Method and apparatus for determining reason for customer call
By constructing feature vectors and rating vectors for customers in customer service channels, and combining a two-layer random forest model with multi-channel data, the problem of low prediction accuracy for the reasons for customer calls due to the scarcity of customer service channel features was solved, achieving higher prediction accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA EVERBRIGHT BANK
- Filing Date
- 2022-08-29
- Publication Date
- 2026-05-12
AI Technical Summary
In existing technologies, model-based collaborative filtering has very low accuracy in predicting the reason for customer calls for customers with very few customer service channel characteristics.
We construct feature vectors and rating vectors for customers in customer service channels, acquire transaction and behavioral data outside of customer service channels, use a two-layer random forest model to filter target features, and combine a user-based collaborative filtering model and softmax regression to model the reasons why customers call.
It improves the predictive accuracy of the reasons why customers rarely call due to customer service channel characteristics, and enhances the predictive ability of the model by integrating multi-channel data.
Smart Images
Figure CN115423012B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing, and more specifically, to a method and apparatus for determining the reason for a customer's call. Background Technology
[0002] To alleviate the service pressure on remote human agents, intelligent voice systems have been introduced to interact with customers through question-and-answer sessions. A common interaction scenario is that when a customer calls in, the intelligent voice system asks if the customer wants to handle a certain type of service. If it guesses correctly, it navigates the customer to the corresponding Interactive Voice Response (IVR) service node.
[0003] Traditional user-based or item-based collaborative filtering models, when predicting the reasons for customer inbound calls, need to calculate the similarity between users or services based on the customer's historical access history to customer service service nodes. Therefore, it is required that the customer has called customer service at least once before recommending the services the customer has accessed. Although model-based collaborative filtering theoretically does not need to calculate similarity, its performance in modeling, decomposing, and restoring sparse matrices is also poor. Therefore, in practice, it is also necessary for the customer to have called customer service before for the recommendation effect to be reliable. However, customer service receives a large number of new cardholders or old customers who have never called before every month. These customers have very few characteristics in the customer service channel, and the service rating matrix is very sparse, making it unsuitable for traditional collaborative filtering models, resulting in a low service rate for intelligent voice guessing.
[0004] For customers with very few customer service channel characteristics in related technologies, model-based collaborative filtering for predicting the reason for customer inbound calls has very low accuracy, and no solution has yet been proposed. Summary of the Invention
[0005] This invention provides a method and apparatus for determining the reason for a customer's call, which at least solves the problem in related technologies where the accuracy of predicting the reason for a customer's call based on model-based collaborative filtering is very low for customers with very few characteristics of customer service channels.
[0006] According to an embodiment of the present invention, a method for determining the reason for a customer call is provided, comprising:
[0007] Construct feature vectors for all customers in the customer service channel, and construct rating vectors for customers on all IVR service nodes, where the rating vectors represent the customer’s tendency to call IVR service nodes.
[0008] Obtain transaction and behavioral data from channels other than the aforementioned customer service channels;
[0009] A two-layer random forest model is used to filter target features from the transaction data and the behavioral data for use by the customer service channel, wherein the target features are correlated with the prediction target of the customer service channel;
[0010] The customer call reason model is obtained by modeling the customer call reason based on the feature vector, the scoring vector and the target feature, and the customer call reason model is used to predict the customer call reason.
[0011] Optionally, the customer call reason model is obtained by modeling the customer call reason based on the feature vector, the scoring vector, and the target feature, including:
[0012] The feature vector and the target feature are divided into real-time features and non-real-time features;
[0013] The non-real-time features are used to determine the similarity matrix among all customers;
[0014] Using a user-based collaborative filtering model, the scoring vector is supplemented according to the similarity matrix to obtain the supplemented first scoring vector;
[0015] The first scoring vector is concatenated with the real-time features to obtain the concatenated vector for all customers;
[0016] Based on the spliced vectors of all customers, softmax regression modeling is performed to obtain the customer call reason model.
[0017] Optionally, the method further includes:
[0018] Determine the similarity between the non-real-time characteristics of new customers and the non-real-time characteristics of other customers;
[0019] The scoring vector is supplemented based on the similarity to obtain the supplemented second scoring vector;
[0020] The second scoring vector is concatenated with the real-time features of the new customer to obtain the concatenated vector of the new customer;
[0021] The concatenated vector of the new customer is input into the customer call reason model for prediction, and the prediction result is obtained.
