Taobao User Purchase Behavior Prediction And Feature Analysis Based On Ensemble Learning

In this paper, we construct a prediction model of user purchase behaviour and apply the SHAP interpretation method to analyze the influence of features to improve the model's reliability. Based on users' accurate historical behaviour data disclosed under the Taobao e-commerce platform, thi...

Full description

Saved in:

Bibliographic Details
Main Authors:	Chengjie, Yang, Wei, Qi
Format:	Conference Proceeding
Language:	eng
Subjects:	Analytical models Behavioral sciences Ensemble learning Forestry Hardware Prediction algorithms Predictive models Purchase behaviour prediction Ensemble Learning SHAP
Online Access:	Request full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	In this paper, we construct a prediction model of user purchase behaviour and apply the SHAP interpretation method to analyze the influence of features to improve the model's reliability. Based on users' accurate historical behaviour data disclosed under the Taobao e-commerce platform, this paper constructs features from five perspectives: user, item, item category, user-item interaction and user-category interaction. Three ensemble learning algorithms are used to construct the model, and the LightGBM model with the best effect is selected to predict user purchase behaviour. The SHAP interpretation method is used to analyze feature variables. The AUC value of the prediction model reaches about 84%, and the characteristics of the user-item interaction and the user significantly impact the prediction. Among them, the number of user-item behaviours on the day before the prediction date is the most crucial feature in predicting the purchasing behaviour of Taobao users. The SHAP interpretation method can consider the accuracy and interpretability of the machine learning prediction model, which is helpful to the following feature and model optimization work, provide a decision-making basis for users' accurate marketing, and make the results of the prediction model more effective and credible.
ISSN:	2472-8527