CN113706195B - An online consumption behavior prediction method and system based on two-stage combination - Google Patents

Abstract

The invention provides an online consumption behavior prediction method and system based on two-stage combination, and relates to the technical field of consumption prediction. The invention comprises a user data input module, a user data processing module and a user data output module. 7 variables screened by the Logistic model can be used as input variables of a decision tree model, the decision tree model is used for carrying out secondary prediction on the behavior of browsing whether goods are consumed by a user, and the probability of purchasing the goods is output; the traditional statistical model is combined with the artificial intelligence method, so that the online consumption behavior prediction model has interpretation and accuracy, and a promotion scheme is formulated for merchants to provide suggestions.

Description

An online consumption behavior prediction method and system based on two-stage combination

Technical field

The present invention relates to the technical field of consumption prediction, and in particular to an online consumption behavior prediction method and system based on a two-stage combination.

Background technique

Qiu et al. (2017) proposed a two-stage online shopping behavior prediction framework. First, calculate the correlation between products. Then, support vector machine (SVM) and hierarchical Bayesian discrete choice model are used to calculate customer preferences for candidate products. Silahtaroglu (2015) et al. collected online shopping behavior and customer demographic information, and used decision trees and neural networks to predict whether users would purchase items in the shopping cart.

Traditional statistical econometric models have good robustness and interpretability, but their prediction accuracy is not high enough and cannot handle high-dimensional data; artificial intelligence methods have high accuracy and do not have strict requirements on data distribution, but their robustness is relatively low. Poor, and the machine learning method is a black box operation, resulting in low interpretability of the output results. Therefore, this paper establishes a two-stage combination model to predict online consumption behavior.

Contents of the invention

In view of the problems existing in the existing technology, the present invention provides an online consumption behavior prediction method and system based on a two-stage combination.

In order to solve the above technical problems, the present invention adopts the following technical solutions:

On the one hand, an online consumption behavior prediction method based on a two-stage combination specifically includes the following steps:

Step 1: Preprocess the historical data of the user's online browsing of goods and consumption behavior; the historical data includes quantitative indicator values and qualitative indicator values;

Step 1.1: Standardize the quantitative index values using the maximum and minimum standardization method;

Step 1.2: Score the qualitative index values according to the qualitative index scoring table;

Step 2: Screen indicator combinations based on the Logistic model;

In the logistic model, it is assumed that n independent indicator variables x={x ₁ ,x ₂ ,...,x _n }, binary response variables y∈{0,1}, y=1 indicates that a user purchases goods, y =0 indicates that a user does not purchase a certain product; assuming that the conditional probability p(y=1|x)=p is the probability of event y=1 occurring under the condition of sample x, then the logistic regression model is expressed as:

Where g(x)=w ₀ +w ₁ x ₁ +…+w _n x _n , w _n represents the weight of the nth independent indicator variable;

Taking the logarithm of the contrast ratio gives:

Find w ₀ , w ₁ ,..., w _n through formula 2. If the weight w _m of the indicator x _m (m=1,2,...,n) is not 0 and the corresponding significance level P value is less than 5%, If so, the indicator is retained; otherwise, the indicator is deleted to perform variable filtering.

Step 3: Use the variables filtered out by the logistic model as input variables of the decision tree model, use the decision tree model to make a secondary prediction of the user's behavior of browsing the product or not, and output the probability of purchasing the product;

The secondary prediction is, the given data set D={(x ₁ , y ₁ ), (x ₂ , y ₂ ),..., (x _i ,y _i ),..., (x _n ,y _n )} , where x _i is the input feature vector, y _i ∈{1,2,...,K} is a categorical variable containing K classes. In the question of whether the consumer purchases, K=2, i=1, 2,…,n, n is the sample size;

The Gini index is used to measure the uncertainty of the data set, and the definition is as shown in Equation (3):

For the two-category problem, that is, whether the consumer buys or not, K=2, then the Gini index is expressed as:

Gini(r)＝2r(1-r) (4)

where r represents the proportion of samples of the kth (k=1,2,…,K) category in node j (j=1,2,…,J);

The smaller the Gini index (Gini) value, the smaller the degree of uncertainty. Select the indicator with the smallest Gini coefficient to branch, then determine whether to purchase, and output the probability of whether the user purchases the product.

Step 4: Output the probability of whether the user purchases or not for merchants to make marketing decisions.

On the other hand, a two-stage combined online consumption behavior prediction system is used to implement the aforementioned two-stage combined online consumption behavior prediction method, including: a user data input module, a user data processing module, and a user data output module.

The user data input module inputs the user's online data of browsing products into the user data input module;

The user data processing module inputs the online data from the user data input module to the user data processing module, and the user data processing model processes the user's online data to predict the probability of the user's online browsing behavior and purchase of goods;

The user data output module and the user data processing module output the user's online consumption, that is, the probability of whether to purchase the product, and the seller makes decisions based on the user's probability of purchasing the product.

