Long-Term Learning from User Behavior in Content-Based Image Retrieval

This article describes a simple algorithm for obtaining knowledge about the importance of features from analyzing user log files of a content-based image retrieval system (CBIRS). The user log files of the usage of the Viper web demonstration system are analyzed over a period of four months. In this time about 3500 accesses to the system were made with 800 multiple image queries. The analysis only takes into account multiple image queries of the system with positive or negative input images, because only these queries contain enough information for the method described in the paper. Features frequently present in images marked together positively in the same query step get a higher weighting whereas features present in an image marked positively and another image marked negatively in the same step get a lower weighting. The Viper system offers a very large number of simple features which allows the creation of feature weightings with high values for important and low values for less important features. These weightings for features can of course differ for several collections and as well for several users. The results are evaluated using the relevance judgments of real users and compared to the system without the long-term learning