Constructing a Question Corpus for Textual Semantic Relations
Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC 2012)
Abstract
Finding useful questions is a challenging task in Community Question Answering (CQA). There are two key issues need to be resolved: 1) what is a useful question to the given reference question; and furthermore 2) what kind of relations exist between a given pair of questions. In order to answer these two questions, in this paper, we propose a fine-grained inventory of textual semantic relations between questions and annotate a corpus constructed from the WikiAnswers website. We also extract large archives of question pairs with user-generated links and use them as labeled data for separating useful questions from neutral ones, achieving 72.2% of accuracy. We find such online CQA repositories valuable resources for related research.