Customer recommendation transaction prediction method and system

Abstract

The invention relates to the technical field of data analysis, and provides a customer recommendation transaction prediction method and system, and the method comprises the following steps: collecting the historical feature data of a customer, carrying out the preprocessing of the collected feature data, and obtaining a training sample set matched with a corresponding customer code and a real estate code; performing significance verification on the feature variables influencing the recommendation transaction probability in the training sample set, and screening significant feature variables; constructing a recommendation transaction probability prediction model based on a binary logistic regression algorithm by taking the significant feature variable as an independent variable and taking whether a transaction is recommended or not as a dependent variable; inputting the training sample set into a recommendation transaction probability prediction model for training; and inputting historical feature data of a target customer into the trained recommendation transaction probability prediction model to obtain a recommendation transaction prediction probability.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and more specifically, to a customer recommendation transaction prediction method and system. Background Art

[0002] As an important pillar industry, the development of the real estate industry is of great significance to the development of the economy and society. However, with the growth of the domestic real estate market, competition is fierce, and traditional marketing methods can no longer effectively meet the needs of the market. In order to promote the development of the market, the identification of customer groups and the analysis of purchasing intentions are conducive to precision marketing.

[0003] In order to promote customer recommendation transactions, there are currently identification and marketing methods for real estate customer groups, which propose to use customer transaction prediction models to predict customer transaction probabilities. Specifically, based on customer behavior data such as customer ID, property ID visited, visit and browsing records, and transaction status, the LightGBM prediction model is selected to predict customer transaction probabilities and output customer transaction probabilities. However, the customer behavior data selected by this method is messy and contains a lot of noise, and does not consider the different degrees of influence of different customer behavior characteristics on transactions. The accuracy of transaction prediction probabilities obtained by directly inputting customer behavior data into the prediction model is low, which in turn affects the marketing effect. Summary of the invention

[0004] In order to overcome the defect in the above-mentioned prior art that a large amount of noise exists in customer behavior data, resulting in low accuracy of transaction prediction probability, the present invention provides a customer recommendation transaction prediction method and system.

[0005] In order to solve the above technical problems, the technical solution of the present invention is as follows: A customer recommendation transaction prediction method comprises the following steps: Collect historical feature data of customers, pre-process the collected feature data, and obtain a training sample set that matches the corresponding customer code and property code; Performing a significance check on the characteristic variables that affect the probability of recommended transaction in the training sample set, and screening the significant characteristic variables; Taking the significant characteristic variables as independent variables and whether to recommend a transaction as a dependent variable, a recommendation transaction probability prediction model based on a binary logistic regression algorithm is constructed; inputting the training sample set into the recommendation transaction probability prediction model for training; The historical feature data of the target customer is input into the trained recommendation transaction probability prediction model to obtain the recommendation transaction prediction probability.

[0006] Furthermore, the present invention also proposes a customer recommendation transaction prediction system, which applies the customer recommendation transaction prediction method proposed in the present invention. The system includes: Data collection module, used to collect historical characteristic data of customers; The data processing module is used to pre-process the collected feature data to obtain a training sample set matching the corresponding customer code and property code; A data screening module, used to perform significance check on the characteristic variables that affect the probability of recommended transaction in the training sample set, and screen significant characteristic variables; A prediction module, which is equipped with a recommendation transaction probability prediction model based on a binary logistic regression algorithm, and is used to input the historical feature data of the target customer into the trained recommendation transaction probability prediction model to output the recommendation transaction prediction probability; The recommendation transaction probability prediction model is trained using the training sample set, wherein the model uses the significant feature variable as an independent variable and whether a transaction is recommended as a dependent variable.

[0007] Compared with the prior art, the technical solution of the present invention has the following beneficial effects: The present invention performs a significance check on the characteristic variables that affect the probability of recommended transactions, and screens the significant characteristic variables for further constructing a recommended transaction probability prediction model based on a binary logistic regression algorithm, which is beneficial for filtering noise in sample data, improving the accuracy of the predicted probability of recommended transactions, and further improving the pertinence and effectiveness of marketing promotion. The present invention can make full use of the accumulated customer data, predict the customer's intention to recommend a deal, locate and recommend a list of customers with high deal intention, and can increase the pertinence and effectiveness of marketing promotion, reduce the cost of promotion, and improve the deal conversion rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 This is a flow chart of the customer recommendation transaction prediction method of Example 1.

[0009] Figure 2 Schematic diagram of the ROC curve of Example 1.

