Flickr search stemming

acme on 2006-05-30T16:14:55

Flickr has recently undergone a redesign. The search page was previouusly very hard to find (it obviously had performance issues) and one of the highlights of the new design for me is the fact that every page now has a search box. This is excellent. I can now easily search for photos of orange, cute lambs and sichuan food. The search results for those searches are wonderful and on-topic. I was looking for photos of yawning things for a presentation recently and I searched for tired - but all I get are pictures of tires! What's going on is that Flickr is stemming the search terms. Both "tired" and "tires" stem to "tire", so I get round things instead of yawning things. Stemming is often used when searching collections - it generally improves accuracy - but in this case it confused me greatly. See what happens when you search for stemming: handle bars, plant stems, stem cells and a "Stem Restaurant" sign. And no diagrams of how stemming systems work...


<aol/>

Dom2 on 2006-05-30T19:43:44

Several of our web sites at $WORK are just going through the process of being de-stemmed at the moment. On some reference sites, it's a really bad idea. We also had to thin out the stop word list significantly.

-Dom

Stemming is fine

drhyde on 2006-05-31T10:56:55

The problem here is not stemming, but illiterate people mis-spelling 'tyre'.