[0022] The model parameters of the softmax regression are updated based on the prediction results to obtain the updated customer call reason model.
[0023] Optionally, using a two-layer random forest model to filter target features from the transaction data and the behavioral data for use by the customer service channel includes:
[0024] The transaction data and the behavioral data were selected as candidate features;
[0025] A two-layer random forest model is used to filter target features from the candidate features for use by the customer service channel.
[0026] Optionally, the method further includes:
[0027] Establish multiple modeling objectives that are associated with predicting the reasons for customer calls;
[0028] The candidate features are randomly mixed and shuffled with the feature vector to form a candidate feature library;
[0029] One modeling objective is randomly selected from the multiple modeling objectives, and the two-layer random forest model is built using the candidate feature library.
[0030] Optionally, the method further includes:
[0031] The modeling objective is randomly selected from the multiple modeling objectives multiple times, and the two-layer random forest model is built using the candidate feature library.
[0032] By using probability adjustment methods, each candidate feature is required to appear at least once in the two-layer random forest model.
[0033] Optionally, before using a two-layer random forest model to filter out target features from the candidate features for use by the customer service channel, the method further includes:
[0034] The importance of each feature in each tree of each random forest in the two-layer random forest model is determined by using the OOB error, and the importance of all features is averaged to obtain the feature importance of each feature.
[0035] The feature with the lowest feature importance in the original customer service channel is used as the threshold feature, and the target feature is selected based on the feature importance of each feature.
[0036] According to another embodiment of the present invention, a customer call reason determination device is also provided, comprising:
[0037] The module is used to construct feature vectors for all customers in the customer service channel and to construct rating vectors for customers on all IVR service nodes, wherein the rating vectors represent the customer’s tendency to call IVR service nodes.
[0038] The acquisition module is used to acquire transaction data and behavioral data from channels other than the customer service channel mentioned above.
[0039] The first screening module is used to use a two-layer random forest model to screen target features from the transaction data and the behavior data for use by the customer service channel, wherein the target features are correlated with the prediction target of the customer service channel;
[0040] The modeling module is used to model the reasons for customer calls based on the feature vector, the scoring vector, and the target features, to obtain a customer call reason model, and to predict the reasons for customer calls based on the reasons in the customer call reason model.
[0041] Optionally, the modeling module includes:
[0042] A segmentation submodule is used to divide the feature vector and the target feature into real-time features and non-real-time features;
[0043] A determination submodule is used to determine the similarity matrix among all customers using the non-real-time features;
[0044] The supplementary submodule is used to supplement the scoring vector according to the similarity matrix using a user-based collaborative filtering model to obtain a supplemented first scoring vector.
[0045] The splicing submodule is used to splice the first scoring vector with the real-time features to obtain the spliced vector of all customers;
[0046] The modeling submodule is used to perform softmax regression modeling on the concatenated vectors of all customers to obtain the customer call reason model.
[0047] Optionally, the device further includes:
[0048] The first determination module is used to determine the similarity between the non-real-time characteristics of a new customer and the non-real-time characteristics of other customers.
[0049] The supplementary module is used to supplement the scoring vector according to the similarity to obtain the supplemented second scoring vector;
[0050] The splicing module is used to splice the second scoring vector with the real-time features of the new customer to obtain the spliced vector of the new customer;
[0051] The prediction module is used to input the concatenated vector of the new customer into the customer call reason model for prediction and obtain the prediction result;
[0052] The update module is used to update the model parameters of the softmax regression based on the prediction results, so as to obtain the updated customer call reason model.
[0053] Optionally, the first filtering module includes:
[0054] A determination submodule is used to determine the transaction data and the behavior data as candidate features;
[0055] The filtering submodule is used to filter target features from the candidate features using a two-layer random forest model for use by the customer service channel.
[0056] Optionally, the device further includes:
[0057] The module defines multiple modeling objectives that are associated with predicting the reasons for customer calls;
[0058] The feature library composition module is used to randomly mix and shuffle the candidate features with the feature vector to form a candidate feature library;
[0059] A module is established to randomly select a modeling target from the multiple modeling targets and use the candidate feature library to establish the two-layer random forest model.
