I would like to ask about possibilities of implementing Sparse Linear Methods (SLIM) recommender in Mahout during GSOC 2014.
The SLIM algorithm generates efficient recommendations and its performance is shown in the original paper (http://glaros.dtc.umn.edu/gkhome/fetch/papers/SLIM2011icdm.pdf). The study demonstrates that SLIM outperforms traditional algorithms (such as itemkNN, userkNN, SVD or Matrix Factorization approaches) on various data-sets in terms of run-time and recommendation quality. The algorithm can be paralellized and Map-Reduce can help us achieve that.
I am aware of real world systems that are using SLIM as a recommendation engine (e.g. Mendeley: http://www.slideshare.net/MarkLevy/efficient-slides) and I think it represents the state-of-the-art in collaborative filtering right now.
Would this be an interesting addition to Mahout and is somebody interested in mentoring this at Google Summer of Code 2014?