Tuesday, March 20, 2012

Collaborative Filtering via Group-Structured Dictionary Learning (implementation)

Zoltan Szabo just sent me the following:
Dear Igor,
We have recently given a new application of group-structured dictionaries for collaborative filtering based recommender systems. May I ask you to post the information to the blog? -- I believe the paper could be of interest to the community. Thank you.
Abstract: Structured sparse coding and the related structured dictionary learning problems are novel research areas in machine learning. In this paper we present a new application of structured dictionary learning for collaborative filtering based recommender systems. Our extensive numerical experiments demonstrate that the presented method outperforms its state-of-the-art competitors and has several advantages over approaches that do not put structured constraints on the dictionary elements.

The extended Technical Report is here. The approach in the paper is using the OSDL (online group-structured dictionary learning) technique. The code available under GPLv3 license on Zoltan's website, direct links to the toolbox:(.rar, .tar, .zip). It is listed in the Matrix Factorization Jungle page.

No comments: