Training method of sorting model, sorting method of contact channels and related devices

Abstract

The invention discloses a training method of a sorting model, a sorting method of contact channels and a related device. Firstly, training data is obtained, the data comprises multiple pieces of sample data with recommendation sequence labels, each piece of sample data comprises a sample user, N contact channels of the user and corresponding trimming information sets, and the recommendation sequence labels are used for indicating recommendation sequences of the contact channels. The trimming information set is extracted from a historical full-amount information set including a basic information set of the customer, a historical insurance policy information set, a historical marketing contact time sequence information set, a historical active interaction time sequence information set and a historical customer contact opportunity time sequence information set. And then inputting the training data into a neural network for training, and finally obtaining a sorting model. By using various data in a historical full-amount information set, the model can determine an optimal contact channel recommendation sequence based on the learned mutual dependency relationship between different channels, so that the marketing effect and the customer satisfaction are improved.

Description

Technical Field

[0001] The present application relates to the field of insurance marketing technology, and in particular to a training method for a ranking model, a ranking method for contact channels, and related devices. Background Art

[0002] In the insurance marketing field, insurance agents may contact customers through a variety of contact channels, including but not limited to one-on-one interviews, regular insurance company activities, online communication software, etc. Customers' preferences for channels also change at different times. For example, before their birthdays, customers may prefer online contact, while they may prefer offline visits when renewing their policies. Each channel has its specific advantages and limitations, and customers have different preferences at different times.

[0003] The multi-channel recommendation problem is a multi-label classification problem. In existing technical solutions, the multi-channel recommendation problem is usually split into multiple binary classification problems and solved by calculating the recommendation score for each channel separately. However, in reality, the customer contact of different channels is not independent of each other, and the optimal customer contact channel for different customers at different times will not be the same. Existing technical solutions do not consider temporal characteristics such as the timing of customer contact, and the scores of different channels lack a unified comparison standard. This method can only provide a list of recommended channels, but cannot determine the priority ranking of these channels, and cannot help marketers choose the best channel to reach customers at the best time.

[0004] Therefore, how to comprehensively consider the interdependence between different channels and the changes in customers' temporal preferences in multi-channel recommendations to determine the recommended order of reaching channels and thus improve marketing effectiveness is a technical problem that technical personnel in this field urgently need to solve. Summary of the invention

[0005] Based on the above problems, the present application provides a training method for a sorting model, a sorting method for contact channels and related devices, which can comprehensively consider the interdependence between different channels and the changes in customers' temporal preferences in multi-channel recommendations to determine the recommended order of contact channels, thereby improving marketing effectiveness.

[0006] The embodiments of the present application disclose the following technical solutions:

[0007] A training method for a sorting model, the method comprising:

[0008] Acquire training data; the training data includes multiple sample data with recommendation sequence labels; one sample data includes a sample user, N contact channels of the sample user, and a balancing information set corresponding to each contact channel; the recommendation sequence label is used to indicate the recommended sequence of the contact channels in the sample data; the balancing information set is obtained from the historical full information set of the contact channels; the historical full information set includes a customer basic information set, a customer historical insurance policy information set, a historical marketing contact time series information set, a customer historical active interaction time series information set, and a historical contact opportunity time series information set; N is a positive integer;

[0009] The training data is input into a neural network for model training to obtain a sorting model.

[0010] In a possible implementation, the process of acquiring the trim information set includes:

[0011] For each of the contact channels, the historical full information set of the sample user is obtained, and feature engineering processing is performed on the information in the historical full information set to obtain a feature information set;

[0012] Normalizing the feature variables in the feature information set to obtain a processed information set;

[0013] Using a correlation analysis method, M characteristic variables related to the contact channel are selected from the processed information as target information; M is a positive integer;

[0014] integrating all the target information of the contact channels into a first target information set;

[0015] A positive and negative sample balancing operation is performed on the feature variables in the first target information set to obtain the balanced information set.

[0016] In a possible implementation, the method further includes:

[0017] Set the binary cross entropy loss function as the target loss function;

[0018] The objective loss function is used to adjust parameters of the sorting model.

