TagClusters: Enhancing Semantic Understanding of Collaborative Tags

Many today’s online communities use TagClouds, an aesthetic and easy to understand visualization, to represent popular tags collaboratively generated by their users. However, due to the free nature of tagging, such collaborative tags have some linguistic problems and certain other intrinsic limitations, such as high semantic density. Moreover, the alphabetical order of TagClouds poorly supports a hierarchical exploration among tags. This paper presents an exploration to support semantic understanding of collaborative tags beyond TagClouds. Based on the results of our survey on people’s practical usages of collaborative tags, we developed a visualization named TagClusters, in which tags are clustered into different groups, with font size representing tag popularity and the spatial distance indicating the semantic similarity between tags. The subgroups in each group and the overlap between groups are highlighted, thus illustrating the underlying hierarchical structure and semantic relations between groups. We conducted a comparative evaluation with TagClouds and TagClusters based on the same tag set. We received overall positive feedback on TagClusters and the results confirmed the advantage of TagClusters in facilitating browsing, comparing and comprehending semantic relations between tags. In future work, besides supporting semantic browsing, we will explore other usages of TagClusters, such as tag suggestions or tag-based Information Retrieval. Keywords: Improvement of TagClouds, collaborative tagging, user-contributed tags, visualization of tags, semantic analysis.