NUROP – Searching for Multimedia News on The Web NUROP CONGRESS PAPER SEARCHING FOR MULTIMEDIA NEWS ON THE WEB

Current search engine is too general and often return too much of irrelevant information in respond to user’s free-text queries. Many redundant documents are returned and this often takes users a long time to find the actual piece of news that they want. The aim of this project is to develop an expanded query system to support the precise retrieval of news from multiple sources and in different medium types from the web. This includes text, images and video. In this project, we will first investigate the use of an existing news video segmentation program to carry out automatic indexing of online news sequences in term of story units. The information to be indexed for each news story includes the audio and visual content features and its associated news transcript generated from a speech recognizer. The second part of the project introduces a way of using the web resources to correct the speech recognition errors in news transcripts. As most errors are of type named-entity (NE); I have called this the Name-Entity (NE) error correction system. This system actually “corrects ” the NE that may be mis-recognized by the speech recognizer. The resulting transcripts contain more correct NEs and thus making the video retrieval more precise