With the development of Natural Language Processing, it’s achievable to review the composition by computer instead of human. The convenience of the technique also bring about large number of users and compositions, it’s useful for improving the users’ writing ability by recommending compositions which the user is interested in accurately. In this article, a new recommendation algorithm which uses probabilistic topic model in personal recommendation is put forward and implemented in the CMET. The article introduces the Collaborative Filtering Recommendation, the Correlated Topic Model and the improvement on the traditional Collaborative Filtering Recommendation Algorithm. Based on the improved Algorithm, the article puts forward a new recommendation algorithm: Collaborative Filtering based on LDA and implements the algorithm.