I use only Mahout and therefore do not know how PredictionIO translates your data to the input and the choice of algorithm.
For Mahout, you will use a detail-based recommendation using logical input and type “listen” as an action with a value of 1 and SIMILARITY_LOGLIKELIHOOD affinity metric. LLR still ignores weights. Scales are used in old style guidelines that try to predict user ratings. Most people these days are more interested in ranking, and the above configuration will give you the best results.
source
share