Refining frequency-based tag reuse predictions by means of time and semantic context

Abstract

In this paper, we introduce a tag recommendation algorithm that mimics the way humans draw on items in their long-term memory. Based on a theory of human memory, the approach estimates a tag’s probability being applied by a particular user as a function of usage frequency and recency of the tag in the user’s past. This probability is further refined by considering the influence of the current semantic context of the user’s tagging situation. Using three real-world folksonomies gathered from bookmarks in BibSonomy, CiteULike and Flickr, we show how refining frequency-based estimates by considering usage recency and contextual influence outperforms conventional “most popular tags” approaches and another existing and very effective but less theory-driven, time-dependent recommendation mechanism. By combining our approach with a simple resource-specific frequency analysis, our algorithm outperforms other well-established algorithms, such as FolkRank, Pairwise Interaction Tensor Factorization and Collaborative Filtering. We conclude that our approach provides an accurate and computationally efficient model of a user’s temporal tagging behavior. We demonstrate how effective principles of recommender systems can be designed and implemented if human memory processes are taken into account.