Entropy-weighted similarity measures for collaborative recommender systems
Soojung Lee · AIP conference proceedings · 2018
Collaborative filtering-based recommender systems have been widely used in many commercial systems, giving a great aid in selecting products to users. These systems, however, still have much to be improved in terms of prediction or recommendation accuracy. This paper intends to improve the previous similarity measures used to find neighbor users which combine heuristic weights with the traditional similarity measures. The proposed method exploits information entropy and incorporates it into the traditional similarity measures, so that the global rating behavior on items by all users can be reflected. The efficiency of the proposed scheme is examined through various experiments in comparison to the previous measures.