Balancing Exploration – Exploitation in Image Retrieval
Dorota Głowacka, Sayantan Hore · Työväentutkimus Vuosikirja · 2014
Abstract. In recent years there has been an increased interest in developing exploration–exploitation algorithms for image search. However, little research has been done as to what type of image search such techniques might be most beneficial. We present an interactive image retrieval system that combines Rein-forcement Learning with an interface designed to allow users to actively engage in directing the search. Reinforcement Learning is used to model the user interests by allowing the system to trade off between exploration (unseen types of image) and exploitation (images the system thinks are relevant). A task-based user study indicates that for certain types of searches a traditional exploitation-based sys-tem is more than adequate, while for others a more complex system trading off exploration and exploitation is more beneficial. Image retrieval techniques operating on meta-data, such as textual annotations, have become the industry standard. However, with the explosive growth of image collections, tagging new images quickly is not always possible. Secondly, there are many instances where image search by query is problematic, e.g. finding an illustration for an article