TWI657393B - Marketing customer group prediction system and method - Google Patents

Abstract

A marketing customer group prediction system includes a customer database storing a plurality of customer data and a model computing server. The model computing server includes a first computing module, a second computing module, and a third computing module. And an analysis module, the first operation module analyzes the customer data according to a first operation model and obtains a first predicted probability value, and the second operation module analyzes the customer data according to a second operation model and obtains A second prediction probability value, the third calculation module uses the first prediction probability value and the second prediction probability value to obtain a prediction range by using an analysis prediction model operation, and the analysis module analyzes from the customer data according to the prediction range Target marketing customers.

Description

Marketing customer group prediction system and method

The invention relates to a marketing customer group prediction system and a method thereof, and particularly to a marketing customer group prediction system and a method thereof using a mixed analysis of multiple supervised models.

For a long time, sellers have always been able to efficiently market their products to customers. For the financial industry, traditionally, they can only rely on the broadcast of various financial media or the advertisement of related financial products on the channel, or passively to each Customers who come to the branches of the banks are often ineffective, and most customers are troubled.

In the era of marketing advertising explosion, in order to reduce the waste of marketing resources and excessive interference with customers to cause customer complaints, how to improve marketing accuracy and find potential customers who may have demand has become an important issue.

Therefore, an object of the present invention is to provide a marketing customer group prediction system that utilizes a mixed analysis of multiple supervised models to improve marketing accuracy.

Therefore, the marketing customer group prediction system of the present invention includes a customer database storing a plurality of customer data and a model computing server. The model computing server includes a first computing module in communication with the customer database, and a customer A second computing module for database communication, a third computing module connected to the first computing module and the second computing module, and an analysis module connected to the third computing module, wherein the first computing module The group analyzes the customer data according to a first operation model and obtains a first predicted probability value. The second operation module analyzes the customer data according to a second operation model and obtains a second predicted probability value. The third operation mode The group obtains a prediction range by calculating the first prediction probability value and the second prediction probability value using an analysis prediction model, and the analysis module analyzes the target marketing customer group from the customer data according to the prediction range.

In one embodiment, the marketing customer group prediction system further includes a transceiver module connected to the analysis module for transmitting the target marketing customer group analyzed by the analysis module.

In one embodiment, the first operation model is a potential customer model, and the second operation model is a potential customer model.

In one embodiment, the analysis and prediction model is a threshold control model. The threshold control model calculates and selects an optimal recall rate according to the first predicted probability value and the second predicted probability value. To determine the prediction range.

In detail, the threshold control model first generates a complex decision tree according to the first predicted probability value and the second predicted probability, and calculates the retrieval rate of each decision tree, and then determines the prediction according to the decision tree with the highest retrieval rate. range.

In one embodiment, the calculation of the retrieval rate is based on the following calculation formula:

The threshold control model generates a predicted value (Predict) and an actual value (Actual) according to the first predicted probability value and the second predicted probability, and True Positive means that the predicted value is 1 and the actual value is 1, False Negative. Refers to a predicted value of 0 and an actual value of 1.

In addition, another object of the present invention is to provide a marketing customer group prediction method that uses a hybrid analysis of multiple supervised models to improve marketing accuracy.

Therefore, the marketing customer group prediction method of the present invention is applied to a prediction system. The method uses a complex operation model to analyze a plurality of pieces of customer data and obtain a prediction probability value respectively, and then uses the analysis probability to predict the prediction probability values. The model calculation obtains a prediction range, and the target marketing customer group is analyzed from the customer data according to the prediction range.

In one embodiment, the number of these computing models is two, which is a potential customer model and a potential customerless model.

In one embodiment, the analysis and prediction model is a threshold control model, and the threshold control model calculates and selects an optimal Recall Rate according to the prediction probability values to determine the prediction range.

In detail, the threshold control model first generates a complex decision tree according to the predicted probability values, calculates the retrieval rate of each decision tree, and then determines the prediction range according to the decision tree with the highest retrieval rate.

In one embodiment, the calculation of the retrieval rate is based on the following calculation formula:

Among them, the threshold control model generates a predicted value (Predict) and an actual value (Actual) respectively according to the predicted probability values, and True Positive means the predicted value is 1 and the actual value is 1, False Negative means the predicted value 0 and the actual value is 1.

