Use Logistic Regression to Predict User’ Behaviors
Bing Huang, Yang Xiang, Zhen Huang · Applied Mechanics and Materials · 2014
Accurately capturing user behaviors over time is a great practical challenge in recommender system. Recently there is an Alibaba competition, which is to predict users’ purchase behave according to users’ previous four months behaviors, including click, purchase, add to collection and add to cart. We find that the brand is of crucial for user and click have a great impact for users. Besides, 1 / 5 users may buy according to previous interest. 4 / 5 users may choose other brands. While we focus on the predicting users for those brands which they have some behaviors. We extract some features, such as the number of click, the most last time for click, the number of purchase, the most last time for purchase. Then we use assemble model to predict users’ behaviors, which is better than other models.