Learning to answer questions from semi-structured knowledge sources

Abstract

Question answering is an efficient and convenient way for humans to make use of the massive amount of information on the Web. I start with an interesting source of information -- infoboxes in Wikipedia that summarize factoid knowledge -- and develop a comprehensive approach to answering questions with high precision. I first build a system to access data in infoboxes in a structured manner. I use the system to construct a crowdsourced dataset of over 15,000 high-quality, diverse questions. With these questions, I train a convolutional neural network model that outperforms models that achieve top results in similar answer selection tasks.