracted to obtain the user feature vector (i.e., the first feature vector) and the time feature vector (i.e., the second feature vector). Then, the user feature vector is linearly transformed and the sigmoid activation function is used to obtain the estimated value of the user's average willingness to connect. (i.e., the first estimated response rate), where The value range is (0,1). At the same time, the user feature vector and the time period feature vector are linearly transformed and the sigmoid activation function is used to obtain the activation result p. The activation result p is converted using the formula a*pa / 2 to obtain the response rate increment of the time period. (i.e., estimated response rate increment), where The value range of is (-a / 2, a / 2), where a can be customized according to the specific scenario; finally, the estimated value of the average user willingness to connect is Estimated response rate increment by time period Add them together to get the user's estimated response rate (i.e., the second estimated response rate)

[0179] Phase 3: Model training process.

[0180] In some embodiments, according to Fig.10 After determining the network structure of the time period prediction model, we can use the training data set defined in stage 1 to train the model and use the formula (1) above as the true response rate y i and estimated response rate The cross entropy Loss (corresponding to the first loss function) is used as the actual average willingness to connect. Estimated average willingness to connect The cross entropy Loss (corresponding to the second loss function) is used, and formula (7) is used as the total loss function of the model.

[0181] In the application phase of the time period prediction model, that is, stage 4: set different time periods and estimate the response rate of users in different time periods.

[0182] In some embodiments, for the user i (i.e., the target user) who needs to make a subsequent call, user characteristics are collected and T time periods are set. The time characteristics of each time period are j, where j=0…T-1. For the T time periods of user i, the estimated answer rate corresponding to time period j is estimated.

[0183] Phase 5: Determine the outbound calling plan.

[0184] In some embodiments, outbound calls are made to users during the time period with the highest estimated answer rate, that is, the outbound call time period is: Make an outbound call to the user in time period j.

[0185] In the absence of a control group, this application creates a fictitious average willingness to connect as a control group, thereby using the Uplift concept to construct an outbound call time period to generate an answer rate gain, and determines the correspondence between the time period and the answer rate through a model; in addition, this application also designs a calibration method for the average willingness to connect, by constructing a target for the average willingness to connect for users with sufficient performance data, and adding an auxiliary loss function for the average willingness to connect, so that the average willingness to connect of the overall users is more accurate, thereby making the corresponding relationship prediction more accurate.

[0186] The following further describes the exemplary structure of the model processing device 543 provided in the embodiment of the present application implemented as a software module. In some embodiments, Figure 2A As shown, the software modules stored in the model processing device 543 of the memory 540 may include: a processing module 5431 and an adjustment module 5432 .

[0187] The processing module 5431 is used to input the user data of the sample user into the first model for first processing to obtain a first estimated response rate and a second estimated response rate; wherein the first processing includes: encoding the user features in the user data to obtain a first feature vector, and encoding the time features in the user data to obtain a second feature vector, the time features including multiple first time periods; determining the first estimated response rate of the sample user in multiple first time periods based on the first feature vector, and selecting a first time period from the multiple first time periods, and determining the second estimated response rate of the sample user in the selected first time period based on the first feature vector and the second feature vector; the adjustment module 5432 adjusts the model parameters of the first model based on the first estimated response rate and the second estimated response rate.

[0188] In some embodiments, the processing module 5431 is also used to perform linear transformation on the first feature vector to obtain a first linear transformation result; perform activation processing on the first linear transformation result to obtain a first activation result, and use the first activation result as the first estimated response rate of the sample user in multiple first time periods.

[0189] In some embodiments, the processing module 5431 is also used to perform a linear transformation on the first feature vector to obtain a first linear transformation result; perform a linear transformation on the sub-feature vector corresponding to the selected first time period in the second feature vector to obtain a second linear transformation result; perform activation processing on the first linear transformation result and the second linear transformation result to obtain a second activation result; and determine the second estimated response rate of the sample user in the selected first time period based on the second activation result and the first estimated response rate of the sample user in multiple first time periods.

[0190] In some embodiments, the processing module 5431 is also used to convert the second activation result to obtain a conversion result, and use the conversion result as the estimated response rate increment of the sample user in the selected first time period; based on the estimated response rate increment of the sample user in the selected first time period, and the first estimated response rate of the sample user in multiple first time periods, determine the second estimated response rate of the sample user in the selected first time period.

