CWI at ImageCLEF 2008
Theodora Tsikrika, Henning Rode, Arjen P. de Vries · 2008
CWI used PF/Tijah, a flexible XML retrieval system, to evaluate image retrieval based on textual evidence in the context of the wikipediaMM task at ImageCLEF 2008. We employed a language modelling framework and found that the text associated with the Wikipedia images is a good source of evidence. We also investigated a length prior and found that biasing towards images with longer descriptions than the ones retrieved by our language modelling approach is not beneficial.