Document expansion for image retrieval

Successful information retrieval requires e�ective matching\ud between the user's search request and the contents of relevant\ud documents. Often the request entered by a user may\ud not use the same topic relevant terms as the authors' of the\ud documents. One potential approach to address problems\ud of query-document term mismatch is document expansion\ud to include additional topically relevant indexing terms in a\ud document which may encourage its retrieval when relevant\ud to queries which do not match its original contents well. We\ud propose and evaluate a new document expansion method\ud using external resources. While results of previous research\ud have been inconclusive in determining the impact of document\ud expansion on retrieval e�ectiveness, our method is\ud shown to work e�ectively for text-based image retrieval of\ud short image annotation documents. Our approach uses the\ud Okapi query expansion algorithm as a method for document\ud expansion. We further show improved performance can be\ud achieved by using a \document reduction" approach to include\ud only the signi�cant terms in a document in the expansion\ud process. Our experiments on the WikipediaMM task at\ud ImageCLEF 2008 show an increase of 16.5% in mean average\ud precision (MAP) compared to a variation of Okapi BM25 retrieval\ud model. To compare document expansion with query\ud expansion, we also test query expansion from an external resource\ud which leads an improvement by 9.84% in MAP over\ud our baseline. Our conclusion is that the document expansion\ud with document reduction and in combination with query expansion\ud produces the overall best retrieval results for shortlength\ud document retrieval. For this image retrieval task, we\ud also concluded that query expansion from external resource\ud does not outperform the document expansion method