Question & Answer (Q&A) behaviors on social media have huge potential as a rich source of information and knowledge online. However, little is known about how much diversity there exists in the topics covered in such Q&As and whether unstructured social media data can be made searchable. This paper seeks the feasibility of utilizing social media data for developing a Q&A service by examining the topic coverage in Twitter conversations. We propose a new framework to automatically extract informative Q&A content using machine learning techniques.