Retrieving Relevant Conversations for Q&A on Twitter.

Community Question and Answering (Qu0026A) sites provide special features for asking questions and receiving answers from users on the Web. Nevertheless, Web users do not restrict themselves to posting their questions exclusively in these platforms. With the massification of on-line social networks (OSN) such as Twitter, users are increasingly sharing their information needs on these websites. Their motivation for doing so is to obtain a timely and reliable answer from their personal community of trusted contacts. Therefore, daily on Twitter, there are hundreds of thousands of questions being shared among users from all over the world. Many of these questions go unanswered, but also an important number receive relevant and complete replies from the network. The problem is that due to the volatile nature of the streaming data in OSN and the high arrival rates of messages, valuable knowledge shared in this Qu0026A interaction lives very shortly in time. This produces high redundancy and similarity in questions which occurs consistently over time. Following this motivation we study Qu0026A conversations on Twitter, with the goal of finding the most relevant conversations posted in the past that answer new information needs posted by users. To achieve this we create a collection of Qu0026A conversation threads and analyze their relevance for a query, based on their contents and relevance feedback from users. In this article, we present our work in progress which includes a methodology for retrieving and ranking Qu0026A conversation threads for a given query. Our preliminary findings show that we are able to use historical conversation on Twitter to answer new queries posted by users. We observe that in general the asker’s feedback is a good indicator of thread relevance. Also, not all of the feedback features provided by Twitter are equally useful for ranking Qu0026A thread relevance. Our current work focuses on determining empirically the best ranking strategy for the recommendation of relevant threads for a new user question. In the future we seek to create an automatic Qu0026A knowledge base that is updated in real-time that allows for preserving and searching human understanding.