Predicting User Preference for Movies using NetFlix database

Online content and service providers deal with the problem of providing “relevant” content on a regular basis, especially due to the sheer volume of data available. This work deals with one such problem, namely, that of predicting user preference for movies using the NetFlix database. We present a memory-based Collaborative Filtering (CF) algorithm that learns the personality traits of the users in a features space we call the Latent Genre Space (LGS). This representation allows us to use traditional clustering algorithms in this space, and overcome one of the biggest problems in these works – that of different lengths of user feature vectors in the vote space. Inference techniques in this space are discussed, and a kd-tree based nearest-neighbor scheme is implemented.