With rapid growth of mobile applications, users of mobile app platforms are facing problem of information overload. Large number of apps make it difficult for users to find appropriate ones, while many long tail apps are submerged in the resource pool and are unknown to most users. Meanwhile, existing recommendation methods usually pay more attention to accuracy than diversity, making popular apps as most recommended items. As a result, the overall exposure rate of mobile apps is low as behavior data accumulated by the system is gradually biased towards popular apps, which leads to a poor recommendation performance in the long run. To solve this problem, this article first proposes two recommendation methods, named LDA_MF and LDA_CF, to improve existing methods. LDA_MF combines user topic model and app topic model with matrix factorization model MF, and LDA_CF takes both tag information of apps and user behavior data into consideration. In order to take advantages of different algorithms and increase the diversity of recommendation results without sacrificing accuracy, a hybrid recommendation algorithm is also provided to combine LDA_MF, LDA_CF and item-based collaborative filtering models. A large real data set is used to evaluate the proposed methods, and the results show that the presented approach achieves better diversity and good recommendation accuracy.