Group recommendation has attracted significant research attention in recent years. In view of the problem of insufficiency of fusion strategy in group recommendation which cause the low accuracy, we propose an novel method to improve preference fusion strategy to increaandse the accuracy. We introduce the concept of item type proportion factor and calculate the type similarity between group type preference and user type preference. Meanwhile we design a score fusion formula to predict item score for the group. Finally, we carry out experiments and compare our method with several classic group recommendation methods using Movielens dataset. The results show that our method achieves higher recommendation accuracy than all the baseline methods.