[0060] Optionally, the establishment module is further configured to repeatedly and randomly select a modeling target from the plurality of modeling targets, and establish the two-layer random forest model using the candidate feature library; and to limit each candidate feature to appear at least once in the two-layer random forest model by means of probability adjustment.
[0061] Optionally, the device further includes:
[0062] The second determination module is used to determine the importance of the features of each tree in each random forest in the two-layer random forest model using OOB error, and to average the importance of all features to obtain the feature importance of each feature.
[0063] The second filtering module is used to select the feature with the lowest feature importance in the original customer service channel as the threshold feature, and filter the target feature based on the feature importance of each feature.
[0064] According to yet another embodiment of the present invention, a computer-readable storage medium is also provided, wherein a computer program is stored therein, wherein the computer program is configured to perform the steps in any of the above method embodiments when executed.
[0065] According to yet another embodiment of the present invention, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.
[0066] This invention constructs feature vectors for all customers in the customer service channel and rating vectors for all IVR service nodes. It acquires transaction and behavioral data from channels other than the customer service channel. A two-layer random forest model is used to filter target features from the transaction and behavioral data for use by the customer service channel. Based on the feature vectors, rating vectors, and target features, a customer call reason model is created, and the call reason model is used to predict the call reason. This addresses the problem in related technologies where collaborative filtering based on models results in low accuracy for predicting call reasons for customers with very few customer service channel features. By using a two-layer random forest model to filter target features from transaction and behavioral data and integrating them with the customer service channel's feature and rating vectors to create a customer call reason model, the invention facilitates prediction of call reasons for customers with very few customer service channel features, significantly improving prediction accuracy. Attached Figure Description
[0067] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:
[0068] Figure 1 This is a hardware structure block diagram of a mobile terminal for a customer call reason determination method according to an embodiment of the present invention.
[0069] Figure 2 This is a flowchart of a method for determining the reason for a customer call according to an embodiment of the present invention;
[0070] Figure 3 A flowchart of a remote customer call cause modeling method according to an embodiment of the present invention;
[0071] Figure 4 This is a flowchart of the alternative feature selection process according to an embodiment of the present invention;
[0072] Figure 5 This is a schematic diagram illustrating the establishment of a customer call reason model according to an embodiment of the present invention;
[0073] Figure 6 This is a block diagram of a customer call reason determination device according to an embodiment of the present invention. Detailed Implementation
[0074] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present application can be combined with each other.
[0075] It should be noted that the terms "first," "second," etc., in the specification, claims, and drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.
[0076] Example 1
[0077] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Taking its operation on a mobile terminal as an example, Figure 1 This is a hardware structure block diagram of a mobile terminal for a customer call reason determination method according to an embodiment of the present invention, as shown below. Figure 1 As shown, a mobile terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. Optionally, the mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the mobile terminal described above. For example, the mobile terminal may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.
[0078] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the customer call reason determination method in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
[0079] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the mobile terminal's communication provider. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module used for wireless communication with the Internet.
[0080] This embodiment provides a method for determining the reason for a customer call, which operates on the aforementioned mobile terminal or network architecture. Figure 2 This is a flowchart of a method for determining the reason for a customer call according to an embodiment of the present invention, such as... Figure 2 As shown, this process, applied to service providers, includes the following steps:
[0081] Step S202: Construct feature vectors for all customers in the customer service channel, and construct rating vectors for all IVR service nodes, wherein the rating vectors represent the customer’s tendency to call into the IVR service node;
[0082] Step S204: Obtain transaction data and behavioral data from channels other than the customer service channel;
[0083] Step S206: Use a two-layer random forest model to filter target features from the transaction data and the behavior data for use by the customer service channel, wherein the target features are correlated with the prediction target of the customer service channel;
[0084] Step S208: Model the customer call reason based on the feature vector, the scoring vector and the target feature to obtain the customer call reason model, and predict the customer call reason based on the customer call reason model.
[0085] Through the above steps S202 to S208, the problem of low prediction accuracy for customers with very few customer service channel features when using collaborative filtering based on models to predict the reason for customer calls can be solved. By using a two-layer random forest model to filter target features from transaction data and behavioral data for use by customer service channels, and integrating them with the feature vector and rating vector of customer service channels to create a model of the reason for customer calls, it is easier to predict the reason for calls of customers with very few customer service channel features, and the prediction accuracy is greatly improved.
