Document System plays an important role in information dissemination and utilization. However, with the emergence of information overload, the utilization rate of data would greatly decrease. To solve this problem, a document recommendation system based on multi-granularity features and Hybrid Algorithms is proposed. User interest and document feature models are established on both phrase and term granularities. Then, the system generates recommendation lists for users based on the combination of content-based and collaborative-filtering algorithms. The tests based on authentic data demonstrate that the document recommendation system has a better performance on precision, recall rate, coverage rate and novelty. The recommendation lists are more in line with users' interests. This helps to increase the utilization rate of data and improves user experience with better performance.