[0019] In a possible implementation, the neural network includes an input layer, a hidden layer, and an output layer;

[0020] Among them, the number of nodes in the input layer is the same as the number of feature variables in the balancing information set; the activation function of the hidden layer is the rectified linear unit ReLU function; the output dimension of the output layer is equal to N, and the activation function of the output layer is the Sigmoid function.

[0021] A method for sorting contact channels, the method comprising:

[0022] Determine a target user, and obtain N second target information sets corresponding to the N contact channels of the target user; one second target information set includes M feature variables related to one contact channel of the target user; M is a positive integer;

[0023] Based on the N second target information sets corresponding to the N contact channels of the target user, the N contact channels of the target user are sorted by a sorting model; the sorting model is trained according to the sorting model training method as described above.

[0024] A training method for a sorting model, the method comprising:

[0025] A first acquisition unit is used to acquire training data; the training data includes a plurality of sample data with recommendation sequence labels; one sample data includes a sample user, N contact channels of the sample user, and a balancing information set corresponding to each contact channel; the recommendation sequence label is used to indicate the recommended sequence of the contact channels in the sample data; the balancing information set is acquired from the historical full information set of the contact channels; the historical full information set includes a customer basic information set, a customer historical insurance policy information set, a historical marketing contact time series information set, a customer historical active interaction time series information set, and a historical contact opportunity time series information set; N is a positive integer;

[0026] The model training unit is used to input the training data into the neural network for model training to obtain a sorting model.

[0027] A method for sorting contact channels, the method comprising:

[0028] A determination unit, used for determining a target user;

[0029] A second acquisition unit is used to acquire N second target information sets corresponding to the N contact channels of the target user; one second target information set includes M feature variables related to one contact channel of the target user; M is a positive integer;

[0030] The second sorting unit is used to sort the N contact channels of the target user through a sorting model based on the N second target information sets corresponding to the N contact channels of the target user; the sorting model is trained according to the sorting model training method as described above.

[0031] A training device for a sorting model comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the training method for the sorting model as described above is implemented.

[0032] A contact channel sorting device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the contact channel sorting method as claimed in any one of claim 5 is implemented.

[0033] A computer-readable storage medium stores instructions, which, when executed on a terminal device, enable the terminal device to execute the above-mentioned method for training a sorting model or the above-mentioned method for sorting contact channels.

[0034] Compared with the prior art, this application has the following beneficial effects:

[0035] The present application provides a method for training a ranking model, a method for ranking contact channels, and related devices. Specifically, when executing the training method for the ranking model provided in the embodiment of the present application, first, training data including multiple sample data with recommended order labels can be obtained, one sample data includes a sample user, N contact channels of the sample user, and a balancing information set corresponding to each contact channel, and the recommended order label is used to indicate the recommended order of the contact channels in the sample data; the balancing information set is obtained from the historical full information set of the contact channel including the customer basic information set, the customer's historical policy information set, the historical marketing contact timing information set, the customer's historical active interaction timing information set, and the historical contact customer opportunity timing information set. Then input the training data into the neural network for model training to obtain a ranking model. The present application uses the customer's basic information, policy information, marketing contact timing information, active interaction timing information, and contact customer opportunity timing information in the historical full information set, so that the model can determine the optimal contact channel recommendation order based on the learned interdependence between different channels. At the same time, combining the historical active interaction timing information and the contact timing information can better understand the changing trends of customers' preferences at different time points, help the model determine the customer's timing preferences, and learn the recommended order of contact channels accordingly, thereby improving marketing effectiveness and customer satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0037] Figure 1 A method flow chart of a method for training a sorting model provided in an embodiment of the present application;

[0038] Figure 2 A method flow chart of a contact channel sorting method provided in an embodiment of the present application;

[0039] Figure 3 A schematic diagram of the structure of a training device for a sorting model provided in an embodiment of the present application;

[0040] Figure 4 A schematic diagram of the structure of a contact channel sorting device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0041] To facilitate understanding of the technical solutions provided by the embodiments of the present application, the background technology involved in the embodiments of the present application will be described below.

