Flexible recommendation using random walks on implicit feedback graph

Realizing context-aware recommender systems (CARS) has been acknowledged as one of the most important topics in the area of recommender systems. CARSs aim to incorporate contextual information in recommendation to achieve better recommendation results. Although existing CARSs deal with multiple dimensions, most of them focus on the problem of predicting users' ratings for one specific recommendation target dimension (e.g. movie) considering fixed contextual dimensions. However, in many real world cases, contextual dimensions and recommendation target dimension can often vary depending on different purposes or needs. In this paper, we show how to transform log data which implies users' preference and contextual information into implicit feedback graph and propose a recommendation method using random walks on the transformed graph. Our method has advantages in that it supports flexibility of varying contextual dimensions and recommendation target dimension without time-consuming learning process. We perform several experiments, and the experimental results show that our method can be actually used for real world cases.