Internet advertising is a fast growing market. Predicting the user behavior is the center of multi-billion dollar internet advertising industry. In this talk, I will introduce the business of Internet advertising and the popular statistical model in the industry. To implement the system, we need to resolve the issue of large-scale data. A general approach is to parallelize existing algorithm or develop new algorithms. In our product, we parallelize the algorithm of matrix multiplication with R so the system fits a model with 109 instances in 1 hour. I will share the experience of programming with R and the reasons and the solutions for some performance issues of R. We are also studying some algorithms from the field of online learning in computer science which might be more suitable for you to tackle the large-scale data. Therefore, I will also share some tips of implementing online algorithm such as stochastic gradient descent.