Abstract:

Most of the latest works on utility mining generates a huge number of candidates in dealing with big data,which suffers from the scalability issue.Some work does not generate candidates,but suffers from the efficiency issue due to lack of strong pruning and high computation overhead.A novel algorithm that finds high utility patterns in a single phase without generating candidates was proposed.The novelties lie in a prefix growth strategy with strong pruning,and a sparse matrix based representation of transactions with pseudo projection.The proposed algorithm works in a depth first manner and does not materialize high utility patterns in memory,which further improves the scalability.Extensive experiments on synthetic and rea1-world data show that the proposed algorithm outperforms the latest works in terms of running time,memory overhead,and scalability.

Key words: big data, utility mining, high utility pattern, frequent pattern