Artificial agents engaged in real world applications require accurate allocation strategies in order to better balance the use of their bounded resources. In particular, they should be capable to filter out all irrelevant information and just to consider what is relevant for the current task that they are trying to solve. The aim of this work is to propose a mechanism of relevance-based belief update to be implemented in a BDI cognitive agent. This in order to improve the performance of agents in information-rich environments. In the first part of the paper we present the formal and abstract model of the mechanism. In the second part we present its implementation in the Jason platform and we discuss its performance in simulation trials.