Research On Chinese Vedio Question Answering

Question Answering is to locate, extract, and represent a specific answer to a user question expressed in natural language, and current question answering systems succeed in many aspects regarding to questions of textual documents. With the development of the internet, In addition to traditional text message, multimedia data has become increasingly important data on the web, which provides both opportunities and challenges for question answering. video is one of the most effective information for capturing the events in the real world. Our framework is based on news video. In all features, transcript is the most important and most readily available video features. Moreover, the input of video question answering(VideoQA) is a short question, so we main employ transcript feature that is gained by ASR.This paper proposes a framework for Chinese Video question answering system. The whole system consists of six modules: video segmentation, speech recognition, question classification, transcript retrieval, answer extraction and video output. But the news transcripts contain numerous speech recognition errors, so we manually correct some errors. In the module of question classification, we employ HowNet to improve the accuracy. VideoQA is to obtain the close video clips, and not just a long story unit, so we need to process and position the close sentences to answer the question. We claim that the best sentence that answers the question should satisfied some conditions which are based on query density, answer type, etc.The main contributions of this paper are: (1) HowNet is employed in QA system; (2)the extension of QA technology to support QA in Chinese news video. Experiments on Chinese news CCTV4 show that our framework is effective.