Continuing talking to myself here -- I reverse myself again. After thinking about this a bit more... the "boolean" preference case doesn't need special attention in the context of item-based recommenders or similarity metrics. It doesn't change the implementation.
So Otis I think you could try item-based recommenders in your case, in the way I suggested above. On Tue, Jan 20, 2009 at 10:09 PM, Sean Owen <[email protected]> wrote: > Er yeah Otis reminds me that I really actually need to make a few more > classes to support item-based recommendation in this case. The code > snippet I gave will work but will be better after I get to writing in > a few more changes to actually support item-based recommenders here. > > On Tue, Jan 20, 2009 at 7:37 PM, Otis Gospodnetic > <[email protected]> wrote: >> Hello, >> >> Boolean* classes sped up things for me: >> >> UserSimilarity similarity = new >> BooleanTanimotoCoefficientSimilarity(model); >> hood = new NearestNUserNeighborhood(HOOD_SIZE, MIN_SIMILARITY, >> similarity, model); >> recommender = new BooleanUserGenericUserBasedRecommender(model, hood, >> similarity); >> >> Sean did recommend using Item-based recommender when the number of items is >> relatively low compared to the number of users, but we only have Boolean >> flavour of User-based recommender in svn for now. >