[0086] In this embodiment of the invention, step S208, which involves modeling the customer call reason based on the feature vector, the scoring vector, and the target feature to obtain the customer call reason model, may specifically include: dividing the feature vector and the target feature into real-time features and non-real-time features; using the non-real-time features to determine a similarity matrix among all customers; using a user-based collaborative filtering model to supplement the scoring vector based on the similarity matrix to obtain a supplemented first scoring vector; concatenating the first scoring vector with the real-time feature to obtain a concatenated vector for all customers; and performing softmax regression modeling based on the concatenated vectors for all customers to obtain the customer call reason model.
[0087] In an optional embodiment, the method further includes: determining the similarity between the non-real-time features of a new customer and the non-real-time features of other customers; supplementing the scoring vector according to the similarity to obtain a supplemented second scoring vector; concatenating the second scoring vector with the real-time features of the new customer to obtain a concatenated vector of the new customer; inputting the concatenated vector of the new customer into the customer call reason model for prediction to obtain a prediction result; and updating the model parameters of the softmax regression according to the prediction result to obtain an updated customer call reason model.
[0088] In this embodiment of the invention, step S206 may specifically include: determining the transaction data and the behavior data as candidate features; and using a two-layer random forest model to filter out target features from the candidate features for use by the customer service channel.
[0089] The importance of each feature in the current softmax regression model is calculated using a feature oscillation method. This facilitates post-evaluation and interpretation with business personnel. The arithmetic mean of the errors across all clients is taken as the feature importance, and the importance of all features is normalized to obtain the feature importance. This feature importance is the post-evaluation importance after the modeling is completed. This process is performed after the two-layer random forest model is completed, to examine the importance of the used features, and to rank them, providing a reference for subsequent feature selection optimization. End-to-end feature oscillation is performed using the final softmax regression model, and the error of the prediction results is then examined to infer the influence of the original features on the final result.
[0090] In an optional embodiment, before using a two-layer random forest model to select target features for the customer service channel from the candidate features, the method further includes: determining the importance of features in each tree of each random forest in the two-layer random forest model using out-of-bag (OOB) error; averaging the importance of all features to obtain the feature importance of each feature; using the feature with the lowest feature importance in the original customer service channel as a threshold feature; and selecting the target features based on the feature importance of each feature. This feature selection process in the two-layer random forest model aims to calculate the domain importance of features and establish a feature library of all possible inclusions in the model before modeling. Features with higher importance calculated in this way may perform well in modeling objectives across multiple customer service domains. Combined with the method of using the last-ranked feature importance in the customer service channel as a threshold to extract all candidate features with importance greater than the threshold, a feature library for inclusion in the model can be established.
[0091] In another optional embodiment, the method further includes: formulating multiple modeling objectives associated with predicting the reasons for customer calls; randomly mixing and shuffling the candidate features and the feature vectors to form a candidate feature library; randomly selecting one modeling objective from the multiple modeling objectives and using the candidate feature library to build the two-layer random forest model; furthermore, the two-layer random forest model can be built by repeatedly randomly selecting one modeling objective from the multiple modeling objectives and using the candidate feature library; and by using probability adjustment methods to ensure that each candidate feature appears at least once in the two-layer random forest model, thereby further improving the accuracy of the two-layer random forest model.
[0092] Figure 3 A flowchart of the remote customer call cause modeling method according to an embodiment of the present invention is shown below. Figure 3 As shown, it includes the following steps:
[0093] Step S301: Construct initial feature vectors and rating vectors. Specifically, construct feature vectors for all customers in the customer service channel (e.g., the number of times they used human assistance in the past month) and rating vectors for all IVR service nodes. The rating vector represents a customer's tendency to call that node and is initialized using a feature-weighted approach. If a customer has never called a certain node, that rating is left blank.
[0094] Step S302 involves introducing data from other channels. A two-layer random forest model is used in conjunction with traditional methods to screen candidate features. Transaction and behavioral data from various channels, including mobile banking, online banking, bank counters, telephone, and WeChat, can be introduced as candidate features. The two-layer random forest model is then used to select new features suitable for use in the customer service channel. These new features must have a certain correlation with the prediction target in the customer service channel; otherwise, it will lead to significant errors in calculating customer similarity. Figure 4 This is a flowchart of the candidate feature selection process according to an embodiment of the present invention, such as... Figure 4 As shown, it includes the following steps:
[0095] Step S401: Define the correlation modeling objectives. Define several modeling objectives that are related to predicting the reasons for customer calls, such as: whether the customer will call the mobile banking customer service next month, whether the customer will enter the human channel next month, etc.
