Experiment for Using Web Information to do Query and Document Expansion

ImageCLEF photo task of this year is a little different from those of previous years. The caption field in image annotations and the narrative field in the text queries are removed, and the visual queries (example images) are also removed from the image collection too. In the new definition, the information that can be employed for queries and images is less than before, so that it becomes harder to match query words and annotations directly. To deal with this issue, we explore the web to expand queries and documents. Many images and text information can be found in the web, but we should face the noise embedded. The experiment shows the query expansion improves performance about 16.11%. The document expansion brings too much noise and the performance decrease 28.24% after expansion. The media mapping method that we proposed in previous years is used for query expansion too. The results of formal runs show this method is still very useful in the new task definitions. ACM