Video reference: Question answering on YouTube

Abstract

Community-based question answering systems have become very popular for providing answers to a wide variety of "how-to" questions. However most such systems present only textual answers. In many cases, users would prefer visual answers such as videos which are more direct and intuitive. Currently, there is very little research on automatically presenting precise reference videos based on user's question. In this paper, we explore how to leverage YouTube video collections as a source of reference to fulfilll such task and develop a novel multimedia application named:Video Reference. There are two steps to generating a video reference. The first is recall-driven video search, which is to increase the coverage of question by finding other similar questions. The second is precision-based video ranking. A three level ranking scheme based on visual analysis, opinion analysis and video redundancy is adopted to find the most relevant video reference from YouTube. Experiments conducted using questions from Consumer Electronics domain of Yahoo! Answers archive show the feasibility and effectiveness of our approach. Copyright 2009 ACM.