TRECVID 2008 NOTEBOOK PAPER: InteractiveSearch Using Multiple Queries and Rough Set Theory.
Akihito Mizui, Kimiaki Shirahama, Kuniaki Uehara · 2008
For the TRECVID2008 submitted runs, cs24 kobe team participated in the interactive search task and we submitted two different runs: • 1 b 1 cs24 kobe2 2: we run the search approach based only on ASR/MT transcripts by using pseudo relevance feedback. • 1 b 2 cs24 kobe1 1: we run the search approach based on an exampled-based topic search method which uses multiple queries to “extensionally ” define topics. Specifically, we use rough set theory for multiple queries. There, we can extract subsets of topics, which are characterized by different combinations of low-level features. Then, by unifying extracted subsets, we conceptualize the topics. From the result, even though our result was less effective than expected, we conducted additional experiment to indicate the possibility of our approach. We have learned our approach is effected by user experience on search performances. So, we are currently developing a new approach which can construct the optimal multiple queries by only selecting a few queries. 1.