[0096] Step S402: Randomly mix customer service channel features with features from other channels to form a candidate feature library.
[0097] Step S403: Randomly select a modeling target and establish a random forest model. That is, randomly select a target from the modeling targets and use the candidate feature library to establish a random forest model.
[0098] Step S404: Determine whether the number of models is greater than or equal to a preset threshold. If the determination result is yes, proceed to step S405; otherwise, return to step S403.
[0099] The above modeling process is repeated multiple times. Furthermore, through probability adjustment, each candidate feature is guaranteed to appear at least once in a given random forest model.
[0100] Step S405: Using OOB error, calculate the importance of each feature in each model and take the average. That is, use OOB error to calculate the feature importance of each tree in each random forest. The average importance of a feature across all random forests and all trees is taken as the importance of that feature.
[0101] Step S406: Use the feature with the lowest importance of the customer service channel as the threshold feature to filter all candidate features that can be included in the model.
[0102] Specifically, the importance of the last feature in the customer service channel can be used as a threshold to extract all candidate features whose importance is greater than the threshold.
[0103] The features selected in the previous step are then filtered again using traditional feature selection methods to finalize the features for inclusion in the model. These methods include, but are not limited to, chi-square test, information entropy, and Pearson correlation coefficient.
[0104] Step S303: The final input features are divided into real-time features and non-real-time features, and modeling is performed according to the following scheme:
[0105] Figure 5 This is a schematic diagram illustrating the establishment of a customer call reason model according to an embodiment of the present invention, such as... Figure 5 As shown, a similarity matrix among all customers is calculated using non-real-time features. Then, a user-based collaborative filtering model is used to complete the rating matrix for all customers' IVR service nodes. The rating vector of a single customer will replace the non-real-time features as the input feature in the model.
[0106] A customer's rating vector is concatenated with their real-time features, and a softmax regression model is then performed. The model is trained and its parameters are tuned based on customers with a history of inbound calls.
[0107] Step S304: Update the model parameters based on the daily forecast results;
[0108] When making predictions, new customers also need to first calculate their similarity to other customers to supplement the rating vector, and then concatenate it with real-time features to perform prediction calculations.
[0109] We collect prediction feedback daily and update the softmax regression model parameters accordingly.
[0110] When delivering a model to business stakeholders, it's often necessary to explain the role of each input feature, thus requiring the calculation of feature importance. In this model, there are two methods for calculating feature importance:
[0111] First, we will explain the feature importance obtained using the OOB method in the previous feature selection process as the feature importance in the general customer service field.
[0112] Secondly, using the final model, a small perturbation is applied to a single feature of a particular customer, and the resulting error in the prediction result is calculated. The arithmetic mean of the errors for all customers is then taken as the importance of that feature. Finally, the importance of all features is normalized to obtain the final feature importance. This feature importance is more suitable for the modeling objectives of this study.
[0113] According to another embodiment of the present invention, a device for determining the reason for a customer call is also provided. Figure 6 This is a block diagram of a customer call reason determination device according to an embodiment of the present invention, such as... Figure 6 As shown, it includes:
[0114] Module 62 is used to construct feature vectors for all customers in the customer service channel and to construct rating vectors for all IVR service nodes, wherein the rating vectors represent the customer’s tendency to call into the IVR service node.
[0115] The acquisition module 64 is used to acquire transaction data and behavioral data from channels other than the customer service channel.
[0116] The first screening module 66 is used to use a two-layer random forest model to screen target features from the transaction data and the behavior data for use by the customer service channel, wherein the target features are correlated with the prediction target of the customer service channel;
[0117] The modeling module 68 is used to model the customer call reason based on the feature vector, the scoring vector and the target feature, to obtain the customer call reason model, and to predict the customer call reason based on the customer call reason model.
[0118] Optionally, the modeling module 668 includes:
[0119] A segmentation submodule is used to divide the feature vector and the target feature into real-time features and non-real-time features;
[0120] A determination submodule is used to determine the similarity matrix among all customers using the non-real-time features;
[0121] The supplementary submodule is used to supplement the scoring vector according to the similarity matrix using a user-based collaborative filtering model to obtain a supplemented first scoring vector.
