Corpora: Information extraction from financial texts

Ken Litkowski (
Mon, 08 Feb 1999 22:16:35 -0500

A correspondent raised the question whether it is possible to extract
"numbers" and other salient information (earnings reports, earnings
surprises, new products etc.) and score the frequency, valence and
strength of these units. He would then want to take these scores and
put them into his genetic algorithm driven neural network as predictors
of a time series of, for example, Dell stock prices. (He has other
variables already being tracked and wants to add these, perhaps using
something like Reuters wire reports.) I responded describing the
paradigms used in the MUC competitions, noting that these paradigms
still fall short of what he wants. I am not familiar with anyone having
taken such next steps.

If anyone is aware of something relevant, you may contact "Lawrence
Weathers, Ph.D" <> directly. I would also be
interested and if there sufficient response, I will post a summary.
Suggestions are also welcome about other lists to post this query.


Ken Litkowski                     TEL.: 301-482-0237
CL Research                       EMAIL:
9208 Gue Road
Damascus, MD 20872-1025 USA       Home Page: