CN-115423012-B - Customer calling reason determining method and device

CN115423012BCN 115423012 BCN115423012 BCN 115423012BCN-115423012-B

Abstract

The invention provides a method and a device for determining a customer call cause, wherein the method comprises the steps of constructing feature vectors of all customers in customer service channels, constructing scoring vectors of the customers on all IVR service nodes, acquiring transaction data and behavior data of other channels except the customer service channels, screening target features for the customer service channels from the transaction data and the behavior data by using a double-layer random forest model, modeling the customer call cause according to the feature vectors, the scoring vectors and the target features, obtaining a customer call cause model, and predicting the customer call cause by the customer call cause model, so that the problem that customers with few customer service channel features in related technologies are predicted by collaborative filtering based on the model, and the prediction accuracy is low is solved, and the prediction accuracy is greatly improved.

Inventors

HUANG CHENG
SHI CHENYANG
Pei Yamin
LIU RUIQUN
YUAN CHUNLEI
XUE MING
ZOU HUA
CHEN XIAOLU
CUI JUAN

Assignees

中国光大银行股份有限公司

Dates

Publication Date: 20260512
Application Date: 20220829

Claims (9)

1. A method for determining a cause of a customer call, comprising: Constructing feature vectors of all clients in customer service channels, and constructing scoring vectors of all interactive voice response IVR service nodes of the clients, wherein the scoring vectors represent tendencies of the clients to call in the IVR service nodes; Acquiring transaction data and behavior data of channels except the customer service channel; Screening target features for the customer service channel from the transaction data and the behavior data by using a double-layer random forest model, wherein the target features have relevance with predicted targets of the customer service channel; modeling a customer call reason according to the feature vector, the scoring vector and the target feature to obtain a customer call reason model, and predicting the customer call reason according to the customer call reason model reason; And modeling a customer call reason according to the feature vector, the scoring vector and the target feature, wherein obtaining the customer call reason model comprises the following steps: dividing the feature vector and the target feature into a real-time feature and a non-real-time feature; Determining a similarity matrix between all clients using the non-real-time features; Supplementing the scoring vector according to the similarity matrix by using a collaborative filtering model based on a user to obtain a supplemented first scoring vector; splicing the first scoring vector with the real-time feature to obtain splicing vectors of all clients; And carrying out softmax regression modeling according to the spliced vectors of all clients to obtain the client electricity-making reason model.
2. The method according to claim 1, wherein the method further comprises: Determining the similarity of the non-real-time characteristics of the new client and the non-real-time characteristics of other clients; Supplementing the scoring vector according to the similarity to obtain a supplemented second scoring vector; Splicing the second scoring vector with the real-time characteristics of the new client to obtain a splicing vector of the new client; Inputting the splicing vector of the new customer into the customer calling cause model for prediction to obtain a prediction result; And updating the model parameters of softmax regression according to the prediction result to obtain an updated customer electricity-making reason model.
3. The method of claim 1, wherein using a two-layer random forest model to screen target features from the transaction data and the behavioral data for use by the customer service channel comprises: Determining the transaction data and the behavior data as alternative features; And screening target features for the customer service channel from the candidate features by using a double-layer random forest model.
4. A method according to claim 3, characterized in that the method further comprises: formulating a plurality of modeling targets associated with predicting a customer incoming call cause; Randomly mixing and disturbing the alternative features and the feature vectors to form an alternative feature library; randomly picking one modeling target from the plurality of modeling targets, and establishing the double-layer random forest model by using the alternative feature library.
5. The method according to claim 4, wherein the method further comprises: Randomly picking one modeling target from the modeling targets repeatedly, and establishing the double-layer random forest model by using the alternative feature library; each candidate feature is defined to appear at least once in the double-layer random forest model by probability adjustment means.
6. A method according to claim 3, wherein prior to using a two-layer random forest model to screen the candidate features for target features for use by the customer service channel, the method further comprises: Determining the importance of the features of each tree in each random forest in the double-layer random forest model by using the OOB error, and averaging the importance of all the features to obtain the feature importance of each feature; and taking the feature with the lowest feature importance in the original customer service channel as a threshold feature, and screening the target feature based on the feature importance of each feature.
7. A customer cause determination apparatus, comprising: The construction module is used for constructing feature vectors of all clients in customer service channels and constructing scoring vectors of the clients on all interactive voice response IVR service nodes, wherein the scoring vectors represent tendencies of the clients to call in the IVR service nodes; The acquisition module is used for acquiring transaction data and behavior data of channels except the customer service channel; The first screening module is used for screening target features for the customer service channel from the transaction data and the behavior data by using a double-layer random forest model, wherein the target features have relevance with a predicted target of the customer service channel; the modeling module is used for modeling the customer calling reason according to the feature vector, the grading vector and the target feature to obtain a customer calling reason model, and predicting the customer calling reason through the customer calling reason model reason; wherein the modeling module comprises: dividing the feature vector and the target feature into a real-time feature and a non-real-time feature; a determining submodule for determining a similarity matrix between all clients by using the non-real-time features; the supplementing sub-module is used for supplementing the scoring vector according to the similarity matrix by using a collaborative filtering model based on a user to obtain a supplemented first scoring vector; The splicing sub-module is used for splicing the first scoring vector with the real-time features to obtain splicing vectors of all clients; And establishing a mold module, wherein the mold module is used for carrying out softmax regression modeling on the spliced vectors of all clients to obtain the client electricity-making reason model.
8. A computer-readable storage medium, characterized in that the storage medium has stored therein a computer program, wherein the computer program is arranged to execute the method of any of the claims 1 to 6 when run.
9. An electronic device comprising a memory and a processor, characterized in that the memory has stored therein a computer program, the processor being arranged to run the computer program to perform the method of any of the claims 1 to 6.

