The basic objective of a Collaborative Filtering (CF) algorithm is to suggest items to a particular user based on his/her
preferences and users with similar interests. Although, there is an apparently strong demand for CF techniques, and many algorithms
have been recently proposed, very few articles comparing these techniques can be found. Our paper is oriented towards the
study of a sample of algorithms to representing differents stages in the evolutive process of CF.
Experiments were conducted on two datasets with different characteristics, using two protocols and three evaluation metrics
for the different algorithms. The results indicate that, in general, the Online-Learning (WMA, MWM) and the Support Vector
Machines algorithms have a better performance that the other algorithms, on both datasets. Considering the amount of information,
the less sparse such information is, the higher the coverage and accuracy of general models tend to be; however, the behavior
under sparse data is closer to what is observed in a real system if we have in mind that users usually rate an amount of records
much smaller than the total available.
Keywords Collaborative Filtering - Online learning - Memorybased models - Dependency Networks - Aspect model - Support Vector Machines