A co-location pattern is a set of spatial features whose objects are frequently located in spatial proximity. Spatial co-location patterns resemble frequent patterns in many aspects. Since its introduction, the paradigm of mining frequent patterns has undergone a shift from a generate-and-test based frequent pattern mining to a projection based frequent pattern mining. However for spatial datasets, the lack of a transaction concept, which is critical in frequent pattern definition and its mining algorithms, makes the similar shift of paradigm in spatial co-location mining very difficult. We investigate a projection based co-location mining paradigm. In particular, we propose a projection based co-location mining framework and an algorithm called FP-CM, for FP-growth Based Co-location Miner. This algorithm only requires a small constant number of database scans. It out-performs the generate-and-test algorithm by an order of magnitude as shown by our preliminary experiment results.
展开▼
