An analysis of probabilistic methods for top-N recommendation in collaborative filtering

In this work we perform an analysis of probabilistic approaches to recommendation upon a different validation perspective, which focuses on accuracy metrics such as recall and precision of the recommendation list. Traditionally, state-of-art approches to recommendations consider the recommendation process from a "missing value prediction" perspective. This approach simplifies the model validation phase that is based on the minimization of standard error metrics such as RMSE. However, recent studies have pointed several limitations of this approach, showing that a lower RMSE does not necessarily imply improvements in terms of specific recommendations. We demonstrate that the underlying probabilistic framework offers several advantages over traditional methods, in terms of flexibility in the generation of the recommendation list and consequently in the accuracy of recommendation.