Experimental study on item-based P-tree collaborative filtering algorithm for Netflix Prize

Recommendation system provides customer with personalized recommendations by analyzing historical transactions hence identifying the mostly possible item(s) that will be of interest to the customer. Collaborative Filtering (CF) algorithm plays an important role in the recommendation system. Item-based collaborative filtering is widely employed over user-based algorithm due to the computational complexity. Item similarity calculation is the first but the most important step in item-based collaborative filtering algorithm. In this paper, we analyze and implement several item similarity measurements in P-Tree format to Netflix Prize data set. Our experiments suggest that adjusted cosine based similarity provides much better RMSE than other item-based similarity measurements. The experimental results provide a guideline for our next step for Netflix Prize.