Brady Forrest

Consider Auto-Tagging

Biography
I was born and raised on the East Coast. I attended RPI, getting a degree in Industrial Engineering. I spent several years doing Supply Chain consulting around the North America & Europe. In 2000, a friend of mine in the Bay Area invited me to join MongoMusic.com as a PM. Six short months and a BurningMan later I found myself moving to Seattle as a result of an Microsoft acquisition. While here I’ve been a part of the first versions of MSN Music, Newsbot, and MSN Search. When not working I build cars (www.gravitybowl.com), keep my friends close, and work on my house (www.roochalet.com).

I’m fascinated by how software can keep us in touch with each other across the globe and learn more about how our society is feeling about a given topic.

Position Paper
Tagging is a relatively new topic. People use this new technique to classify information for personal organization and from that we attain an aggregate view of what the net is reading, writing, hoping to do, or photographing.

I see these aggregate results on zeitgeist pages. Typically, I get a quick view of the most popular tags; their size tells me their relative importance. Color could be used to show rising or falling in popularity. This is an amazing view of what the collective consciousness of the users of these services think is important.

However, tagging is not without its drawbacks. One issue is timeliness. Tagging does not react quickly to issues; enough people need to see and tag it for it to make it onto an aggregate page. Another issue is completeness; all of the web is not tagged (many would argue that untagged pages are probably of lower quality) and the majority of the pages tagged are in English. I don’t think that humans will ever tag the web by themselves and the resulting scarcity hinders their usefulness.

To tackle these issues, auto-tagging or keywording could be used. A crawler could be used to tag web pages; a news aggregator already produces this data in the process of clustering. Would we lose the human element entirely? The articles themselves are written by humans, and humans choose the vocabulary of the article, thus influencing the tags. If the tags were displayed on discovery page, one would be able to quickly see which tags were the most popular, which were rising to the top and would gain insight into what was new on the web. With a more complete set of tags they could be used for navigation in search.

Back to Social Computing Symposium 2005