With the rapid development of Internet and Internet penetration into the automobile industry, more and more people search and browse automobile related information on the Internet before making a decision of purchase. This has formed a fertile ground to study automobile purchase intention by using user online activity data. In this paper, we focus on the task of predicting whether a user has the intention to purchase a particular make of automobile mainly based on the Deep Packet Inspection data from ISPs. We extracted 3-month user activity data from DPI data and collected automobile related information by the Web crawler on 5 leading automobile websites in China. The prediction problem was formulated as a typical classification problem in practice. And we paid a great deal of attention to the feature engineering. We proposed a feature engineering method by combining vector representation for user visiting sequence and statistical features related to users as well as automobiles. We trained various classification models with the combined features by traditional statistical methods and our method. The experimental results show that the features generated by our method perform better than the features only by statistical methods.
展开▼