The effect of the present invention is that it can greatly improve marketing accuracy, effectively find potential target marketing customer groups, and help marketing personnel to achieve the best customer operating results at the lowest cost.

Before the present invention is described in detail, it should be noted that in the following description, similar elements are represented by the same numbers.

Refer to FIG. 1, which is a schematic circuit block diagram of an embodiment of a marketing customer group prediction system according to the present invention. The marketing customer group prediction system 100 is used by a bank (or an enterprise) to analyze customers through a variety of monitoring models to find potential customers. Compared with the traditional method of using only one monitoring model to find potential customers, it can help marketers achieve the best customer operation results at the lowest cost.

The marketing customer group prediction system 100 includes a customer database 10 and a model calculation server 20 in communication with the customer database 10.

The customer database 10 stores a plurality of customer data, each of which includes personal information such as the customer's name, age, marital status, financial transaction records, loan transaction records, etc. The customer database 10 of this embodiment is a big data customer data system, of which Record the personal data of each customer from the past to the present. The customer database 10 can be integrated with the model calculation server 20 in the same electronic device, or it can be located on a different electronic device (server) and transmitted through the network communication method. Information.

The model computing server 20 includes a first computing module 21 in communication with the customer database 10, a second computing module 22 in communication with the customer database 10, a first computing module 21 and a second computing module 22 is connected to the third computing module 23, and an analysis module 24 is connected to the third computing module 23.

The first computing module 21 analyzes the customer data in the customer database 10 according to a first computing model and obtains a first predicted probability value. In this embodiment, it is described by taking an example that a bank wants to predict a customer group with a capital demand (that is, a loan demand), but it is not limited thereto. Therefore, the first operation model of this embodiment is a potential customer model, which can analyze the probability of a customer group with potential capital needs.

The second operation module 22 analyzes the customer data in the customer database 10 and obtains a second predicted probability value according to a second operation model. The second operation model of this embodiment is a potential no-demand customer model, which can analyze the potential Probability of customer groups without funding requirements.

The third calculation module 23 obtains a prediction range by calculating the first prediction probability value and the second prediction probability value using an analysis prediction model. The analysis prediction model in this embodiment is a threshold control model. The prediction principle of the threshold control model is Based on the probability values of the first and second computing models, the system calculates itself and automatically selects the optimal Recall Rate drop point to determine the threshold. The calculation formula is as follows:

The threshold control model will generate a predictive value and an actual value according to the first predicted probability value and the second predicted probability in cooperation with the analyzed customer data, and combine them to form an error matrix. An example of the error matrix is as follows:

Among them, True Positive means the predicted value is 1 and the actual value is 1; False Positive means the predicted value is 1 but the actual value is 0; False Negative means the predicted value is 0 but the actual value is 1; True Negative means the predicted value 0 and the actual value is 0. The so-called forecast value of 1 means that the model predicts that the client should have capital needs, and the forecast value of 0 means that the model predicts that the client should have no capital needs, while the actual value of 1 means that the client actually has Funding requirements (for example: the client has already borrowed money from a bank). An actual value of 0 means that the client does not actually have funding requirements (for example: the client has indicated that no loan is needed). Therefore, based on the above error matrix, by predicting the number of customers with loan demand (that is, the predicted value is 1) and the number of customers who actually have loans (that is, the actual value is 1), the withdrawal rate can be calculated to be 136 / (136 + 20) = 87.18%.

Furthermore, the threshold control model will first generate a plurality of Decision Trees, which means that the first prediction probability value generated by the first operation module 21 and the second prediction generated by the second operation module 22 The probability value is divided into multiple sections. In this embodiment, the number of decision trees is three, which are respectively defined as a first decision tree, a second decision tree, and a third decision tree. The rule of the first decision tree is that the first predicted probability value is less than 0.3 and the second predicted probability value is greater than 0.7; the rule of the second decision tree is that the first predicted probability value is 0.3 to 0.69 and the second predicted probability value is less than 0.3; The rules of the third decision tree are that the first predicted probability value is greater than 0.69 and the second predicted probability value is 0.3-0.7. The above is just an example, and the number of decision trees and the manner of dividing sections are not limited.

