A Social and Popularity-based Tag Recommender

Modou Gueye (1, 2) , Talel Abdessalem (1, 2) , Hubert Naacke (3)

1 DIG - Data, Intelligence and Graphs
2 INFRES - Département Informatique et Réseaux
3 BD - Bases de Données

Tag recommendation aims to recommend to a userthe most suited tags for a given item. It is an important functionality of resource sharing systems. In this paper we propose a recommendation algorithm, called FasTag, that links the relevance of the tags to both their popularity and the opinions of the user's neighbors. FasTag assumes that the users are organized in a weighted graph representing, for instance, the similarity or the trust between them.The two salient aspects of FasTag are the scoring function it uses to evaluate the relevance of the tags, and its low computation cost. Thus, FasTag can make online recommendations, even for large datasets.Moreover, we improve the accuracy of its recommendations by adjusting automatically the size of the recommended list of tags. The experiments we did on several datasets show a significant improvement of the accuracy of the recommendations.

Mots clés: Accuracy; Scalability; Social network; Tag recommendation

Domaines: Recherche d'information [cs.IR]; Autre [cs.OH]

Modou Gueye, Talel Abdessalem, Hubert Naacke. A Social and Popularity-based Tag Recommender. SocialCom 2014 - The Seventh IEEE International Conference on Social Computing and Networking, Dec 2014, Sidney, Australia. pp.318-325, ⟨10.1109/BDCloud.2014.44⟩. ⟨hal-01113540⟩