Relevance, novelty, diversity and personalization in tag recommendation

Abstract

The design and evaluation of tag recommendation methods have historically focused on maximizing the relevance of the suggested tags for a given object, such as a movie or a song. However, relevance by itself may not be enough to guarantee recommendation usefulness. In this dissertation, we aim at proposing novel solutions that effectively address multiple aspects related to the tag recommendation problem, notably, relevance, novelty, diversity, and personalization. Towards that goal, we (1) propose and combine various tag quality attributes by means of heuristics and learning-to-rank (L2R) techniques, and (2) extend our best methods to address personalization, novelty (tag's specificity), and diversity (topic coverage), considering different scenarios of interest. Our evaluation, performed with data from five Web 2.0 applications, demonstrates the effectiveness of our new methods, and attest the viability to increase novelty and diversity with only a slight impact on relevance.