The difficulty of automatic extraction of named entities from mid-19th century American newspapers discussed in the first paper from Project Perseus.
Futures include how to interate user corrections etc. (wikipedia-like additions as an adjunct/correction to autogeneration)

?s How to deal with new names particularly in modern and expanding universes.
Machine learning systems vs rule writing. The Perseus folks are using rule writing, but machine learning is getting better and PP will prolly use one of the Open Source versions.
Have news correspondents’ use of pseudonyms caused problems and/or have they been identified? No.
Information network overlays using RDF and Fedora could be very interesting.

Learning to Deduplicate. from two Brazilian universities. Use of fitness functions discussed as a way to deduplicate.

Why people with common names need ClaimId! Lots of looking at the ACM Digital Library to notice multiple entities for a single author and also listings of common names (especially but from from exclusively Asian names — Jeff Ullman is a recurring example). ClaimId isn’t mentioned but it would help greatly. All of this work is highly mathematical and doesn’t involve humans — hmmmm.

“Also by the Same Author” UK-based project to move to the semantic web called Advanced Knowledge Technology. how to interate Citeseer metadata with ontology. Disambiguation amongst 3000 plus citations. Observation people cite their own work so if they cite an author with a similar name then they asre likely the same person (98% of the time). So you can use that to get at a bunch of varieties of names that way. So self-citation is greatly helpful in increasing precision. Why would it ever fail, the 2 – 8%? Many many co-authored papers are the biggest problem especially if the co-authors are very diverse and the work is “not so coherent as to subject matter.”

Leave a Reply