Much research has been done on the calculation of trust, impression and reputation, as well as using these information to decide whether to cooperate with other agents in cooperative environments. However, little is about how to use these information to help agents make decision on whether to believe a particular message when the message sender has intention to be honest as well as dishonest, and make decision on whether to lie. In this paper, we describe a framework to help agents make these decisions in a semi-competitive environment, and show that agents adopting the proposed model have better performance than agents adopting previous models or strategies.