[0042] Binary classification problems are a fundamental type of problem in machine learning where the goal is to classify input data into one of two mutually exclusive categories. In other words, binary classification problems involve classifying data to determine which of two predefined categories it belongs to.

[0043] For example, in the field of insurance marketing, a binary classification problem may involve determining whether a customer will respond positively to a particular channel (such as online communication or face-to-face visits). Specifically, this problem can be formulated as:

[0044] Positive response: The customer shows interest or responds positively to the channel.

[0045] Negative response: The customer has no interest or positive response to the channel.

[0046] For example, suppose there is a binary classification problem where the goal is to predict whether a customer will respond to online communication:

[0047] Input features: customer’s age, gender, historical purchase history, recent interaction behavior, etc.

[0048] Output label: 1 (indicating response) or 0 (indicating no response).

[0049] By training the model to learn the relationship between these features and labels, it is possible to predict whether a new customer will respond to online communication.

[0050] As described above, the insurance marketing sector faces the challenge of reaching customers through multiple channels, where customer preferences change over time. For example, customers may prefer online interactions before birthdays, but prefer offline visits when renewing policies. Existing technologies often simplify the multi-channel recommendation problem into an independent binary classification, without considering the correlation between different channels and the temporal characteristics of customers. This results in a lack of a unified comparison standard for recommendation results, only providing a list of channels without being able to sort priorities, and being unable to help marketers choose the best contact method at the best time, highlighting the technical problems that need to be solved.

[0051] In order to solve this problem, a training method for a sorting model, a sorting method for contact channels and related devices are provided in an embodiment of the present application. First, training data is obtained, which includes multiple sample data with recommended order labels, each sample data includes a sample user, N contact channels of the user and its corresponding balancing information set, and the recommended order label is used to indicate the recommended order of the contact channel. The balancing information set is extracted from a historical full information set including a customer basic information set, a customer historical policy information set, a historical marketing contact timing information set, a customer historical active interaction timing information set and a historical customer contact timing information set. Then these training data are input into a neural network for training, and finally a sorting model is obtained. This application enables the model to learn the interdependence between different channels by utilizing the customer basic information, policy information, marketing contact timing information, active interaction timing information and customer contact timing information in the historical full information set, so as to determine the best contact channel recommendation order, and combine the historical active interaction timing information and customer contact timing information to better understand the customer's preference change trend at different time points, help the model determine the customer's timing preference, and thus improve marketing effectiveness and customer satisfaction.

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0053] See also Figure 1 , which is a flow chart of a method for training a sorting model provided in an embodiment of the present application, such as Figure 1 As shown, the training method of the sorting model may include steps S101-S102:

[0054] S101: Obtain training data.

[0055] To train the ranking model, you first need to obtain a high-quality data set for training to build a ranking model that can effectively recommend the order of contact channels. Specifically, the training data contains multiple sample data, each of which includes the following key components:

[0056] Sample user: refers to a specific user object in the training data.

[0057] N contact channels of the sample user: represents the N different contact channels involved in the user, where N is a positive integer.

[0058] Balancing information set: Each contact channel corresponds to a balancing information set, which is extracted from the historical full information set. The historical full information set contains the following types of data:

[0059] Customer basic information set: records the customer's basic personal information;

[0060] Customer historical insurance policy information set: records the customer's historical insurance policy information;

[0061] Historical marketing contact time series information set: records the interactions between past marketing activities and customers;

[0062] Customer historical active interaction time series information set: records the historical records of customers actively interacting with the company or product;

[0063] Historical contact timing information set: records past contact behaviors and the time points when they occurred.

[0064] Recommended order label: used to indicate the order in which each contact channel should be recommended in a given sample data.

[0065] In this way, each sample in the training data contains not only specific information about the user and channel, but also the recommendation order labels of these channels, allowing the model to learn the dependencies between different channels and optimize the recommendation order accordingly to improve marketing effectiveness and customer satisfaction.

[0066] In a possible implementation, the process of acquiring the trim information set includes A1-A5:

[0067] A1: For each of the contact channels, the historical full information set of the sample user is obtained, and feature engineering processing is performed on the information in the historical full information set to obtain a feature information set.

