My Photo

Your email address:


Powered by FeedBlitz

April 2018

Sun Mon Tue Wed Thu Fri Sat
1 2 3 4 5 6 7
8 9 10 11 12 13 14
15 16 17 18 19 20 21
22 23 24 25 26 27 28
29 30          
Blog powered by Typepad

Become a Fan

« “Macro Trends: The Privacy and Civil Liberties Consequences … and Comments on Responsible Innovation” – My DHS DPIAC Testimony, September 2008 | Main | Nation At Risk: Policy Makers Need Better Information to Protect the Country »

March 03, 2009

Comments

Feed You can follow this conversation by subscribing to the comment feed for this post.

amy

Thanks for educating me what this word actually means!
I couldn't agree more with your assessment. We like to categorize knowledge domains, but often it's inappropriate. So much information is inter-related and benefits from the wisdom of other areas of knowledge. Categorization is useful as a structure for understanding ideas, but if it creates separate theories for similar situations, it's detrimental instead of helpful.

Daniel Tunkelang

Jeff, I share your frustration with folks who over-complicate the tagging process. Automatic and untrained are good things. But I think it's also good to leverage whatever human-supplied information you can get for free.

You might want to check out work that my colleagues and I at Endeca did with the Association of Computing Machinery (ACM) to tag articles in their digital library. We distilled a vocabulary from the dirty, sparse set of author-supplied tags and then used it as a basis for automatically tagging the collection as a whole. The results were very nice, and we presented them at HCIR '08. We also applied a similar technique to a leading sports programming network.

http://research.microsoft.com/en-us/um/people/ryenw/hcir2008/

David Allsopp

It's precisely because 1) change happens and 2) there is no single version of truth, that I find ontologies useful. This is because the alternative is often to build your domain model into code, which is much harder to inspect and change than an ontology. By pulling the model out into an ontology, then building code that introspects onto that model, change is easier to cope with.

The comments to this entry are closed.