Next, the third computing module 23 will calculate the Recall Rate of each decision tree. Taking the above example as an example, it is assumed that the recall rate of the first decision tree is 77.86% and that of the second decision tree. The retrieval rate is 87.17%. The retrieval rate of the third decision tree is 69.12%. Finally, the third operation module 23 will use the decision tree with the highest retrieval rate (that is, the second decision tree) to determine the prediction range, that is, The first predicted probability value is 0.3 to 0.69 and the second predicted probability value is less than 0.3.

The analysis module 24 will select customers from the customer database 10 that meet the customer data of the first predicted probability value of 0.3 to 0.69 and the second predicted probability value of less than 0.3 according to the predicted range. These customers are the capital requirements (ie (With loan needs).

In addition, the analysis module 24 may also exclude a part of the customer list in the target marketing customer group according to an exclusion condition to obtain the final target marketing customer group. In this embodiment, the exclusion conditions are blacklisted customers, customers who do not receive advertising marketing, dead households, and bank stakeholders, but not limited to this. The analysis module 24 can communicate with a levy center server 30 to obtain the levy records of these customers in the target marketing customer group. If the customer's levy records are not good, it will be blacklisted and the analysis module 24 This customer will be excluded from the target marketing audience. Similarly, if the customer has indicated that he / she is unwilling to receive advertising and marketing through telephone voice or Internet settings, or the customer has died or is a bank stakeholder, the analysis module 24 will exclude these customers to find The best target marketing customer base.

It is added that because each model cannot be 100% correct, the marketing customer group prediction system 100 uses a cross-hybrid operation of multiple models (potential demand customer model and potential no demand customer model) to improve forecast accuracy. , The number of models used is not limited to two. As shown in FIG. 2, FIG. 2 is a total distribution diagram of probability values of the two models (potential demand customer model and potential no demand customer model). The horizontal axis is the customer list. Because the higher the first predicted probability value, the higher the probability that the customer has a loan demand, and the higher the second predicted probability value, the higher the probability that the customer has no loan demand. Therefore, if the sum of the two probability values has passed High (for example: close to 2) or too low (for example: close to 0), indicating that the error rate of the analysis result is higher (it is impossible for the same customer to have a loan demand and no loan demand), so the threshold control model will eliminate the totalizer The customers whose rate values deviate too much, and select the matching list, as shown in the dotted line in Figure 2.

Furthermore, if the marketing customer groups to be predicted are different (for example: forecasting customer groups with credit card demand), the computing models used by the first computing module 21 and the second computing module 22 can be adjusted correspondingly, and The amount of customer data that needs to be analyzed (only some or all of the customer data can be analyzed) and content can also be changed in accordance with different marketing customer groups.

Referring to FIG. 1 and FIG. 3, FIG. 3 is a flowchart of a marketing customer group prediction method according to the present invention. The method can be applied to the marketing customer group prediction system 100 or any prediction system for predicting the customer group. The prediction method uses a complex number. The computing model analyzes a plurality of customer data and obtains a predicted probability value, and then uses an analytical prediction model to calculate the predicted probability values to obtain a predicted range, and analyzes the target from the customer data according to the predicted range. Marketing customer base.

In detail, in step S10, the first operation module 21 and the second operation module 22 use the first operation model and the second operation model to analyze the plurality of customer data and obtain the first prediction probability value and the second prediction respectively. Probability value.

In step S20, the third calculation module 23 obtains a prediction range by calculating the first prediction probability value and the second prediction probability value using an analysis prediction model. With reference to FIG. 4, the analysis and prediction model (threshold control model) first generates a complex decision tree according to the first predicted probability value and the second predicted probability (eg, step S21), and then calculates the retrieval rate of each decision tree (eg, Step S22). Finally, the prediction range is determined according to the decision tree with the highest retrieval rate (such as step S23).

In step S30, the analysis module 24 analyzes the target marketing customer group from the customer data according to the predicted range and exclusion conditions. It is specifically stated that the exclusion conditions can be determined or adjusted by the bank (or corporate side), or the exclusion conditions are not required, that is, the target marketing customer group is analyzed from such customer data based on the forecast range only.

In addition, the model calculation server 20 may further include a transceiver module 25 connected to the analysis module 24. The analysis module 24 may output the analyzed target marketing customer group into a final list, and the final The list is sent back to the customer database 10 or to another electronic device (for example, an enterprise host).

In summary, the marketing customer group prediction system 100 of the present invention finds potential customers through a mixed analysis of multiple monitoring models, which is more accurate than traditional traditional methods that use only one monitoring model to find potential customers, which can help marketing personnel at the lowest cost. To achieve the best customer management results, it can indeed achieve the purpose of cost invention.

