A Categorical Model for Discovering Latent Structure in Social Annotations


  • Said Kashoob Texas A&M University
  • James Caverlee Texas A&M University
  • Ying Ding Indiana University




Social, Annotation, Latent, Tags, Topic , Models


The advent of social tagging systems has enabled a new community-based view of the Web in which objects like images, videos, and Web pages are annotated by thousands of users. Understanding the emergent semantics inherent in the socially-generated collection of annotations has important research implications for information discovery and knowledge sharing. To this end, we propose a novel probabilistic generative model for discovering latent structure in large-scale social annotations. The generative model identifies latent community-based ``categories'' of interest that can be used to group semantically-related tags and to augment traditional content-based information search and discovery. We illustrate the proposed approach over large collections of Web objects annotated by the Flickr and Delicious communities. Additionally, we show how to integrate the annotation-based categorical model with traditional content-based approaches for the effective focused discovery and exploration of Web objects.




How to Cite

Kashoob, S., Caverlee, J., & Ding, Y. (2009). A Categorical Model for Discovering Latent Structure in Social Annotations. Proceedings of the International AAAI Conference on Web and Social Media, 3(1), 82-89. https://doi.org/10.1609/icwsm.v3i1.13943