Using a Divergence Model for MediaEval's Tagging Task (Professional Version)
Christian Wartena · 2010
Novay participated in MedieaEval Tagging Task (professional version). For this task videos have to be ranked according to their relevance for a number of different concepts. Our approach was based solely on the abstracts of the videos. A divergence model has been used for retrieval in which both the query and the document model are extended by a Markov chain. The results could be improved by using a small set of synonyms to represent each concept, by enlarging the basis for computing the language models of the query terms, and by also taking into account the rank the concept has for the abstract. The best result was obtained by combining all of these possibilities.