In decentralized interaction-based reputation systems, nodes store information about the past interactions of other nodes. Based on this information, they compute reputations in order to take decisions about future interactions. Computing the reputations with the complete history of interactions is inefficient due to its resource requirements. Furthermore, the complete history of interactions accumulates old information, which may impede the nodes from capturing the dynamic behavior of the system when computing reputations. In this paper, we propose a scheme for reducing the amount of history maintained in decentralized interaction-based reputation systems based on elements such as the age of nodes, and we explore its effect on the computed reputations showing its effectiveness in both synthetic and real-world graphs.