7. Rank the top M answers according to the prob-
uation is costly and time-consuming, so we use au-
ability of the first word.
tomatic evaluation results, i.e., exact matching re-
sults and partial matching results, as a pseudo lower-
This approach is designed to extract only the most
highly probable answers. However, pin-pointing
bound and upper-bound of the performances. Inter-
estingly, the manual evaluation results of MRR and
only answers is not an easy task. To select the top
Top5 are nearly equal to the average between exact
five answers, it is necessary to loosen the condition
and partial evaluation.
for extracting answers. Therefore, in the execution
phase, we only give label O to a word if its probabil-
To confirm that the QBTE ranks potential answers
ity exceeds 99%, otherwise we give the second most
to the higher rank, we changed the number of para-
graphs retrieved from a large corpus from N =
probable label.
As a further relaxation, word sequences that in-
Bạn đang xem 7. - BÁO CÁO KHOA HỌC QUESTION ANSWERING AS QUESTION BIASED TERM EXTRACTION A NEW APPROACH TOWARD MULTILINGUAL QA DOC