PLIERS: PopuLarity-based ItEm Recommender System

In the last few years social networking platforms proliferated, providing the users with the possibility to generate items and define tags. Social tagging systems help users and the system identify useful content with respect to the users interests. The relations between users, items, and tags created by these tagging systems form a special kind of taxonomy, called folksonomy, that can be naturally represented as a graph. To further help the users to find out interesting contents, tag-based recommender systems have been successfully defined as reasoning algorithms on these types of graphs. However, the two most popular solutions proposed in literature suffer from strong limitations. They mainly do not take into account the characteristics of the user’s interests (e.g., their distribution on the network or their semantic meaning), but they tend to recommend either content with minimum or maximum popularity among those available in the network. To overcome this limitation, a hybrid algorithm that linearly combines the two solutions has been recently proposed, but it also inherits, in part, their shortcomings. In fact, the recommendations it produces are not always compatible, in terms of popularity, with the items already owned by the users. In addition, it requires the definition of a specific parameter that identifies the relevance of these algorithms on each specific evaluation. In this paper, we propose a novel tag-based recommender system, called PLIERS, that solves the dilemma between the choice of popular or unpopular items, without requiring any parameters to tune, and ensuring that the popularity of recommended items is always compatible with the popularity of items already adopted by the users. Simulation analyses on four large-scale datasets obtained from real social networks demonstrates that our approach outperforms the state of the art.