[0010] Figure 3 This is an architecture diagram of the customer recommendation transaction prediction system of Example 2. DETAILED DESCRIPTION

[0011] The accompanying drawings are only used for illustrative purposes and are not to be construed as limiting the present invention; It is understandable to those skilled in the art that some well-known descriptions may be omitted in the drawings.

[0012] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0013] Example 1 This embodiment proposes a customer recommendation transaction prediction method, such as Figure 1As shown, it is a flow chart of the customer recommendation transaction prediction method of this embodiment.

[0014] The customer recommendation transaction prediction method proposed in this embodiment includes the following steps: S100, collecting historical feature data of customers, preprocessing the collected feature data, and obtaining a training sample set matching corresponding customer codes and property codes; S200, performing significance check on the characteristic variables that affect the recommended transaction probability in the training sample set, and screening significant characteristic variables; S300, using the significant characteristic variables as independent variables and whether to recommend a deal as a dependent variable, constructing a recommended deal probability prediction model based on a binary logistic regression algorithm, and inputting the training sample set into the recommended deal probability prediction model for training; S400: Input the historical feature data of the target customer into the trained recommendation transaction probability prediction model to obtain the recommendation transaction prediction probability.

[0015] In this embodiment, a significance check is performed on the characteristic variables that affect the recommendation transaction probability, and significant characteristic variables are screened for further construction of a recommendation transaction probability prediction model based on a binary logistic regression algorithm, which is beneficial for filtering noise in sample data, improving the accuracy of the recommendation transaction prediction probability, and further improving the pertinence and effectiveness of marketing promotion.

[0016] In an optional embodiment, the historical characteristic data includes one or more of satisfaction data, behavior data, basic personal data and real estate data.

[0017] The satisfaction data includes satisfaction characteristics collected in the form of a questionnaire link; the satisfaction characteristics include one or more of property service satisfaction, sales service satisfaction, contract communication satisfaction, contract service satisfaction and delivery service satisfaction; The behavior data includes behavior characteristics obtained from the client interface; the behavior characteristics include whether to subscribe, the number of subscriptions, whether to forward articles, the number of forwarded articles, whether to report, the number of reports, whether to recommend a transaction, the number of recommended transactions, the life cycle when reporting, and the life cycle when the transaction is recommended for the first time. The basic personal data includes the basic characteristics of the customer obtained from the client interface; the basic characteristics of the customer include one or more of ID, age, gender and place of origin; The real estate data includes the real estate characteristics of the customer obtained from the sales end interface; the real estate characteristics include one or more of price, payment method, product type, area and apartment type.

[0018] As an exemplary description, in a specific implementation process, the selected historical feature data is shown in Table 1 below.

[0019] Table 1 Example of historical feature data

[0020] Since the historical feature data is directly collected through the survey questionnaire link, client interface, and sales interface, there is a lot of redundant information in the collected data. Directly inputting it into the prediction model for predicting the probability of recommended transaction will affect the accuracy of the prediction results. In this regard, this embodiment performs a significance check on the training samples composed of a large amount of redundant information, selects the feature variables that affect the probability of recommended transaction in the samples as significant feature variables, and further constructs and updates the recommended transaction probability prediction model to improve the accuracy of the prediction results.

[0021] Furthermore, in an optional embodiment, preprocessing the collected data includes the following steps: S110, setting customer codes and property codes for different customers; S120, associating and matching the collected feature data with the corresponding customer code and property code to form a training sample set consisting of a plurality of sample data; S130: Remove noise from the training sample set.

[0022] In this embodiment, a customer code is set according to the customer ID, and a property code is set according to the property owned by the customer. The collected feature data is then associated and matched with the customer code and property code of the customer to obtain a sample data about the customer, forming a training sample set.

[0023] Exemplarily, when noise removal is performed on the training samples in step S130, noise removal may be performed based on a preset rule strategy, or automatic data cleaning may be performed based on a clean-label algorithm or the like.

[0024] In an optional embodiment, the step of performing a significance check on the characteristic variables affecting the recommendation transaction probability in the training sample set and screening the significant characteristic variables includes: S210, selecting a verification method according to the variable type of the characteristic variable to obtain a significance value of the characteristic variable; wherein the Mann-Whitney U test method is used for continuous variables, and the Pearson chi-square test method is used for categorical variables; S220: retain the feature variables whose significance values ​​are less than a preset threshold as significant feature variables in the training sample set, and delete other feature variables in the training sample set.

[0025] It should be noted that continuous variables are variables that can take any value within a certain interval, while categorical variables are fixed-value variables that appear as incompatible categories or attributes.

