An enhanced distance based similarity measure for user based recommendations

Abstract

© Springer International Publishing AG 2017. Internet users are overwhelmed with a large number of choices, consequently, there is a need to filter and prioritize relevant information. Recommender System (RS) solves this problem by searching through information provided by users similar to the active user. Precise determination of similar users is the keystone to accuracy of personalized recommendation and in this regard, the contribution of this paper is two-fold. First, an enhanced Distance based similarity measure is introduced. Second, a systematic evaluation is presented of the predictive performance of the proposed similarity measure against different similarity measures in recommendations based on user based Collaborative Filtering (CF). The evaluation encompasses both numeric and non-numeric measures against the proposed measure. The performance metrics are the recommendation accuracy (statistical and decision-making) and coverage. Experimental results on three real-world datasets show that the enhanced Distance based similarity outperforms all other similarity measures for user based recommendations in respect of the recommendation accuracy and coverage.