Abstract: Shared knowledge is focused on the problems of knowledge discovery shared by two(or more) applications/datasets,and it can help users investigate a poorly understood dataset from a well understood dataset by analogy and transferring knowledge．However,with the advent of the era of big data,the existing algorithms based on serialized can not able to meet the need of the rapid data growth and the serial algorithm has low efficiency in big datasets．The Parallel Shared Decision Tree algorithm(PSDT) based on Hadoop was introduced by using cloud computing．The traditional attributes list structure is uses in the PSDT．But the overmuch operations of I/O occur in the parallel processes of this algorithm,which will influence the algorithm’s performance．A new Hybrid Parallel Shared Decision Tree algorithm(HPSDT),which uses hybrid data structures,was proposed．The algorithm applies attributes-list to compute the split indicators parallelly,and data records-structure to split in the splitting procedure．Data analysis indicates that the algorithm of HPSDT simplifies the splitting process,and its operations of I/O are 0.34times of the PSDT．The experimental results show that PSDT and HPSDT have good parallelism and scalability,furthermore,the performance of HPSDT is better than that of PSDT．Especially,the superiority is even more obvious with the increase of the dataset size.

Key words: Shared decision tree,Parallel shared decision tree,Hybrid data structure,Cloud computing,Hadoop