[0026] In this embodiment, different significance test methods are adopted according to different variable types. The Mann-Whitney U test method is used for continuous variables such as "number of subscriptions", "number of forwarded articles", "age", "housing price", "area" and satisfaction, and the Pearson chi-square test method is used for categorical variables such as "whether to subscribe", "whether to forward articles", "gender", "payment method", "product type" and "housing type".

[0027] As an exemplary illustration, the preset threshold in step S220 is set to 0.05. When the significance value of the characteristic variable is less than 0.05, it means that the characteristic variable has a significant impact on the dependent variable of the recommended transaction probability prediction model and can be included in the model independent variable as a significant characteristic variable.

[0028] Furthermore, in an optional embodiment, the expression of the recommendation transaction probability prediction model based on the binary logistic regression algorithm is: ; ; in, P Predict the probability of a recommended transaction; is a linear combination of significant characteristic variables, Indicates n The value of a significant characteristic variable, is the bias term, Indicates n The regression coefficients of the significant characteristic variables; n is the total number of significant characteristic variables that have been screened.

[0029] When the predicted probability of transaction is recommended P If it is greater than the predicted classification standard value, the predicted result is that the customer is willing to recommend the transaction, otherwise the predicted result is that the customer is not willing to recommend the transaction.

[0030] Furthermore, in an optional embodiment, the method further comprises the following steps: S310: Using the ROC curve (Receiver Operating Characteristic Curve) to judge the prediction effect of the recommended transaction probability prediction model: When the area under the ROC curve is greater than the preset threshold, it indicates that the prediction effect of the current recommended transaction probability prediction model meets the requirements, and the prediction result output by the current recommended transaction probability prediction model is output as the recommended transaction prediction probability; Otherwise, the regression coefficient of the recommended transaction probability prediction model is adjusted using the maximum likelihood estimation method, the training sample set is input into the recommended transaction probability prediction model for retraining, and the recommended transaction probability prediction model is updated.

[0031] The ROC curve is obtained by plotting the True Positive Rate (TPR) and False Positive Rate (FPR) at different classification thresholds. TPR measures the proportion of samples correctly identified by the classifier among all samples that are actually positive examples, while FPR measures the proportion of samples that are incorrectly identified as positive examples among all samples that are actually negative examples. In the ROC curve, the vertical axis is TPR and the horizontal axis is FPR. Each point on the curve corresponds to a different classification threshold. By adjusting the threshold, different combinations of TPR and FPR can be obtained, thereby determining different points on the ROC curve. The area under the ROC curve (AUC) is an important indicator for evaluating model performance. The closer the AUC value is to 1, the better the model performance.

[0032] As an exemplary explanation, the preset threshold in step S310 is set to 0.6-0.75.

[0033] Optionally, the preset threshold in step S310 is set to 0.75, that is, when the AUC is greater than 0.75, it is judged that the prediction ability of the current recommended transaction probability prediction model meets the requirements.

[0034] As an exemplary explanation, when the AUC is less than or equal to a preset threshold, the regression coefficient, bias term and other parameters of the recommended transaction probability prediction model are combined with the log-likelihood loss function, and optimization training is performed using an optimization algorithm such as a gradient descent algorithm to update the recommended transaction probability prediction model.

[0035] In the specific implementation process, after collecting data and forming a training sample set, the characteristic variables that affect the probability of recommended transaction are tested for significance, and the non-parametric test results of the characteristic variables are obtained as shown in Table 2 below.

[0036] Table 2 Results of nonparametric tests of characteristic variables

[0037] It can be seen that the significant characteristic variables finally screened include: overall satisfaction, complaint handling satisfaction, maintenance service satisfaction, sales service satisfaction, house quality satisfaction, willingness to recommend, whether recommended, gender, age, number of properties, housing prices, area, life cycle at the time of reporting, whether subscribed, number of reports, and whether to forward the article.

[0038] The significant characteristic variables are used as independent variables, whether to recommend a transaction is used as the dependent variable, and the training sample set containing the significant characteristic variables is input into the recommended transaction probability prediction model based on the binary logistic regression algorithm for training, thereby obtaining a recommended transaction probability prediction model that has completed the training.

[0039] In a specific implementation process, the parameter estimation results shown in Table 3 below are obtained.

