Volume 5 Issue 1
Spring 2009
ISSN 1937-7266

Exploiting Log Files in Video Retrieval

Frank Hopfgartner, Thierry Urruty and Joemon M. Jose

University of Glasgow
Glasgow, United Kingdom
{hopfgarf, thierry, jj}@dcs.gla.ac.uk

Most interactive video retrieval systems are benchmarked in laboratory based user experiments. However, to make a robust measurement, the evaluation must be based on a large user population, which is very expensive. An alternative way of evaluating such systems is the use of simulations.

We adapted such evaluation methodology in simulating users interacting with a facet-based video retrieval interface by exploiting log files from a user-study. We used the log data to try and develop an approach which is more effective than the introduced system B. In the scope of this work, we investigated if the content of other facets can help improve the retrieval results of each facet. Therefore, we incorporated the content of other facets in a simulation run S. In the simulation, we took advantage of the explicit relevance feedback given by each user in marking shots as relevant for a facet. We used these shots as a query expansion source and determined query candidate terms for each facet. If a term appeared in more than one facet, we removed it from the facet which contained the most candidate terms and used these terms as a new search query. In other words, we reduced the number of query terms in a facet, when the query term was used in another facet with less query terms at the same time. This results in a more focused retrieval for the facets, as double entries are avoided. Afterwards, we compared the retrieval results of the original queries used in B with the result list coming from S.

For a larger view of the poster, click here. After the image opens, click on it to zoom.