Trust evaluation is a key issue in the interaction between network entities in open networks. The attacks of malicious entities have become a major obstacle to the development of open networks. Few traditional trust models have considered incorporating incentive mechanisms to reduce the influence of recommendation values from malicious entities in trust evaluation. This paper proposed a dynamic evaluation of recommendation trust model, considering the interaction procedure between entities, introducing reward-punishment factor and evaluation reliability factor. The function of reward-punishment factor is to reward honest interactions between entities while punishing fraudulent interactions. The evaluation reliability factor is used to decide whether to accept the recommendations from the recommending entities. Simulation results show that the model could effectively reduce the influence of malicious entities in trust evaluation. The proposed model could accurately and reliably identify the access behaviour of malicious entities, and adopt appropriate processing countermeasure to ensure the accuracy and fault tolerance of calculation.