[0191] In some embodiments, the adjustment module 5432 is also used to substitute the first estimated response rate and the first actual response rate of the sample users in multiple first time periods into the first loss function to obtain a first error; substitute the second estimated response rate and the second actual response rate of the sample users in the selected first time period into the second loss function to obtain a second error; determine the total error based on the first error and the second error; and adjust the model parameters of the first model based on the total error.

[0192] In some embodiments, the adjustment module 5432 is also used to perform a weighted sum of the first error and the second error, and use the weighted sum result as the total error; the adjustment module 5432 is also used to perform back propagation based on the total error, and update the model parameters of the first model layer by layer during the back propagation process.

[0193] In some embodiments, Figure 2B As shown, the software modules stored in the outbound calling device 643 of the memory 650 may include: a generating module 6431, an encoding module 6432, a predicting module 6433 and an outbound calling module 6434.

[0194] A generation module 6431 is used to set N second time periods for making outbound calls to the user, where N is a positive integer greater than 1; an encoding module 6432 is used to obtain user features from the user data of the user, and encode the user features to obtain a third feature vector, and encode the N second time periods respectively to obtain N fourth feature vectors; a prediction module 6433 is used to perform prediction processing based on the third feature vector and the N fourth feature vectors through a pre-trained first model to obtain an estimated answer rate of the user in each second time period; an outbound call module 6434 is used to initiate an outbound call to the user in the second time period with the largest estimated answer rate.

[0195] It should be noted that the description of the device in the embodiment of the present application is similar to the description of the method embodiment above, and has similar beneficial effects as the method embodiment, so it will not be repeated. Figure 3 , Figure 4 , Figure 5 , Figure 6 , Figure 7 , Figure 8 ,or Fig. 9The present invention should be understood by referring to the description of any one of the accompanying drawings.

[0196] The embodiment of the present application provides a computer program product, which includes a computer program or a computer executable instruction, and the computer program or the computer executable instruction is stored in a computer-readable storage medium. The processor of the computer device reads the computer executable instruction from the computer-readable storage medium, and the processor executes the computer executable instruction, so that the computer device executes the model processing method or the outbound call method described in the embodiment of the present application.

[0197] The embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, the processor will be caused to execute the model processing method or the outbound call method provided in the embodiment of the present application, for example, Figure 3 , Figure 4 , Figure 5 , Figure 6 , Figure 7 , Figure 8 The model processing method shown in any of the accompanying drawings, or Fig. 9 The outbound calling method is shown.

[0198] In some embodiments, the computer-readable storage medium may be a ferroelectric random access memory (FRAM), ROM, programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory, magnetic surface memory, optical disk, or compact disc read-only memory (CD-ROM) and other memories; it may also be various devices including one or any combination of the above memories.

[0199] In some embodiments, executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0200] As an example, the executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one site, or on multiple electronic devices distributed at multiple sites and interconnected by a communication network.

[0201] In summary, the embodiments of the present application have the following beneficial effects:

[0202] (1) By setting up an average connection willingness control group and taking different time periods as controls, a corresponding relationship between time periods and connection willingness is established, thereby improving the incremental prediction effect of time periods on connection willingness;

[0203] (2) By establishing true labels for the average willingness to connect of users with sufficient performance, the learning of the average willingness to connect is assisted, making the average willingness to connect of the entire user base more accurate, thereby making the incremental prediction effect of the willingness to connect in each time period more accurate.

[0204] The above is only an embodiment of the present application and is not intended to limit the protection scope of the present application. Any modifications, equivalent substitutions and improvements made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. A model processing method, characterized in that: The method comprises: Inputting user data of the sample user into the first model for first processing to obtain a first estimated response rate and a second estimated response rate; wherein the first processing includes: encoding user features in the user data to obtain a first feature vector, and encoding time features in the user data to obtain a second feature vector, wherein the time feature includes a plurality of first time periods; determining a first estimated response rate of the sample user in the plurality of first time periods based on the first feature vector; selecting a first time period from the plurality of first time periods, and determining a second estimated response rate of the sample user in the selected first time period based on the first feature vector and the second feature vector; Model parameters of the first model are adjusted based on the first predicted response rate and the second predicted response rate.

2. The method according to claim 1, characterized in that: The determining, based on the first feature vector, a first estimated response rate of the sample user in the plurality of first time periods comprises: Performing a linear transformation on the first eigenvector to obtain a first linear transformation result; An activation process is performed on the first linear transformation result to obtain a first activation result, and the first activation result is used as a first estimated response rate of the sample user in the plurality of first time periods.

