INVESTIGATING DIFFERENT SIMILARITY METRICS USED IN VARIOUS RECOMMENDER SYSTEMS TYPES: SCENARIO CASES

Abstract. A recommendation system represents a very efficient way to propose solutions adapted to customers needs. It allows users to discover interesting items from a large amount of data according to their preferences. To do this, it uses a similarity metric, which determines how similar two users or products are. In the case of recommender systems, similarity computation is a practical step. The calculation of similarity may be used for both items and users. Following the similarity calculation, a user or item with a comparable computation value can be recommended together with the goods to a user with similar preferences. The user’s requirements influence the choice of similarity metric. This paper explores various similarity measurement methods employed in recommender systems. We compare correlation and distance techniques to determine the capabilities of different similitude calculation algorithms and synthesize which similarity measure is adapted for which type of recommendation.