UNED at iCLEF 2009: Analysis of Multilingual Image Search Sessions

In this paper we summarize the analysis performed on the logs of multilingual image search provided by iCLEF09 and its comparison with the logs released in the iCLEF08 campaign. We have processed more than one million log lines in order to identify and characterize 5, 243 individual search sessions. We focus on the analysis of users’ behavior and their performance trying to find possible correlations between: a) the language skills of the users and the annotation language of the target images; and b) the final outcome of the search session. We have observed that the proposed task can be considered as easy, even though users with no competence in the annotation language of the images tend to perform more interactions and to use cross-language facilities more frequently. Usage of relevance feedback is remarkably low, but successful users use it more often.