With the ever increasing popularity of photo sharing sites -- such as Flickr -- the volume of photos online growing rapidly. This huge volume of photos calls for new tools for searching and browsing photos. TagExplorer is an application for browsing Flickr photos using the collective knowledge in the Flickr community.

The TagExplorer application allows for browsing of Flickr photos using the textual annotations of the photos – i.e., the tags. The user can either start browsing using a keyword query or by choosing one of the popular tags used by Flickr users. Given the user’s query the TagExplorer generates a set of tags related to the query. The tag-set is displayed to the user using the well-know tag-cloud paradigm. Additionally, the tag-cloud is broken up into several parts by grouping together tags that belong to the same syntactic category.

Here is an example. Suppose the user decides to browse the tag London. The user is of course shown the top photos relevant to her query. Additionally, the TagExplorer shows a cloud of related tags, such as England, United Kingdom, Southwark, City, Big Ben, London Eye, Thames, party and travel. The tags England, United Kingdom, Southwark, and City are grouped together as locations; the tags Big Ben, London Eye, and Thames are grouped together as subjects; and the tags party and travel are grouped together as activities.

Given the tag-cloud of related tags the user can continue her browsing by choosing to add to her query a term from the tag cloud or start a new query based on one of the tags in the cloud. Given the new query a new tag-cloud is generated and the photo list is updated. For example, suppose the user chooses to add the tag Thames to her London query. The tag will now show different tag-cloud relevant to the new query, including the tags Tower Bridge, bridge, river, South Bank, etc. Now the user can again choose to add tags to her query or browse other tags. This process can then – in principle
-- continue forever.

Both the generation of dynamic tag-clouds and the tag classification is based on capturing the collective knowledge in a large pool of user generated content. The dynamic tag-clouds are generated using information about how tags co-occur in photo annotations across the whole set of public Flickr photos. The basic assumption is that two tags are likely to be related to each other if they appear frequently together in the annotation of photos and are used together by many different users (see [1] for details). The tag classification uses structural patterns of Wikipedia – the user generated encyclopedia
-- to classify the Flickr tags into syntactic categories (see [2] for details).

That is enough about the technology behind TagExplorer … it is time to explore.

Börkur
Sigurbjörnsson

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

[2] Overell, Sigurbjörnsson and van Zwol, Classifying Tags using Open Content Resources, In Proceedings of the Second ACM International Conference on Web Search and Data Mining, Barcelona, Spain, 2009.