The Internet has accelerated the public opinion formation, and public opinion discussions are happening all the time. In order to explore the influence of heterogenous agents on opinion evolution, this paper modifies the traditional DeGroot model and considers the opinion evolution where different kinds of agents coexist in the social network. Agents are divided into authority and non-authority. From the perspective of social psychology, the non-authoritative agents are randomly classified into conformists, anticonformists and independent agents. To overcome the shortcoming of the traditional DeGroot model in determining weights, a new weight determination method is proposed by introducing the eigenvector centrality of nodes. There are two stages in each time step of the proposed model. In the first stage, all agents update their opinions through the DeGroot method. In the second stage, the anticonformists and independent agents are assumed to adjust their opinions obtained in the first stage. Simulation results in the connected Erdddotos-Rényi networks show that the conformists reach a consensus with the authority and the consensus is determined by the authority's opinion. The anticonformists form their opinion clusters that are far away from the conformists and the authority. The stable opinions of the independent agents are scattered between the anticonformists and authority and are closer to the consensus of the conformists and the authority.