While at Yahoo! Research I was involved in various projects on analyzing, extending and utilizing tagging of Flickr photos. We analyzed what type of tags users applied to their photos and devised a method for classifying those tags into semantic categories. We developed two prototypes, one for tag recommendation and and one for tag based browsing of photos.

We started our tag analysis by mapping tags to WordNet broad categories and found out that while users annotations are sparse, they are semantically varied [1]. We used the sparseness to motivate the need for a semi automatic tagging system that helped users annotate their photos, using a global aggregation of tag co-occurrence statistics from a large set of user annotations. We later extended the tag recommendation experiments with using information from the users’ social network [2]. Finally we built a prototype system called Fannr that aided users in annotating their photos by recommending tags from various level of the social network as well as using geo information [5].

When mapping Flickr tags to WordNet we found out that a large majority of the tags was not covered and we were not able to map them to a broad semantic category. This lead us to building a tag classification system that used information from Wikipedia to significantly improve the coverage of the tag categorization [3].

Finally, using the extended tag categorization system together with the tag co-occurrence analysis we built a prototype for browsing photos. TagExplorer provided intelligent photo browsing interface through semantically organized tag clouds [4,6].

Related publications

[1] Börkur Sigurbjörnsson, and Roelof van Zwol. Flickr Tag Recommendation based on Collective Knowledge. Proceedings of the 17th International World Wide Web Conference (WWW’08). 2008.

[2] Adam Rae, Roelof van Zwol, and Börkur Sigurbjörnsson. Improving Tag Recommendation Using Social Networks. RIAO’2010 : Recherche d’Information Assistée par Ordinateur / Adaptativity, Personalization and Fusion of Heterogeneous Information. 2010.

[3] Simon Overell, Börkur Sigurbjörnsson, and Roelof van Zwol. Classifying Tags using Open Content Resources. Proceedings of the Second ACM International Conference on Web Search and Data Mining (WSDM 2009). 2009.

[4] Börkur Sigurbjörnsson and Roelof van Zwol. TagExplorer: Faceted Browsing of Flickr Photos. Yahoo! Labs Technical Report (YL-2010-005). 2010.

Related prototypes

[5] TagExplorer — Faceted browsing of Flickr photos

[6] Fannr — Flickr Annotator