Develops an automated log processing approach combined with several representational learning frameworks to extract valuable signals from the entirety of lengthy log data of online behavioral data. Compared to conventional models using manual feature engineering to analyze only short data sequences, the improved models improves predictions and reduces time and costs required.