However, the above are only examples of the present invention. When the scope of implementation of the present invention cannot be limited in this way, any simple equivalent changes and modifications made in accordance with the scope of the patent application and the content of the patent specification of the present invention are still Within the scope of the invention patent.

100 ···· Marketing customer group prediction system 10 ····· Customer database 20 ····· Model operation server 21 ····· First operation module 22 ·· ····················································· 25 ··· Steps S10 ～ S30 of SJSC server Steps S21 ～ S23         

Other features and effects of the present invention will be clearly presented in the embodiments with reference to the drawings, in which: FIG. 1 is a schematic circuit block diagram of an embodiment of the marketing customer group prediction system of the present invention; FIG. 2 is a potential customer model Distribution diagram of the sum of the probability values of the potential and no-demand customer model; FIG. 3 is a flowchart of the marketing customer group prediction method of the present invention; and FIG. 4 is a calculation flowchart of the threshold control model of this embodiment.

Claims (7)

A marketing customer group prediction system includes: a customer database that stores a plurality of customer data; and a model operation server including a first operation module that communicates with the customer database, and a communication module that communicates with the customer database A second computing module, a third computing module connected to the first computing module and the second computing module, and an analysis module connected to the third computing module, wherein the first computing module Analyze the customer data according to a first operation model and obtain a first predicted probability value, the second operation module analyzes the customer data according to a second operation model and obtain a second predicted probability value, and the first The operation model is a potential customer model, which can analyze the probability of a customer group with potential capital needs to obtain the first predicted probability value. The second operation model is a potential customerless model, which can analyze potential To obtain the second predicted probability value by the probability of the customer group without capital requirements, the third computing module uses the first predicted probability value and the second predicted probability value to use an analytical prediction mode The operation obtains a prediction range, and the analysis prediction model is a threshold control model, and the threshold control model calculates and selects the best one retrieval rate based on the first prediction probability value and the second prediction probability value ( Recall Rate) to determine the forecast range, and the analysis module analyzes the target marketing customer group from the customer data according to the forecast range. The marketing customer group prediction system according to claim 1, wherein the threshold control model first generates a plurality of decision trees according to the first prediction probability value and the second prediction probability, and calculates the rebates of the decision trees respectively. Rate, and then the prediction range is determined according to the decision tree with the highest retrieval rate. The marketing customer group prediction system according to claim 2, wherein the calculation of the retrieval rate is based on the following calculation formula: Wherein, the threshold control model generates a predicted value (Predict) and an actual value (Actual) according to the first predicted probability value and the second predicted probability, and True Positive means that the predicted value is 1 and the actual value is 1, False Negative means that the predicted value is 0 and the actual value is 1. The marketing customer group prediction system according to claim 1, further comprising a transceiver module connected to the analysis module for transmitting the target marketing customer group analyzed by the analysis module. A marketing customer group forecasting method is applied to a forecasting system. The method uses a plurality of computing models to separately analyze a plurality of customer data and obtain a forecast probability value, and then obtains these forecast probability values by using an analysis forecasting model operation to obtain A forecast range, and the target marketing customer group is analyzed from the customer data according to the forecast range; wherein the computing models include a potential demand customer model and a potential no demand customer model, the potential demand customer model may be Analyze the probability of a customer group with potential funding needs to obtain a first predicted probability value. The potential no-demand customer model can analyze the probability of a customer group with potential non-funded needs to obtain a second predicted probability value; the analysis predicts The model is a threshold control model. The threshold control model calculates and selects an optimal recall rate according to the first predicted probability value and the second predicted probability value to determine the prediction range. The marketing customer group prediction method according to claim 5, wherein the threshold control model first generates a complex decision tree according to the first predicted probability value and the second predicted probability value, and calculates the response of each decision tree separately. Take the rate, and then determine the prediction range according to the decision tree with the highest retrieve rate. The marketing customer group prediction method according to claim 6, wherein the calculation of the retrieval rate is based on the following calculation formula: Wherein, the threshold control model generates a predicted value and an actual value respectively according to the first predicted probability value and the second predicted probability value, and True Positive means that the predicted value is 1 and the actual The value is 1, False Negative means that the predicted value is 0 and the actual value is 1.