[0068] For each contact channel, data is obtained from the historical full information set of sample users, and feature engineering is performed on these data to generate feature information sets. These steps include extracting basic customer information, historical insurance policy information, historical marketing contact time series information, historical active interaction time series information, and historical contact timing time series information.

[0069] A2: Normalize the feature variables in the feature information set to obtain a processed information set.

[0070] All feature variables in the feature information set are normalized to eliminate the dimensional differences between different features and obtain the processed information set, namely the processed information set.

[0071] A3: Utilize a correlation analysis method to select M characteristic variables related to the contact channel from the processed information as target information.

[0072] Using the correlation analysis method, multiple feature variables related to contact channels are selected from the processed information set as target information. This step ensures a high correlation between the selected features and the contact channels.

[0073] In a possible implementation, step A3: using a correlation analysis method to select M characteristic variables related to the contact channel from the processed information as target information can be implemented through the following steps:

[0074] (1) Calculate the correlation between the feature and the target variable: Use information value (IV value) and correlation methods to quantify the degree of correlation between each feature variable and the contact channel.

[0075] The IV value is an important indicator to measure the predictive ability of feature variables on target variables and is often used in classification problems, especially in the fields of credit scoring and marketing.

[0076] Correlation can be calculated using statistical methods such as Pearson correlation coefficient and Spearman correlation coefficient.

[0077] (2) Sort feature variables based on relevance: Sort all feature variables based on their relevance to the contact channel. Features with high IV values ​​or high relevance indicate that these features have stronger predictive power for the contact channel.

[0078] This ranking can help us identify which characteristics have a greater impact on the contact channel.

[0079] (3) Select important features: Assuming that we need to select M features as target information, then according to the above sorting results, select the first M features.

[0080] These selected features should have high IV values ​​or high correlations, indicating that they have a greater impact on the contact channels and are important feature variables.

[0081] (4) Screen out irrelevant features: Some features, such as gender and name, may have low relevance to the contact channel. These features may not contribute to the prediction of the target variable and may even introduce noise.

[0082] Therefore, these features are screened out to reduce the complexity of the model and improve its accuracy and generalization ability.

[0083] For example, suppose you are analyzing a marketing dataset that contains feature variables such as the user's age, gender, income, education level, and whether they clicked on an ad (exposure channel). The goal is to find the features that are most correlated with a user clicking on an ad.

[0084] Calculate IV value and correlation: First, calculate the IV value of each feature variable (such as age, income, etc.) and calculate the correlation with clicking ads. For example, age and income may have high IV values ​​and correlations, while gender and name may have low IV values ​​and correlations.

[0085] Sort feature variables: Sort these feature variables according to IV value and correlation. For example, if the IV values ​​of age and income are 0.5 and 0.4 respectively, and the correlations are 0.6 and 0.7 respectively, these two features will be ranked first.

[0086] Select important features: If you assume that you need to select three features, then based on the sorting results, you may choose age, income, and education level.

[0087] Screen out irrelevant features: Features such as gender and name are excluded due to their low relevance. In this way, you can ensure that the selected features are truly relevant to the contact channel, thereby improving the predictive power and efficiency of the model.

[0088] Through this method, not only the most important feature variables can be screened out, but also those insignificant features can be eliminated, thereby building a more effective and concise prediction model.

[0089] Exemplarily, the filtered target information may include:

[0090] ① Customer contact opportunities: birthdays in the next three months, policy renewals in the next three months, benefits expiring in the next three months, etc.

[0091] ② Historical marketing contact time series information: number of marketing times in the past year, number of days since the most recent marketing contact, most recent marketing channel category, etc.

[0092] ③Customer insurance policy purchase information: how long ago the most recent policy was issued, whether there will be policy renewals within the next three months, etc.

[0093] ④Customer personal information: the level of the city where the customer is located, etc.

[0094] A4: Integrate all the target information of the contact channel into a first target information set.

[0095] All target information of the contact channels is integrated into a target information set to form a set containing key features (ie, the first target information set).

[0096] A5: Perform a positive and negative sample balancing operation on the feature variables in the first target information set to obtain the balanced information set.

