Question answering on large news video archive

Abstract

Many users are interested in looking for precise answers when querying the news video archives, while current systems are designed to return only video sequences. This research explores the use of question-answering (QA) to support personalized news video retrieval. Based on the user's short, imprecise natural language question, with implicit constraints on contents, context, duration and genre of expected videos, the system returns short precise news video fragments as answers. The system uses multi-modal features including visual, audio, speech-to-text, and external on-line news articles to perform the analysis. The system has been tested using 7 days of news video and has been found to be effective.