[0040] Table 3 Prediction model parameter estimation results

[0041] Furthermore, overall satisfaction, complaint handling satisfaction, maintenance service satisfaction, sales service satisfaction, housing quality satisfaction, recommendation willingness, whether recommended, gender, age, number of properties, housing price, area, life cycle at the time of reporting, whether subscribed, number of reports, and whether forwarded articles were used as significant feature variables and input into the recommendation transaction probability prediction model based on the binary logistic regression algorithm. The recommendation transaction probability prediction model was constructed, and its expression is: Logit( P )= =-0.49×R+0.27×(G==2)-0.0242×N-2.64×10-8+0.00311×M-0.0045×TJ+0.0117 ×(TJG==1)-1.93×(DY==1)-0.0148×Z+0.271×TS+0.248×WX+0.329×X+0.406×FZ; Where Logit( P ) represents the binary logistic regression function.

[0042] Finally, by substituting actual customer satisfaction, customer personal basic data, property data and customer behavior data parameters into the model, real-time prediction of customer recommendation transaction intention can be performed.

[0043] The ROC curve is further used to judge the prediction effect of the model, and the following is obtained: Figure 2 As shown in the ROC curve diagram, its AUC=0.9299>0.75 indicates that the current model has good prediction ability and can be directly applied to the prediction of recommended transaction probability.

[0044] This embodiment makes full use of the accumulated customer historical data, predicts the customer's intention to recommend a deal, and locates a list of customers with high intention to recommend a deal, which can increase the pertinence and effectiveness of marketing promotion, reduce promotion costs, and improve deal conversion rates.

[0045] Example 2 This embodiment proposes a customer recommendation transaction prediction system, which applies the customer recommendation transaction prediction method proposed in Embodiment 1. Figure 3 As shown, it is the architecture diagram of the customer recommendation transaction prediction system of this embodiment.

[0046] The customer recommendation transaction prediction system proposed in this embodiment includes: Data collection module, used to collect historical characteristic data of customers; The data processing module is used to pre-process the collected feature data to obtain a training sample set matching the corresponding customer code and property code; A data screening module, used to perform significance check on the characteristic variables that affect the probability of recommended transaction in the training sample set, and screen significant characteristic variables; A prediction module, which is equipped with a recommendation transaction probability prediction model based on a binary logistic regression algorithm, and is used to input the historical feature data of the target customer into the trained recommendation transaction probability prediction model to output the recommendation transaction prediction probability; The recommendation transaction probability prediction model is trained using the training sample set, wherein the model uses the significant feature variable as an independent variable and whether a transaction is recommended as a dependent variable.

[0047] It can be understood that the system of this embodiment corresponds to the method of the above-mentioned embodiment 1, and the options in the above-mentioned embodiment 1 are also applicable to this embodiment, so they are not described repeatedly here.

[0048] Example 3 This embodiment proposes a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the customer recommendation transaction prediction method proposed in Example 1.

[0049] Example 4 This embodiment proposes a storage medium on which computer-readable instructions are stored, wherein the computer-readable instructions, when executed by a processor, implement the steps of the customer recommendation transaction prediction method proposed in Embodiment 1.

[0050] Exemplarily, the storage medium includes, but is not limited to, a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0051] Exemplarily, the instructions, programs, code sets or instruction sets may be implemented using conventional programming languages.

[0052] Exemplarily, the processor includes but is not limited to a smart phone, a personal computer, a server, a network device, etc., and is used to execute all or part of the steps of the customer recommendation transaction prediction method described in Example 1.

[0053] The terms in the drawings are only used for illustrative purposes and are not to be construed as limiting the present invention; Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A customer recommendation transaction prediction method, characterized in that: The following steps are involved: Collect historical feature data of customers, pre-process the collected feature data, and obtain a training sample set that matches the corresponding customer code and property code; Performing a significance check on the characteristic variables that affect the probability of recommended transaction in the training sample set, and screening the significant characteristic variables; Taking the significant characteristic variables as independent variables and whether to recommend a deal as a dependent variable, a recommendation deal probability prediction model based on a binary logistic regression algorithm is constructed, and the training sample set is input into the recommendation deal probability prediction model for training; The historical feature data of the target customer is input into the trained recommendation transaction probability prediction model to obtain the recommendation transaction prediction probability.

2. The customer recommendation transaction prediction method according to claim 1, characterized in that: The preprocessing of the collected data comprises the following steps: Set customer codes and property codes for different customers; The collected feature data are associated and matched with the corresponding customer code and property code to form a training sample set consisting of several pieces of sample data; Noise removal is performed on the training sample set.

3. The customer recommendation transaction prediction method according to claim 1, characterized in that: The step of performing a significance check on the characteristic variables that affect the recommendation transaction probability in the training sample set and screening the significant characteristic variables includes: Selecting a verification method according to the variable type of the characteristic variable to obtain a significance value of the characteristic variable; wherein the Mann-Whitney U test method is used for continuous variables, and the Pearson chi-square test method is used for categorical variables; The feature variables whose significance values ​​are less than a preset threshold are retained in the training sample set as significant feature variables, and other feature variables in the training sample set are deleted.

