Time and Local Popularity in top-N Recommendation
share
Items popularity is a strong signal in recommendation algorithms. It affects collaborative filtering algorithms and it has been proven to be a very good baseline in terms of results accuracy. Even though we miss an actual personalization, global popularity of items in a catalogue can be used effectively to recommend items to users. In this paper we introduce the idea of a form of personalized popularity also considering how its changes over time affect the accuracy of recommendation results. Although the proposed approach results quite light in terms of computational effort, its accuracy results highly competitive compared to state of the art model-based collaborative approaches.
READ FULL TEXT