3. The method according to claim 1, characterized in that The determining, based on the first feature vector and the second feature vector, a second estimated response rate of the sample user in the selected first time period includes: Performing a linear transformation on the first eigenvector to obtain a first linear transformation result; Performing a linear transformation on the sub-feature vector corresponding to the selected first time period in the second feature vector to obtain a second linear transformation result; Performing activation processing on the first linear transformation result and the second linear transformation result to obtain a second activation result; Based on the second activation result and the first estimated response rates of the sample users in the plurality of first time periods, a second estimated response rate of the sample users in the selected first time period is determined.

4. The method according to claim 3, characterized in that The determining, based on the second activation result and the first estimated response rates of the sample users in the plurality of first time periods, a second estimated response rate of the sample users in the selected first time period comprises: Performing conversion processing on the second activation result to obtain a conversion result, and using the conversion result as the estimated response rate increment of the sample user in the selected first time period; A second estimated response rate of the sample user in the selected first time period is determined based on the estimated response rate increment of the sample user in the selected first time period and the first estimated response rates of the sample user in the plurality of first time periods.

5. The method according to claim 1, characterized in that: The adjusting the model parameters of the first model based on the first estimated response rate and the second estimated response rate includes: Substituting the first estimated response rate and the first true response rate of the sample user in the plurality of first time periods into a first loss function to obtain a first error; Substituting the second estimated response rate and the second true response rate of the sample user in the selected first time period into a second loss function to obtain a second error; determining a total error based on the first error and the second error; Model parameters of the first model are adjusted based on the total error.

6. The method according to claim 5, characterized in that The determining of the total error based on the first error and the second error comprises: Performing a weighted summation on the first error and the second error, and taking the weighted summation result as a total error; The adjusting the model parameters of the first model based on the total error comprises: Back propagation is performed based on the total error, and model parameters of the first model are updated layer by layer during the back propagation process.

7. An outbound calling method, characterized in that: The method comprises: Setting N second time periods for making outbound calls to the user, where N is a positive integer greater than 1; Acquire a user feature from the user data of the user, encode the user feature to obtain a third feature vector, and encode the N second time periods respectively to obtain N fourth feature vectors; Based on the third eigenvector and the N fourth eigenvectors, performing prediction processing through a pre-trained first model to obtain an estimated response rate of the user in each of the second time periods; Initiate an outbound call to the user during the second time period when the estimated answer rate is the largest.

8. A model processing device, characterized in that: The device comprises: A processing module, configured to input user data of a sample user into a first model for first processing to obtain a first estimated response rate and a second estimated response rate; wherein the first processing includes: encoding user features in the user data to obtain a first feature vector, and encoding time features in the user data to obtain a second feature vector, wherein the time feature includes a plurality of first time periods; determining a first estimated response rate of the sample user in the plurality of first time periods based on the first feature vector; selecting a first time period from the plurality of first time periods, and determining a second estimated response rate of the sample user in the selected first time period based on the first feature vector and the second feature vector; An adjustment module is used to adjust a model parameter of the first model based on the first estimated response rate and the second estimated response rate.

9. An outbound calling device, characterized in that: The device comprises: A generating module, used to set N second time periods for making outbound calls to the user, where N is a positive integer greater than 1; an encoding module, configured to obtain user features from the user data of the user, encode the user features to obtain a third feature vector, and encode the N second time periods respectively to obtain N fourth feature vectors; A prediction module, configured to perform prediction processing based on the third eigenvector and the N fourth eigenvectors through a pre-trained first model to obtain an estimated response rate of the user in each of the second time periods; An outbound calling module is used to initiate an outbound call to the user during the second time period when the estimated answer rate is the largest.

10. An electronic device, characterized in that: include: A memory for storing executable instructions; The processor is used to implement the model processing method described in any one of claims 1 to 7, or the outbound calling method described in claim 8 when executing the executable instructions stored in the memory.

11. A computer-readable storage medium storing computer-executable instructions, characterized in that: When the computer executable instructions are executed by a processor, the model processing method described in any one of claims 1 to 7 is implemented, or the outbound calling method described in claim 8 is implemented.

12. A computer program product comprising a computer program or computer executable instructions, characterized in that: When the computer program or computer executable instructions are executed by a processor, the model processing method described in any one of claims 1 to 7 is implemented, or the outbound calling method described in claim 8 is implemented.

Images