In this paper we addressed the problem how social interactions, especially P2P (Peer to Peer) communication, set trends. We made a multiagent-based model to investigate the issue. The agents are computer programs that act autonomously and behave individually. In this artificial society, the agents demand goods that they want to consume as well as they can produce and consume goods. They try to barter their goods with one of the other agents that is selected randomly. They make their own evaluation of each good based not on global information in the market but on local information shared with their trading partners: They assume the goods, which their trading partners demand, to be popular. They can also demand the goods that they think very popular by themselves and exchange their evaluation of the popularity when they trade with other agents. We pointed out that, in this situation, agents' evaluation of the popularity can be concentrated on some good one after another, which can explain a mechanism of the concentration of popularity.