Scalable and practical One-Pass clustering algorithm for recommender system

Recommender systems apply artificial intelligence techniques for filtering unseen information and predict whether a user would like/dislike a given item. K-Means clustering-based recommendation algorithms have been proposed claiming to increase the scalability of recommender systems. One potential drawback of these algorithms is that they perform training offline and hence cannot accommodate the incremental updates with the arrival of new data, making them unsuitable for the dynamic environments. From this line of research, a new clustering algorithm called One-Pass is proposed, which is a simple realtime algorithm that maintains a good level of accuracy, scale well with data, and build the training model incrementally with the arrival of new data. We run One-Pass algorithm on four different datasets (MovieLens, Film Trust, Book Crossing, and Last-FM) and empirically show that the proposed algorithm outperforms K-Means in terms of recommendation and training time. Moreover, One-Pass algorithm is comparable to K-Means in term of accuracy and cluster quality.