[0122] The splicing submodule is used to splice the first scoring vector with the real-time features to obtain the spliced vector of all customers;
[0123] The modeling submodule is used to perform softmax regression modeling on the concatenated vectors of all customers to obtain the customer call reason model.
[0124] Optionally, the device further includes:
[0125] The first determination module is used to determine the similarity between the non-real-time characteristics of a new customer and the non-real-time characteristics of other customers.
[0126] The supplementary module is used to supplement the scoring vector according to the similarity to obtain the supplemented second scoring vector;
[0127] The splicing module is used to splice the second scoring vector with the real-time features of the new customer to obtain the spliced vector of the new customer;
[0128] The prediction module is used to input the concatenated vector of the new customer into the customer call reason model for prediction and obtain the prediction result;
[0129] The update module is used to update the model parameters of the softmax regression based on the prediction results, so as to obtain the updated customer call reason model.
[0130] Optionally, the first filtering module 66 includes:
[0131] A determination submodule is used to determine the transaction data and the behavior data as candidate features;
[0132] The filtering submodule is used to filter target features from the candidate features using a two-layer random forest model for use by the customer service channel.
[0133] Optionally, the device further includes:
[0134] The module defines multiple modeling objectives that are associated with predicting the reasons for customer calls;
[0135] The feature library composition module is used to randomly mix and shuffle the candidate features with the feature vector to form a candidate feature library;
[0136] A module is established to randomly select a modeling target from the multiple modeling targets and use the candidate feature library to establish the two-layer random forest model.
[0137] Optionally, the establishment module is further configured to repeatedly and randomly select a modeling target from the plurality of modeling targets, and establish the two-layer random forest model using the candidate feature library; and to limit each candidate feature to appear at least once in the two-layer random forest model by means of probability adjustment.
[0138] Optionally, the device further includes:
[0139] The second determination module is used to determine the importance of the features of each tree in each random forest in the two-layer random forest model using OOB error, and to average the importance of all features to obtain the feature importance of each feature.
[0140] The second filtering module is used to select the feature with the lowest feature importance in the original customer service channel as the threshold feature, and filter the target feature based on the feature importance of each feature.
[0141] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.
[0142] Embodiments of the present invention also provide a computer-readable storage medium storing a computer program configured to perform the steps in any of the above method embodiments when executed.
[0143] Optionally, in this embodiment, the storage medium may be configured to store a computer program for performing the following steps:
[0144] S1, construct feature vectors for all customers in the customer service channel, and construct rating vectors for all IVR service nodes, where the rating vectors represent the customer’s tendency to call into the IVR service node;
[0145] S2, Obtain transaction data and behavioral data from channels other than the customer service channel;
[0146] S3, using a two-layer random forest model to filter target features from the transaction data and the behavior data for use by the customer service channel, wherein the target features are correlated with the prediction target of the customer service channel;
[0147] S4. Based on the feature vector, the scoring vector, and the target feature, model the reason for customer calls to obtain a customer call reason model, and predict the reason for customer calls based on the reason for customer calls in the customer call reason model.
[0148] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.
[0149] Embodiments of the present invention also provide an electronic device including a memory and a processor, the memory storing a computer program and the processor being configured to run the computer program to perform the steps in any of the above method embodiments.
[0150] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.
[0151] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:
[0152] S1, construct feature vectors for all customers in the customer service channel, and construct rating vectors for all IVR service nodes, where the rating vectors represent the customer’s tendency to call into the IVR service node;
[0153] S2, Obtain transaction data and behavioral data from channels other than the customer service channel;
[0154] S3, using a two-layer random forest model to filter target features from the transaction data and the behavior data for use by the customer service channel, wherein the target features are correlated with the prediction target of the customer service channel;
[0155] S4. Based on the feature vector, the scoring vector, and the target feature, model the reason for customer calls to obtain a customer call reason model, and predict the reason for customer calls based on the reason for customer calls in the customer call reason model.
[0156] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.
[0157] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.
