Video Ranking with Multi-Evidence Combination
Zhu Zhenfeng · Dianzi xuebao · 2010
According to the analysis on users'query behavior,we found that the users are rarely patient to go through all the returned results.Usually,they only check the most top returned results,so it is vital for a search engine to rank high relevant results to the top.In order to move vigorously toward this goal,we proposed a video ranking method based on multi-evidence combination.Its theoretical foundation is Dempster-Shafer evidence combination theory,which can fuse the evidence from distinct views and cooperatively infer the most likely relevant results.If the evidence from multiple aspects reaches a consensus on the conclusion that a certain video shot is relevant to query,the video shot is treated as the most likely relevant shot and is ranked to the top of the returned result list.On the contrary,a lack of consensus leads to the judgment that the video shot is not highly relevant to the query and is ranked to the bottom of the returned result list.Experimental results show that the search quality,especially on the most top-ranked results,is significantly improved after using the proposed scheme.