3 Harper, F. Maxwell Moy, D. Konstan, J.A. 2009 ACM SIGCHI Conference on Human Factors in Computing Systems <p>Tens of thousands of questions are asked and answered<br />every day on social question and answer (Q&amp;A) Web sites<br />such as Yahoo Answers. While these sites generate an<br />enormous volume of searchable data, the problem of<br />determining which questions and answers are archival<br />quality has grown. One major component of this problem is<br />the prevalence of conversational questions, identified both<br />by Q&amp;A sites and academic literature as questions that are<br />intended simply to start discussion. For example, a<br />conversational question such as “do you believe in<br />evolution?� might successfully engage users in discussion,<br />but probably will not yield a useful web page for users<br />searching for information about evolution. Using data from<br />three popular Q&amp;A sites, we confirm that humans can<br />reliably distinguish between these conversational questions<br />and other  informational questions, and present evidence<br />that conversational questions typically have much lower<br />potential archival value than informational questions.<br />Further, we explore the use of machine learning techniques<br />to automatically classify questions as conversational or<br />informational, learning in the process about categorical,<br />linguistic, and social differences between different question<br />types. Our algorithms approach human performance,<br />attaining 89.7% classification accuracy in our experiments.</p>