"Supertagger" behavior in building folksonomies

Abstract

A folksonomy is ostensibly an information structure built up by the "wisdom of the crowds", but is the "crowd" really doing the work? Tagging is in fact a sharply skewed process in which a small minority of users generate an overwhelming majority of the annotations. Using data from the social music site Last.fm as a case study, this paper explores the implications of this tagging imbalance. Partitioning the folksonomy into two halves -one created by the prolific minority and the other by the non-prolific majority of taggers -we examine the large-scale differences in these two sub-folksonomies and the users generating them, and then explore several possible accounts of what might be driving these differences. We find that prolific taggers preferentially annotate content in the long-tail of less popular items, use tags with higher information content, and show greater tagging expertise. These results indicate that "supertaggers" not only tag more than their counterparts, but in quantifiably different ways.

The Wisdom of the Few? “Supertaggers” in Collaborative Tagging Systems

“…Figure 10 presents average user SPEAR expertise scores as a function of user annotation count. Consistent with our previous work (Lorince et al, 2014), we find an overall positive relationship in both Last.fm and Delicious such that average user expertise increases with number of annotations, although the data is noisy for high annotation counts. This suggests that supertaggers are not only more prolific, but also more expert, in their tagging behavior, at least as defined by SPEAR.…”

“…Many efforts have been made to infer taxonomies from folksonomies, synthesizing the advantages of controlled vocabulary and crowdsourced curation (Kubek et al, 2010;Niepert et al, 2007). Due to their low economic costs of implementation and curation, folksonomies have been implemented in diverse domains, including Flickr (photos, Nov et al, 2008;Nov and Ye, 2010), Delicious (web bookmarks, Golder and Huberman, 2006), Last.fm (music, Lorince and Todd, 2013;Lorince et al, 2014), and Bibsonomy (academic papers, Hotho et al, 2006). Reviews of many early social tagging systems can be found in Marlow et al (2006) and Sen et al (2006).…”

“…The Last.fm dataset, on the other hand, was collected by the current authors. It is an expanded version of that first presented in Lorince and Todd (2013) and later analyzed in Lorince et al (2014).…”

“…Expanding on an earlier version of this paper (Lorince et al, 2014) 1 that analyzed tagging in Last.fm, here we address these questions across two additional large-scale tagging datasets from Delicious and Flickr. After presenting related work (Section 2) and an overview of the datasets (Section 3), we formalize our definition of supertaggers and illustrate their disproportionate tagging contribution in Section 4 (RQ1).…”