Adapted Collaborative Filtering Algorithms through Diversity and Novelty

In today’s society, the quantity of available data is exploding. Recommender systems are tools that enable processing those data. With the use of such tools, drawbacks about the quality of the recommended content - e.g. poor diversity and novelty - are envisioned. This paper presents two approaches that modify a classical user-based collaborative filtering process with the aim of improving diversity and/or novelty while maintaining a good level of recall. The first approach uses information about item popularity to alter the process. Only unpopular items are taken into account in he neighborhood determination, leading to more novelty in the recommendation lists. The second approach - a reranking technique- allows deciding the level of similarity within the recommendation lists proposed to users. That criteria switch impacts both diversity and novelty