Description: A novel method employing machine-based learning to identify messages related to other messages is described and evaluated. This technique may enable an analyst to identify and correlate a small number of related messages from a large sample of individual messages. The classic machine learning techniques of decision trees and naive Bayes classification are seeded with few (or no) messages of interest and 'learn' to identify other related messages. The performance of this approach and these specific learning techniques are evaluated and generalized.
Date: October 1, 2004
Creator: Nove, Charles E.; Maclin, Richard Frank; Theuninck, Andrew K. (University of Minnesota-Duluth, Duluth, MN); Newland, Jeremy L. (University of Minnesota-Duluth, Duluth, MN); Torrey, Lisa A. (University of Wisconsin-Madison, Madison, WI) & Robinson, Eric R. (University of Wisconsin-Madison, Madison, WI)
Item Type: Refine your search to only Report
Partner: UNT Libraries Government Documents Department