The beneficial effects produced by the present invention are:

The present invention proposes an online consumption behavior prediction method and system based on a two-stage combination, which has the following beneficial effects:

1. The present invention finds indicators that have a greater influence on whether users (consumers) purchase goods, and these indicators provide a basis for merchants to take corresponding marketing measures.

2. The present invention found that the peak consumption of products occurs in August, September, October and November. Therefore, the inventory is increased in these months to prevent shortages, and promotions are carried out in other months. At the same time, the shopping proportion of consumers who find this product directly enter the product homepage from other websites exceeds the shopping proportion who often jump back and forth between different products on the website. Therefore, it is recommended that the seller of this product increase the stickiness of the product so that consumers can directly Visit this product page.

3. The system and method used in the present invention can be applied to other products sold online. The system and method can be used for each product to find out the factors that affect product sales, and provide targeted marketing suggestions.

Description of the drawings

Figure 1 is an overall flow chart of a two-stage combined online consumption behavior prediction method in an embodiment of the present invention;

Figure 2 is an overall structural block diagram of a two-stage combined online consumption behavior prediction system in an embodiment of the present invention.

Detailed ways

Specific implementations of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the invention but are not intended to limit the scope of the invention.

On the one hand, an online consumption behavior prediction method based on a two-stage combination, as shown in Figure 1, specifically includes the following steps:

Step 1: Preprocess the historical data of the user's online browsing of goods and consumption behavior; the historical data includes quantitative indicator values and qualitative indicator values;

Step 1.1: Standardize the quantitative index values using the maximum and minimum standardization method;

Step 1.2: Score the qualitative indicator value according to the qualitative indicator scoring table; the higher the probability of purchasing goods in the indicator category, the higher the score. For example, the higher the probability of purchasing the goods in February in the first indicator Month, the higher the score.

Table 1 Qualitative indicator scoring table

Step 2: Screen indicator combinations based on the Logistic model;

In the logistic model, it is assumed that n independent indicator variables x={x ₁ ,x ₂ ,...,x _n }, binary response variables y∈{0,1}, y=1 indicates that a user purchases goods, y =0 means that a user does not purchase a certain product; let the conditional probability p(y=1|x)=p be the probability of event y=1 occurring under the condition of sample x, where the sample is the user and the event is the user purchasing the product, then The logistic regression model is expressed as:

Where g(x)=w ₀ +w ₁ x ₁ +…+w _n x _n , w _n represents the weight of the nth independent indicator variable;

Taking the logarithm of the contrast ratio gives:

Find w ₀ , w ₁ ,..., w _n through formula 2. If the weight w _m of the indicator x _m (m=1,2,...,n) is not 0 and the corresponding significance level P value is less than 5%, If so, the indicator is retained; otherwise, the indicator is deleted to perform variable filtering.

Step 3: Use the 7 variables screened out by the Logistic model as the input variables of the decision tree model, use the decision tree model to make a secondary prediction of the user's behavior of browsing the product or not, and output the probability of purchasing the product;

The decision tree model is a model with a top-down tree structure, which includes root nodes, internal nodes, leaf nodes and branches. When building a decision tree, the root node of the tree is the optimal attribute selected based on a certain criterion. Each internal node represents a test of an attribute, and each branch represents the result of the test. Leaf nodes represent classes or class distributions. After the root node, the optimal attribute among the remaining attributes is selected as the test for the next node. This process continues until all attributes have been compared, or there are no attributes remaining to further segment the sample.

Given the data set D={(x ₁ ,y ₁ ),(x ₂ ,y ₂ ),…,( _xi ,y _i ),…,(x _n ,y _n )}, where _xi is the input The feature vector, y _i ∈{1,2,...,K} is a categorical variable containing K classes. In the question of whether the consumer buys, K=2, i=1,2,...,n,n is the sample size; the modeling process of the classification tree is to recursively select an attribute variable and determine the conditions for segmentation based on this attribute. Commonly used standards for measuring the impurity Q _j of node j include misclassification rate, entropy (entory), Gini index (Gini), etc. The Gini index (Gini) is used here.

The Gini index is used to measure the uncertainty of the data set, and the definition is as shown in Equation (3):

For the two-category problem, that is, whether the consumer buys or not, K=2, then the Gini index is expressed as:

Gini(r)＝2r(1-r) (4)

where r represents the proportion of samples of the kth (k=1,2,…,K) category in node j (j=1,2,…,J);

The smaller the Gini index value, the smaller the degree of uncertainty. Choose the indicator with the smallest Gini coefficient to branch, and then determine whether to buy. The advantages of decision trees in terms of ease of understanding and intuitive expression are very significant; the difference from other analysis methods is that decision trees can quickly analyze a large amount of data-based and conventional data to obtain accurate results. The decision tree model can output the probability of whether the user purchases the product.

Step 4: The merchant inputs the user's real-time data into the user data input module, and then the user data processing model processes the user's input data. The user data output model outputs the probability of whether the user purchases, for the merchant to make marketing decisions.

On the other hand, an online consumption behavior prediction system based on a two-stage combination is used to implement the aforementioned online consumption behavior prediction method based on a two-stage combination, as shown in Figure 2, including: user data input module, user data processing Module, user data output module.

The user data input module inputs the user's online data of browsing products into the user data input module;

The user data processing module inputs the online data from the user data input module to the user data processing module, and the user data processing model processes the user's online data to predict the probability of the user's online browsing behavior and purchase of goods;

The user data output module and the user data processing module output the user's online consumption, that is, the probability of whether to purchase the product, and the seller makes decisions based on the user's probability of purchasing the product.

In this embodiment, the logistic regression model and the decision tree model are selected to evaluate online consumer purchasing behavior, and a two-stage combination model is established in a series manner. The logistic regression algorithm is combined with variable selection and parameter estimation to obtain the highest prediction accuracy with the smallest feature variables. It is very robust and interpretable when there is correlation between independent variables. The decision tree model has high prediction accuracy, but poor robustness, which is not conducive to the generalization of the model. At the same time, due to the existence of black box operations, it cannot be explained. This article uses the variables selected by logistic regression as input variables of the decision tree to obtain the prediction results of this model. The comparison table of the two models in this article is shown in 2.

Table 2 Accuracy comparison table of different models

From the perspective of accuracy, the accuracy of logistic regression is 0.8847, and the accuracy of the two-stage combination model (this model) is 0.9019. The lowest type 1 error rate is the combined model, and the lowest type 2 error rate is also the combined model. After the above analysis, in terms of accuracy, type I error rate and type II error rate, this model has the best effect, which proves that this model can effectively judge whether consumers purchase a certain product and provide decision-making basis for merchants' promotion plans.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that it can still be used Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent substitutions are made to some or all of the technical features; however, these modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the scope of the claims of the present invention.

Claims (3)

1. An online consumption behavior prediction method based on a two-stage combination, which is characterized by including the following steps: Step 1: Preprocess the historical data of the user's online browsing of goods and consumption behavior; the historical data includes quantitative indicator values and qualitative indicator values; Step 2: Screen indicator combinations based on the Logistic model; In the logistic model, it is assumed that n independent indicator variables x={x ₁ ,x ₂ ,...,x _n }, binary response variables y∈{0,1}, y=1 indicates that a user purchases goods, y =0 indicates that a user does not purchase a certain product; assuming that the conditional probability p(y=1|x)=p is the probability of event y=1 occurring under the condition of sample x, then the logistic regression model is expressed as: Where g(x)=w ₀ +w ₁ x ₁ +…+w _n x _n , w _n represents the weight of the nth independent indicator variable; Taking the logarithm of the contrast ratio gives: Find w ₀ , w ₁ ,..., w _n through formula 2. If the weight w _m of the indicator x _m is not 0 and the corresponding significance level P value is less than 5%, where m = 1, 2,..., n, If not, the indicator will be retained; otherwise, the indicator will be deleted to perform variable filtering; Step 3: Use the variables filtered out by the logistic model as input variables of the decision tree model, use the decision tree model to make a secondary prediction of the user's behavior of browsing the product or not, and output the probability of purchasing the product; The secondary prediction is, the given data set D={(x ₁ , y ₁ ), (x ₂ , y ₂ ),..., (x _i ,y _i ),..., (x _n ,y _n )} , where x _i is the input feature vector, y _i ∈{1,2,...,K} is a categorical variable containing K classes. In the question of whether the consumer purchases, K=2, i=1, 2,…,n, n is the sample size; The Gini index is used to measure the uncertainty of the data set, and the definition is as shown in Equation (3): For the two-category problem, that is, whether the consumer buys or not, K=2, then the Gini index is expressed as: Gini(r)＝2r(1-r) (4) where r represents the proportion of samples of the kth category in node j, where j=1,2,…,J, k=1,2,…,K; The smaller the value of the Gini index, the smaller the degree of uncertainty. Choose the indicator with the smallest Gini coefficient to branch, then determine whether to purchase, and output the probability of whether the user purchases the product; Step 4: Output the probability of whether the user purchases or not for merchants to make marketing decisions. 2. An online consumption behavior prediction method based on a two-stage combination according to claim 1, characterized in that said step 1 specifically includes the following steps: Step 1.1: Standardize the quantitative index values using the maximum and minimum standardization method; Step 1.2: Score the qualitative indicator values according to the qualitative indicator scoring table. 3. An online consumption behavior prediction system based on a two-stage combination, used to implement an online consumption behavior prediction method based on a two-stage combination according to claim 1, characterized in that it includes: a user data input module, a user data Processing module, user data output module; The user data input module inputs the user's online data of browsing products into the user data input module; The user data processing module inputs the online data from the user data input module to the user data processing module, and the user data processing model processes the user's online data to predict the probability of the user's online browsing behavior and purchase of goods; The user data output module and the user data processing module output the user's online consumption, that is, the probability of whether to purchase the product, and the seller makes decisions based on the user's probability of purchasing the product.