Comparative analysis of neighborhood-based approache and matrix factorization in recommender systems
Оleg Chertov, Armelle Brun, Anne Boyer, Marharyta Aleksandrova · Eastern-European Journal of Enterprise Technologies · 2015
Unlike other works, this paper aims at searching a connection between two most popular approaches in recommender systems domain: Neighborhood-based (NB) and Matrix Factorization-based (MF). Provided analysis helps better understand advantages and disadvantages of each approach as well as their compatibility.While NB relies on the ratings of similar users to estimate the rating of a user on an item, MF relies on the identification of latent features that represent the underlying relation between users and items. However, as it was shown in this paper, if latent features of Non-negative Matrix Factorization are interpreted as users, the processes of rating estimation by two methods become similar. In addition, it was shown through experiments that in this case elements of NB and MF are highly correlated. Still there is a major difference between Matrix Factorization-based and Neighborhood-based approaches: the first one exploits the same set of base elements to estimate unknown ratings (the set of latent features), while the second forms different sets of base elements (in this case neighbors) for each user-item pair.