Data feature oriented data partition and weighted data mining

It is comprehensible that to find as much interesting knowledge as possible is the initial and main aim to mine data, no matter which pattern (parallel or sequential) is utilized in data mining, though parallelism is practically important as well. We present a principle, called DFDP, for partitioning large dataset-the first step for parallelization. Data subsets after partitioning are treated tendentiously for possible parallel or distributed processing. One feasible logical structure for parallel processing is recommended in the paper. Also experimental comparisons are reported in the paper, which shows that weighted data mining will find more interesting rules from data.