Large-scale online sequential behavior analysis with latent graphical model

Nowadays large amounts of data on peoples´ online activities, especially web-browsing data, have become available. Exploitation on such data can benefit a lot of real-life applications, such as user behavior identification, online customers classification and targeted advertisement. However, how to extract features on user behaviors from large amount of time series data is still a challenge due to its high complexity. In this work, we study the problem of inferring users´ instantaneous actions from their sequential online-shopping data. We propose a graphical hidden state model based on statistical features and integrate all available information sources to simulate the decision making process. Experimental results show that the proposed algorithm lead to nearly 30% of improvement on the million-clicks data sets.