Abstract:
|
Recently there has been a growing demand to develop efficient recommender systems, which track users' preferences and recommend items of interest. We propose a group-specific method which utilizes dependency information from similar users and items under the matrix factorization framework. The new approach is effective for the "cold-start" problem, where majority responses in the testing set are obtained from new users whose preferences are not available. Another novelty is that we incorporate non-random missingness information through clustering, based on the numbers of ratings from each user and variables associated with missing patterns. Computationally, we propose a scalable algorithm that embeds backfitting into alternating least squares. This avoids large matrices operation and big memory storage. Our simulation studies and MovieLens data analysis both indicate that the proposed method improves prediction accuracy significantly compared to existing competitive approaches.
|