4. The customer recommendation transaction prediction method according to claim 3, characterized in that: The expression of the recommended transaction probability prediction model based on the binary logistic regression algorithm is: in, P Predict the probability of a recommended transaction; is a linear combination of significant characteristic variables, Indicates n The value of a significant characteristic variable, Indicates n The regression coefficients of the significant characteristic variables; n is the total number of significant characteristic variables that have been screened.

5. The customer recommendation transaction prediction method according to any one of claims 1 to 4, characterized in that: The method further comprises the following steps: The ROC curve is used to judge the prediction effect of the recommended transaction probability prediction model: When the area under the ROC curve is greater than a preset threshold, it indicates that the prediction effect of the current recommended transaction probability prediction model meets the requirements, and the prediction result output by the current recommended transaction probability prediction model is output as the recommended transaction prediction probability; Otherwise, the regression coefficient of the recommended transaction probability prediction model is adjusted using the maximum likelihood estimation method to update the recommended transaction probability prediction model.

6. The customer recommendation transaction prediction method according to any one of claims 1 to 4, characterized in that: The historical characteristic data includes one or more of satisfaction data, behavior data, basic personal data and real estate data; wherein: The satisfaction data includes satisfaction characteristics collected in the form of a questionnaire link; the satisfaction characteristics include one or more of property service satisfaction, sales service satisfaction, contract communication satisfaction, contract service satisfaction and delivery service satisfaction; The behavior data includes behavior characteristics obtained from the client interface; the behavior characteristics include whether to subscribe, the number of subscriptions, whether to forward articles, the number of forwarded articles, whether to report, the number of reports, whether to recommend a transaction, the number of recommended transactions, the life cycle when reporting, and the life cycle when the transaction is recommended for the first time. The basic personal data includes the basic characteristics of the customer obtained from the client interface; the basic characteristics of the customer include one or more of ID, age, gender and place of origin; The real estate data includes the real estate characteristics of the customer obtained from the sales end interface; the real estate characteristics include one or more of price, payment method, product type, area and apartment type.

7. A customer recommendation transaction prediction system, using the customer recommendation transaction prediction method according to any one of claims 1 to 6, characterized in that: include: Data collection module, used to collect historical characteristic data of customers; The data processing module is used to pre-process the collected feature data to obtain a training sample set matching the corresponding customer code and property code; A data screening module, used to perform significance check on the characteristic variables that affect the probability of recommended transaction in the training sample set, and screen significant characteristic variables; A prediction module, which is equipped with a recommendation transaction probability prediction model based on a binary logistic regression algorithm, and is used to input the historical feature data of the target customer into the trained recommendation transaction probability prediction model to output the recommendation transaction prediction probability; The recommendation transaction probability prediction model is trained using the training sample set, wherein the model uses the significant feature variable as an independent variable and whether a transaction is recommended as a dependent variable.

8. The customer recommendation transaction prediction system according to claim 7, characterized in that: The data processing module performs the following steps in the process of preprocessing the collected feature data: Set customer codes and property codes for different customers; The collected feature data are associated and matched with the corresponding customer code and property code to form a training sample set consisting of several pieces of sample data; Noise removal is performed on the training sample set.

9. The customer recommendation transaction prediction system according to claim 7, characterized in that: The data screening module performs the following steps in the process of performing significance check on the characteristic variables affecting the recommended transaction probability in the training sample set: Selecting a verification method according to the variable type of the characteristic variable to obtain a significance value of the characteristic variable; wherein the Mann-Whitney U test method is used for continuous variables, and the Pearson chi-square test method is used for categorical variables; The feature variables whose significance values ​​are less than a preset threshold are retained in the training sample set as significant feature variables, and other feature variables in the training sample set are deleted.

10. The customer recommendation transaction prediction system according to any one of claims 7 to 9, characterized in that: The system also includes a prediction effect evaluation module, which is used to use the ROC curve to judge the prediction effect of the recommended transaction probability prediction model; when the area under the ROC curve is greater than a preset threshold, it means that the prediction effect of the current recommended transaction probability prediction model meets the requirements, and the prediction result output by the current recommended transaction probability prediction model is used as the recommended transaction prediction probability output; otherwise, the regression coefficient of the recommended transaction probability prediction model is adjusted using the maximum likelihood estimation method, and the recommended transaction probability prediction model is updated.

Images