Recently, researchers have started to model interactions between users and search engines as an online learning ranking. Such systems obtain feedback only on the few top-ranked documents results. To obtain feedbacks on other documents, the system has to explore the non-top-ranked documents that could lead to a better solution. However, the system also needs to ensure that the quality of result lists is high by exploiting what is already known. Clearly, this results in an exploration/exploitation dilemma. We introduce in this paper an algorithm that tackles this dilemma in Context-Based Information Retrieval (CBIR) area. It is based on dynamic exploration/exploitation and can adaptively balance the two aspects by deciding which user's situation is most relevant for exploration or exploitation. Within a deliberately designed online framework we conduct evaluations with mobile users. The experimental results demonstrate that our algorithm outperforms surveyed algorithms