two new recommenders, a new similarity measure, a couple of new evaluation metrics

[paraschakis]

• Recommenders:
• Bigram Rules recommender. Simple association rules of type item1 -> item2, where rule's confidence and support are used for scoring candidate items (currently, the linear combination of confidence and support is used, but other options are also possible).
• RSNMF recommender for rating prediction. See Luo et al. (2014): "An efficient non-negative matrix-factorization-based approach to collaborative filtering for recommender systems". Currently, only the "train" and "predict" methods are implemented.
• "LogLikelihood" similarity measure, based on Mahout implementation.
• Evaluation metrics: F1@5, F1@10, R-precision (precision at the level of recall).