Existing pattern mining algorithms typically assume that the dataset can fit into the main memory, while large graph datasets cannot satisfy this condition. Thus mining patterns in large-scale linked data is still a challenge. In this paper we propose a new partition-based approach for pattern mining in linked data which is composed of three steps: dividing linked data into connected typed object graphs, clustering graphs into clusters according to shared patterns and partitioning clusters into size-limited units. A global pattern mining algorithm is proposed, which is used to merge local link patterns into global patterns. Experiments on Semantic Web Dog Food dataset show that our approach is feasible and promising.