A Multi-Armed Bandit Recommender System Based on Context and KNN

In this paper a multi-armed bandit Recommender System (RS) is proposed based on contextual content and knn (k-nearest neighbors) algorithm. The multi-armed bandit problem is considered as a recommending problem and it transfers the cold start problem into EE (explore exploit) problem. By using the users observed features as contextual content, the RS can compute the difference between users and get the k nearest neighbors of a giving user. The neighbors rate all the candidate items based on the contextual content. With the analysis of the neighbors ratings, the system can make some good recommendations to the user. The experiments on real datasets from movielens and jester are conducted. It shows that the contextual multi-armed bandit Recommender System based on knn performs better compared with other baseline approaches. And the proposed algorithm is particularly effective in solving the cold start problem.