How to Choose a Recommender System: Insights and Experiences for Large-Scale User Personalization
Rohit Parimi, Toma Trepka, Doina Caragea, Cody Bennett · 2015
Given the large number of items (web-pages, videos) available on the Web, users benefit from being shown only items of potential interest to them. Numerous collaborative filtering (CF) approaches have been proposed in the literature to address this information overload problem. However, with increasingly large datasets, it is often not possible to experiment with every approach and choose the one that best fits an application domain. In this work, we study two CF algorithms, Adsorption and Matrix Factorization, considered to be state-of-the-art approaches, and summarize our experiences and lessons learned from experiments on three implicit feedback data domains. Specifically, we suggest that the characteristics of a domain (e.g., Close connections versus loose connections among users) or characteristics of the data available (e.g., Density of the feedback matrix) can be useful in selecting the most suitable CF approach to use for a particular recommendation problem. Furthermore, we suggest that similar information can also be useful in selecting the best approach to constructing user neighborhoods for Adsorption. Finally, for domains with time information, we show that short user histories can be more effective than long user histories.