Description

Customer calling reason determining method and device Technical Field The invention relates to the field of data processing, in particular to a method and a device for determining a customer calling reason. Background In order to relieve the service pressure of remote manual agents, an intelligent voice system is introduced to conduct question-answer interaction with clients. One common interaction scenario is when a customer calls in, the intelligent voice system asks the customer if he or she is going to transact a certain type of service, and if so, navigates the customer to the corresponding interactive voice response (INTERACTIVE VOICE RESPONSE, abbreviated IVR) service node. When a traditional collaborative filtering model based on a user or an article predicts the reason of the incoming call of a client, the similarity between users or services needs to be calculated according to the history of the client accessing a telephone customer service node, so that the client is required to call the telephone customer service at least once before making a recommendation for the service accessed by the client. Although the collaborative filtering based on the model does not need to calculate the similarity theoretically, the modeling, decomposing and restoring effects on the sparse matrix are poor, so that clients are actually required to call in customer service once, and the recommendation effect is credible. However, customer service has a large number of new card users or old customers who never call in each month, the customers have few characteristics in customer service channels, the service scoring matrix is very sparse, and the service rate of intelligent voice guessing is low because the service scoring matrix is not suitable for a traditional collaborative filtering model. Aiming at the problem that the prediction accuracy is low in the related art for the clients with few customer service channel characteristics, the model-based collaborative filtering prediction of the incoming call reasons of the clients is not provided. Disclosure of Invention The embodiment of the invention provides a method and a device for determining a customer call cause, which are used for at least solving the problems of low prediction accuracy of a customer with few customer service channel characteristics in the related technology, and predicting the customer call cause based on collaborative filtering of a model. According to an embodiment of the present invention, there is provided a method for determining a cause of customer call, including: Constructing feature vectors of all clients in customer service channels, and constructing scoring vectors of the clients to all IVR service nodes, wherein the scoring vectors represent tendencies of the clients to call in the IVR service nodes; Acquiring transaction data and behavior data of channels except the customer service channel; Screening target features for the customer service channel from the transaction data and the behavior data by using a double-layer random forest model, wherein the target features have relevance with predicted targets of the customer service channel; And modeling a customer call reason according to the feature vector, the scoring vector and the target feature to obtain a customer call reason model, and predicting the customer call reason according to the customer call reason model reason. Optionally, modeling the customer calling reason according to the feature vector, the scoring vector and the target feature, and obtaining the customer calling reason model includes: dividing the feature vector and the target feature into a real-time feature and a non-real-time feature; Determining a similarity matrix between all clients using the non-real-time features; Supplementing the scoring vector according to the similarity matrix by using a collaborative filtering model based on a user to obtain a supplemented first scoring vector; splicing the first scoring vector with the real-time feature to obtain splicing vectors of all clients; And carrying out softmax regression modeling according to the spliced vectors of all clients to obtain the client electricity-making reason model. Optionally, the method further comprises: Determining the similarity of the non-real-time characteristics of the new client and the non-real-time characteristics of other clients; Supplementing the scoring vector according to the similarity to obtain a supplemented second scoring vector; Splicing the second scoring vector with the real-time characteristics of the new client to obtain a splicing vector of the new client; Inputting the splicing vector of the new customer into the customer calling cause model for prediction to obtain a prediction result; And updating the model parameters of softmax regression according to the prediction result to obtain an updated customer electricity-making reason model. Optionally, the screening target features for use by the customer service chan