[0158] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for determining the reason for a customer's call, characterized in that, include: Construct feature vectors for all customers in the customer service channel, and construct customer rating vectors for all interactive voice response (IVR) service nodes, where the rating vectors represent the customer’s tendency to call into IVR service nodes. Obtain transaction and behavioral data from channels other than the aforementioned customer service channels; A two-layer random forest model is used to filter target features from the transaction data and the behavioral data for use by the customer service channel, wherein the target features are correlated with the prediction target of the customer service channel; The customer call reason model is obtained by modeling the customer call reason based on the feature vector, the scoring vector and the target feature, and the customer call reason model is used to predict the customer call reason. The customer call reason model is obtained by modeling the customer call reason based on the feature vector, the scoring vector, and the target feature, including: The feature vector and the target feature are divided into real-time features and non-real-time features; The non-real-time features are used to determine the similarity matrix among all customers; Using a user-based collaborative filtering model, the scoring vector is supplemented according to the similarity matrix to obtain the supplemented first scoring vector; The first scoring vector is concatenated with the real-time features to obtain the concatenated vector for all customers; Based on the spliced vectors of all customers, softmax regression modeling is performed to obtain the customer call reason model.
2. The method according to claim 1, characterized in that, The method further includes: Determine the similarity between the non-real-time characteristics of new customers and the non-real-time characteristics of other customers; The scoring vector is supplemented based on the similarity to obtain the supplemented second scoring vector; The second scoring vector is concatenated with the real-time features of the new customer to obtain the concatenated vector of the new customer; The concatenated vector of the new customer is input into the customer call reason model for prediction, and the prediction result is obtained. The model parameters of the softmax regression are updated based on the prediction results to obtain the updated customer call reason model.
3. The method according to claim 1, characterized in that, Using a two-layer random forest model, target features for use by the customer service channel are selected from the transaction data and the behavioral data, including: The transaction data and the behavioral data were selected as candidate features; A two-layer random forest model is used to filter target features from the candidate features for use by the customer service channel.
4. The method according to claim 3, characterized in that, The method further includes: Establish multiple modeling objectives that are associated with predicting the reasons for customer calls; The candidate features are randomly mixed and shuffled with the feature vector to form a candidate feature library; One modeling objective is randomly selected from the multiple modeling objectives, and the two-layer random forest model is built using the candidate feature library.
5. The method according to claim 4, characterized in that, The method further includes: The modeling objective is randomly selected from the multiple modeling objectives multiple times, and the two-layer random forest model is built using the candidate feature library. By using probability adjustment methods, each candidate feature is required to appear at least once in the two-layer random forest model.
6. The method according to claim 3, characterized in that, Before using a two-layer random forest model to filter target features from the candidate features for use by the customer service channel, the method further includes: The importance of each feature in each tree of each random forest in the two-layer random forest model is determined by using the OOB error, and the importance of all features is averaged to obtain the feature importance of each feature. The feature with the lowest feature importance in the original customer service channel is used as the threshold feature, and the target feature is selected based on the feature importance of each feature.
7. A device for determining the reason for a customer call, characterized in that, include: The module is used to construct feature vectors for all customers in the customer service channel and to construct customer rating vectors for all interactive voice response (IVR) service nodes, wherein the rating vectors represent the customer’s tendency to call into the IVR service node. The acquisition module is used to acquire transaction data and behavioral data from channels other than the customer service channel mentioned above. The first screening module is used to use a two-layer random forest model to screen target features from the transaction data and the behavior data for use by the customer service channel, wherein the target features are correlated with the prediction target of the customer service channel; The modeling module is used to model the reasons for customer calls based on the feature vector, the scoring vector and the target features, to obtain a customer call reason model, and to predict the reasons for customer calls based on the reasons in the customer call reason model. The modeling module includes: A segmentation submodule is used to divide the feature vector and the target feature into real-time features and non-real-time features; A determination submodule is used to determine the similarity matrix among all customers using the non-real-time features; The supplementary submodule is used to supplement the scoring vector according to the similarity matrix using a user-based collaborative filtering model to obtain a supplemented first scoring vector. The splicing submodule is used to splice the first scoring vector with the real-time features to obtain the spliced vector of all customers; The modeling submodule is used to perform softmax regression modeling on the concatenated vectors of all customers to obtain the customer call reason model.
8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, wherein the computer program is configured to execute the method described in any one of claims 1 to 6 when it is run.
9. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method of any one of claims 1 to 6.