... DepartmentUniversity of Californiaat BerkeleyBerkeley, CA, 94720zhiheng@eecs.berkeley.eduAbstractWe present a graph -based semi-supervised learning for the question-answering (QA)task for ranking candidate ... unlabeled data insemi-supervised learning (SSL) environment, withan emphasis on graph -based methods, can im-prove the performance of information extractionfrom data for tasks such as question classifica-tion ... we used one feature for each and up to three for M-Match features. Thefeature values vary based on matching type, i.e.,exact match, containment, synonym match, etc. For example, the S-Match...