2004), we demonstrated an increase in precision
from .43 to .95, with only a 30% drop in recall.
o Through a structured repository, such as a
Although the reciprocal questions seem to be
knowledge-base of real-world information
o Through statistical techniques tied to a machine-
symmetrical and thus redundant, their power stems
from the differences in the search for answers inher-
learning algorithm, and a text corpus.
ent in our system. The search is primarily based on
We think that all three methods are appropriate,
the expected answer type (STATE vs. CAPITAL in
but we initially concentrate on the first for practical
the above example). This results in different docu-
reasons. Most TREC-style factoid questions are
ment sets being passed to the answer selection mod-
about people, places, organizations, and things, and
ule. Subsequently, the answer selection module
we can generate generic auxiliary question sets for
each of these classes. Moreover, the purpose of this
works with a different set of syntactic and semantic
relationships, and the process of asking a reciprocal
paper is to explain the QDC methodology and to
question ends up looking more like the process of
investigate its value.
asking an independent one. The only difference be-
Bạn đang xem 2004) - BÁO CÁO KHOA HỌC QUESTION ANSWERING USING CONSTRAINT SATISFACTION QA BY DOSSIER WITH CONSTRAINTS DOCX
