Automatically Generating Extraction Patterns from Untagged Text

Riloff 1996

Riloff describes AutoSlog-TS, an unsupervised algorithm for extracting patterns from tagged texts. These patterns are meant for use in finding & categorizing noun phrases. Example patterns are "x was murdered" and x bombed." AutoSlog-TS first finds patterns for every NP in the text, using pattern templates (such as "subj passive-verb"). It then evaluates the patterns, by testing how well they do at classifying the texts. In particular, AutoSlog-TS selects patterns that maximize precision and coverage. It continues selecting patterns until the precision/coverage become too low. After it has selected and ordered the patterns, they can be manually filtered.


    author = "Ellen Riloff",
    title = "Automatically Generating Extraction Patterns from Untagged Text",
    booktitle = "{AAAI}/{IAAI}, Vol. 2",
    pages = "1044-1049",
    year = "1996",
    url = "citeseer.nj.nec.com/riloff96automatically.html"