[0097] The positive and negative sample balancing operation is performed on the feature variables in the first target information set to ensure that the number of positive and negative samples in the data set is balanced, thereby preventing the model from being biased towards a certain type of sample during the training process, and finally obtaining a balanced information set.

[0098] S102: Input the training data into a neural network for model training to obtain a sorting model.

[0099] After obtaining the training data, the training data is input into a well-designed neural network for model training. The architecture of the neural network includes input layer, hidden layer and output layer.

[0100] Input layer: The number of nodes in the input layer is the same as the number of feature variables in the balanced information set. This means that each feature variable will have a corresponding input node, ensuring that the model can fully receive and process all relevant information.

[0101] Hidden layer: The number of nodes in the hidden layer is adjustable, and the hidden layer uses the rectified linear unit (ReLU) as the activation function. The ReLU function can effectively capture the complex nonlinear relationship between features and improve the learning and generalization capabilities of the model.

[0102] Output layer: The output dimension of the output layer is equal to N (i.e. the number of contact channels), and the Sigmoid function is used as the activation function. The Sigmoid function can convert the output of the model into a value between 0 and 1, which is convenient for interpretation and application, such as determining the recommendation order by comparing the output values ​​of different contact channels.

[0103] The specific steps are as follows:

[0104] Data input: Input the balanced information set in the training data into the input layer of the neural network to ensure that each feature variable can be received by the model.

[0105] Model training: During the training process, the neural network continuously adjusts weights and biases through forward propagation and back propagation to minimize prediction errors. The hidden layer uses the ReLU activation function to enhance the model's ability to capture complex patterns, and the output layer uses the Sigmoid activation function to generate normalized output values.

[0106] Model output: After multiple iterations of optimization, the neural network generates a ranking model that can predict the best order of contact channel recommendations based on the input user and contact channel information.

[0107] In short, by inputting the prepared training data into a neural network with a specific structure for training, a ranking model can be obtained that can accurately predict the recommended order of contact channels, thereby providing strong support for marketing strategies.

[0108] In one possible implementation, a grid retrieval method can be used to debug the number of layers, number of nodes, learning rate, batch, and other factors that affect model performance of the neural network.

[0109] In a possible implementation, the method further includes:

[0110] The binary cross entropy loss function is set as the target loss function, and then the target loss function is used to adjust the parameters of the sorting model.

[0111] Based on the contents of S101-S102, it can be seen that this application uses the basic customer information, policy information, marketing contact timing information, active interaction timing information and customer contact timing information in the historical full information set to enable the model to learn the interdependence between different channels, thereby determining the best contact channel recommendation order. At the same time, by analyzing the historical active interaction timing information and customer contact timing information, it is possible to more accurately grasp the changes in customer preferences at different time points, help the model identify the customer's timing preferences, and then optimize the recommended order of contact channels, improve marketing effectiveness and customer satisfaction.

[0112] Based on the above-mentioned embodiment of the training method of the sorting model, the present application embodiment also provides a method for sorting contact channels. Figure 2 , Figure 2 A method flow chart of a method for sorting contact channels provided in an embodiment of the present application. Figure 2 As shown, the method includes S201-S202:

[0113] S201: Determine a target user, and obtain N second target information sets corresponding to N contact channels of the target user.

[0114] First, determine a specific target user. This target user can be any user who needs marketing activities or service recommendations. For each contact channel of the target user, obtain the second target information set related to it. Each second target information set contains M feature variables related to the contact channel, where M is a positive integer. These feature variables may include but are not limited to basic customer information, historical policy information, historical marketing contact timing information, historical active interaction timing information, and historical contact timing information.

[0115] Specifically, assuming that the target user has N contact channels, there are N second target information sets, each of which contains M feature variables. For example:

[0116] Contact channel 1: [characteristic variable 1, characteristic variable 2, ..., characteristic variable M]

[0117] Contact channel 2: [characteristic variable 1, characteristic variable 2, ..., characteristic variable M] ...

[0119] Contact channel N: [characteristic variable 1, characteristic variable 2,..., characteristic variable M].

[0120] It should be noted that the types of the M feature variables are consistent with the data categories of the M feature variables related to the contact channels in "using the correlation analysis method to screen out M feature variables related to the contact channels from the processed information as target information" during the sorting model training.

[0121] S202: Based on the N second target information sets corresponding to the N contact channels of the target user, sort the N contact channels of the target user by using a sorting model.

[0122] The N second target information sets obtained above are input into the pre-trained sorting model. The sorting model is trained by the training method described above and can effectively predict the recommended order of contact channels. The sorting model processes the input N second target information sets, and generates a recommendation score or probability value for each contact channel based on the internal learning and optimization mechanism. These scores or probability values ​​reflect the priority of different contact channels. The model sorts the N contact channels based on the generated scores or probability values. The sorted results will indicate which contact channels should be recommended to the target users first.

[0123] Based on the contents of S201-S202, it can be seen that the target user is first determined, and the second target information set corresponding to the N contact channels of the target user is obtained, each of which contains characteristic variables related to a contact channel of the target user. Through this information, the N contact channels of the target user can be optimized and sorted using the sorting model. In addition, the sorting model is obtained according to the previously trained method, which can provide a more accurate and actual sorting result, thereby improving the recommendation accuracy and customer satisfaction.

[0124] See also Figure 3 , Figure 3 A schematic diagram of a training device for a sorting model provided in an embodiment of the present application. Figure 3 As shown, the training device of the sorting model includes:

[0125] The first acquisition unit 301 is used to acquire training data; the training data includes multiple sample data with recommendation sequence labels; one sample data includes a sample user, N contact channels of the sample user, and a balancing information set corresponding to each contact channel; the recommendation sequence label is used to indicate the recommended sequence of the contact channels in the sample data; the balancing information set is obtained from the historical full information set of the contact channel; the historical full information set includes a customer basic information set, a customer historical insurance policy information set, a historical marketing contact time series information set, a customer historical active interaction time series information set, and a historical contact opportunity time series information set; N is a positive integer;

[0126] The model training unit 302 is used to input the training data into the neural network for model training to obtain a sorting model.

[0127] In a possible implementation manner, the device further includes:

[0128] A second acquisition unit, configured to acquire the historical full information set of the sample user for each contact channel;

[0129] A feature engineering processing unit, which performs feature engineering processing on the information in the historical full information set to obtain a feature information set;

[0130] A normalization processing unit, used for performing normalization processing on the feature variables in the feature information set to obtain a processed information set;

[0131] A screening unit, used for screening M characteristic variables related to the contact channel from the processed information as target information by using a correlation analysis method; M is a positive integer;

[0132] an integration unit, configured to integrate all the target information of the contact channel into a first target information set;

[0133] The sample balancing unit is used to perform a positive and negative sample balancing operation on the feature variables in the first target information set to obtain the balanced information set.

[0134] In a possible implementation manner, the device further includes:

[0135] A loss function setting unit, used to set the binary cross entropy loss function as the target loss function;

[0136] A parameter adjustment unit is used to adjust the parameters of the sorting model using the target loss function.

[0137] In a possible implementation manner, the device further includes:

[0138] The neural network includes an input layer, a hidden layer and an output layer;

[0139] Among them, the number of nodes in the input layer is the same as the number of feature variables in the balancing information set; the activation function of the hidden layer is the ReLU function; the output dimension of the output layer is equal to N, and the activation function of the output layer is the Sigmoid function.

[0140] See also Figure 4 , Figure 4 This is a schematic diagram of the structure of a contact channel sorting device provided in an embodiment of the present application. Figure 4 As shown, the contact channel sorting device includes:

[0141] A determination unit 401, configured to determine a target user;

[0142] The second acquisition unit 402 is used to acquire N second target information sets corresponding to the N contact channels of the target user; one second target information set includes M feature variables related to one contact channel of the target user; M is a positive integer;

[0143] The second sorting unit 403 is used to sort the N contact channels of the target user through a sorting model based on the N second target information sets corresponding to the N contact channels of the target user; the sorting model is trained according to the sorting model training method according to any one of claims 1-4.

[0144] In addition, an embodiment of the present application also provides a training device for a sorting model, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the training method for the sorting model as described above is implemented.

