Abstract
Many recommender systems collect online users' activity and infer from it users' preferences. They record user actions of various types (e.g. clicks, views), and predict unknown, possibly future, interactions between users and items, mostly using Collaborative Filtering (CF) or Sequence Mining (SM) techniques. While both techniques have their advantages, in this paper, we show that improved prediction accuracy can be achieved by hybridizing them. The proposed hybrid model uses first an SM model to augment an existing actions' data set and then uses collaborative filtering in the final prediction step. The empirical evaluation, which was conducted on a large real-world dataset, showed that the proposed hybrid model outperforms both stand-alone SM and CF.