Online Clustering for Collaborative Filtering

We study two online clustering methods for collaborative filtering. In the first method, we assume that each user is equally likely to belong to one of m clusters of users and that the user’s rating for each item is generated randomly according to a distribution that depends on the item and the cluster that the user belongs to. In the second method, we assume that each user is equally likely to belong to one ofm clusters of users while each item is equally likely to belong to one of n clusters of items. The rating for a user item pair is generated randomly according to a distribution that depends on the cluster that the user belongs to and the cluster that the item belongs to. We derive performance bounds for Bayesian sequential probability assignment for these two methods to elucidate the trade offs involved in using these methods. Bayesian sequential probability assignment does not appear to be computationally tractable for these model classes. We propose heuristic approximations to Bayesian sequential probability assignment for the model classes and performed experiments on a movie rating data set. The proposed algorithms are fast, perform well and the results of the experiments agree with the insights derived from the theoretical considerations.