Abstract
If we develop a Website which automatically collects questions and asks for answers, people can respond efficiently. Against this background, we developed a method to semi-automatically collect questions, constructed a Website in which they were shown, and evaluated its effectiveness. Our method consists of two steps, i.e., (1) by submitting words which are characteristic to questions into search engine, we obtain Blog articles and tweets which are likely to contain questions, and (2) based on automatic text classification, we extract Blog articles and tweets with questions. It was found that (1) the characteristic words "taitoru-ga-omoidase-nai" produced the best precision and (2) Boosting and decision tree produced the best classification for Blogs and Twitter. When we showed 30 articles and 31 tweets in our Website, six and five were answered, respectively, which satisfied the authors ofquestions.
