SCARS: A scalable context-aware recommendation system

Abstract

Recommender Systems (RS) are used to provide personalized suggestions for information, products and services that are not already used or experienced by a user, but are very likely to be preferred by him/her. Most of the existing RS employ variations of Collaborative Filtering (CF) for suggesting items relevant to users' interests. However, CF requires similarity computations that grows polynomially with the number of users and items in the database. In order to handle this scalability problem and speeding up the recommendation process, we propose a clustering based recommendation method. The proposed work utilizes the different user attributes such as age, gender, occupation, etc. as contextual features and then partitions the users' space on the basis of these attributes. We divide the entire users' space into smaller clusters based on the context, and then apply the recommendation algorithm separately to the clusters. This helps us to reduce the running time of the algorithm as we avoid computations over the entire data. In this work, we present a scalable CF framework that extends the traditional CF algorithms by incorporating users context into the recommendation process. While recommending to a target user in a specific cluster, our approach uses the ratings of the target user as well as the rating history of the other users in that cluster. One of the main objectives of our work is to reduce the running time without compromising the recommendation quality much. This ensures scalability, allowing us to tackle bigger datasets using the same resources. We have tested our algorithm on the MovieLens dataset, however, our recommendation approach is perfectly generalized. Experiments conducted indicate that our method is quite effective in reducing the running time.