View On Data

Managing personal collections of data is becoming more and more problematic with the drastic increase in digital information. Collaborative Tagging, as seen e.g. on Flickr, Del.ici.ous, or Last.fm, is a relatively new and user-driven way to organise data by assigning labels to data items. Although there are no rules and no pre-defined categories, some form of convergence to a shared vocabulary of tags can be observed. The single reason for this fascinating behaviour is the exposure of the tags to other users of the web site. With users now providing high-level descriptions of data, the tagging movement questions the traditional view on information retrieval. We believe that the new challenge consists in integrating tagging and data analysis for an intuitive and effective data navigation. Integrated tools will not only help to overcome shortcomings inherent to solely tag-based systems, but will also contribute to keep tagging alive.

Our prototype system for image navigation uses photos and tags from Flickr. Browsing by tags is enhanced by similarity retrieval based on image analysis. This research shows that our approach can overcome problems inherent to using tags only, such as synonymy, homonymy, and spelling mistakes. On the other hand, a great simplification over image search solely based on image features is achieved.