[0145] In addition, an embodiment of the present application also provides a contact channel sorting device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the contact channel sorting method described above is implemented.

[0146] In addition, an embodiment of the present application also provides a computer-readable storage medium, in which instructions are stored. When the instructions are executed on a terminal device, the terminal device executes the training method of the sorting model as described above, or executes the sorting method of the contact channels as described above.

[0147] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. Some or all of them can be selected according to actual needs to achieve the purpose of the embodiment scheme. Ordinary technicians in this field can understand and implement it without paying creative work.

[0148] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A training method for a sorting model, characterized in that: The method comprises: Acquire training data; the training data includes multiple sample data with recommendation sequence labels; one sample data includes a sample user, N contact channels of the sample user, and a balancing information set corresponding to each contact channel; the recommendation sequence label is used to indicate the recommended sequence of the contact channels in the sample data; the balancing information set is obtained from the historical full information set of the contact channels; the historical full information set includes a customer basic information set, a customer historical insurance policy information set, a historical marketing contact time series information set, a customer historical active interaction time series information set, and a historical contact opportunity time series information set; N is a positive integer; The training data is input into a neural network for model training to obtain a sorting model.

2. The method according to claim 1, characterized in that The process of acquiring the balancing information set includes: For each of the contact channels, the historical full information set of the sample user is obtained, and feature engineering processing is performed on the information in the historical full information set to obtain a feature information set; Normalizing the feature variables in the feature information set to obtain a processed information set; Using a correlation analysis method, M characteristic variables related to the contact channel are selected from the processed information as target information; M is a positive integer; integrating all the target information of the contact channels into a first target information set; A positive and negative sample balancing operation is performed on the feature variables in the first target information set to obtain the balanced information set.

3. The method according to claim 1, characterized in that The method further comprises: Set the binary cross entropy loss function as the target loss function; The objective loss function is used to adjust parameters of the sorting model.

4. The method according to claim 1, characterized in that The neural network includes an input layer, a hidden layer and an output layer; Among them, the number of nodes in the input layer is the same as the number of feature variables in the balancing information set; the activation function of the hidden layer is the rectified linear unit ReLU function; the output dimension of the output layer is equal to N, and the activation function of the output layer is the Sigmoid function.

5. A method for sorting contact channels, characterized in that: The method comprises: Determine a target user, and obtain N second target information sets corresponding to the N contact channels of the target user; one second target information set includes M feature variables related to one contact channel of the target user; M is a positive integer; Based on the N second target information sets corresponding to the N contact channels of the target user, the N contact channels of the target user are sorted by a sorting model; the sorting model is trained according to the sorting model training method according to any one of claims 1 to 4.

6. A method for training a sorting model, characterized in that: The method comprises: A first acquisition unit is used to acquire training data; the training data includes a plurality of sample data with recommendation sequence labels; one sample data includes a sample user, N contact channels of the sample user, and a balancing information set corresponding to each contact channel; the recommendation sequence label is used to indicate the recommended sequence of the contact channels in the sample data; the balancing information set is acquired from the historical full information set of the contact channels; the historical full information set includes a customer basic information set, a customer historical insurance policy information set, a historical marketing contact time series information set, a customer historical active interaction time series information set, and a historical contact opportunity time series information set; N is a positive integer; The model training unit is used to input the training data into the neural network for model training to obtain a sorting model.

7. A method for sorting contact channels, characterized in that: The method comprises: A determination unit, used for determining a target user; A second acquisition unit is used to acquire N second target information sets corresponding to the N contact channels of the target user; one second target information set includes M feature variables related to one contact channel of the target user; M is a positive integer; A second sorting unit is used to sort the N contact channels of the target user through a sorting model based on the N second target information sets corresponding to the N contact channels of the target user; the sorting model is trained according to the training method of the sorting model as described in any one of claims 1-4.

8. A training device for a sorting model, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the training method of the sorting model as described in any one of claims 1 to 4 is implemented.

9. A contact channel sorting device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the contact channel sorting method as described in any one of claim 5 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed on a terminal device, the terminal device executes the training method of the sorting model as described in any one of claims 1 to 4, or executes the contact channel sorting method as described in claim 5.

Images