Generating Questions from Web Community Contents

Large amounts of knowledge exist in the user-generated contents of web communities. Generating questions from such community contents to form the question-answer pairs is an effective way to collect and manage the knowledge in the web. The parser or rule based question generation (QG) methods have been widely studied and applied. Statistical QG aims to provide a strategy to handle the rapidly growing web data by alleviating the manual work. This paper proposes a deep belief network (DBN) based approach to address the statistical QG problem. This problem is considered as a three-step task: question type determination, concept selection and question construction. The DBNs are introduced to generate the essential words for question type determination and concept selection. Finally, a simple rule based method is used to construct questions with the generated words. The experimental results show that our approach is promising for the web community oriented question generation.