The traditional recommendation system is mainly for a single user,but people are increasingly participating in activities in the form of multiple users.Group recommendation is intended to serve groups of multiple users and has already become one of the hotspots of current research.Aiming at the low accuracy of current group recommendation and the difficulty of integrating preference conflicts among group members,this paper proposes a new consensus model strategy,which combines group leader impact factor and project heat impact factor,and is based on K Nearest Neighbor recommendation.A group recommendation algorithm based on preference aggregation is designed to find the neighbor groups for the target group and draw on the preference of the neighbor groups.The experimental results on the MovieLens dataset show that the aggregation strategy proposed in this paper has better performance than traditional preference aggregation strategy.The overall average performance of the recommended accuracy (measured by normalized Discounted Cumulative Gain,nDCG) is increased by about 13%,and the overall average performance of the recommendation list diversity is improved by about 10%.