Flickr Does Search Well With Clustering
Recently, two posts have given me an increased appreciation for what Flickr has accomplished with their clustering feature.
In his post Random Acts of Sensless Tagging, DonXML talks about the weakness of simple tagging schemes to understand the semantics of a tag. Today, Jeff Atwood follows up this thought by illustrating how a Google search for a single word also returns results that ignore other possible meanings of the word. He presents eBay as a better example with its “quasi-hierarchical” category results in the side bar.
Jeff even suggests a better approach using Markov chain probabilities to automatically suggest alternat semantics. For example, you search on “Jaguar” and in the search results you get a suggestion, “Did you mean: Jaguar cat, Jaquar Automobile, OSX Jaguar...”
Well this is exactly what Flickr does when searching for tags. Try searching for the tag “rock”. Flickr returns a set of clusters around the term. In the top cluster (at the time of this writing), there are pictures of bands music bands and guitarists. The other clusters involve stones, ocean, beaches as one might expect. The last one interestingly enough associates “rock” with “hard” and “cafe”.