Study on Distributed Sequential Pattern Discovery Algorithm
Xiang ZOU · Journal of Software · 2005
Algorithm FDMSP (fast distributed mining of sequential patterns) is proposed in order to deal with mining sequential patterns in distributed environment and its properties are analyzed. The algorithm utilizes prefix-projected technique to divide the pattern searching space, utilizes polling site associated with prefix to get a global support, and utilizes local pruning, poll pruning and count pruning to decrease candidate sequences. It is divided into three sub-procedures which run asynchronously. As a result, the algorithm has lower I/O cost, memory cost and communication cost, and global sequential patterns are generated with higher efficiency. The experiments show that it outperforms the algorithm GSP after centralizing data by 68.5% to 99.5% and scaleable over LAN with huge amount of data.