Papers by Author: Ke Ming Tang

Paper TitlePage

Authors: Shi Xi Tang, Ke Ming Tang
Abstract: To solve the problem that the duration is too long when the change of the measuring process data of industrial real-time database appears slow, the double compression algorithm with linear interpolation of dynamic threshold is advanced. Good compression quality and reduction effect of process data is accessed. The compression and fitting times are less than 1ms, and the compression ratio is 6:1 on the 6000 real-time data compression. The compression time is short, and the compression ratio is high. The fitting error between each fitted value and the original value is within the threshold 1.0. The comprehensive performance is better than SLIM and Swinging Door compression algorithm.
Authors: Ke Ming Tang, Cai Yan Dai, Ling Chen
Abstract: Mining closed frequent itemsets in data streams is an important task in stream data mining. Most of the traditional algorithms for mining closed frequent itemsets are Apriori-based which find the frequent itemsets from large amount of candidates, and needs a great deal of time and space. In this paper, an algorithm ItemListFCI for mining closed frequent itemsets in data stream is proposed. The algorithm is based on the sliding window model, and uses a ItemList where the transactions and itemsets are recorded by the column and row vectors respectively. The algorithm first builds the ItemList for the first sliding window. Frequent closed itemsets can be detected by pair-test operations on the binary numbers in the Table. After building the first ItemList, the algorithm updates the ItemList for each sliding window. The frequent closed itemsets in the sliding window can be identified from the ItemList. Algorithms are also proposed to modify ItemList when adding and deleting a transaction. The experimental results on synthetic and real data sets indicate that the proposed algorithm needs less CPU time and memory than other similar methods.
Showing 1 to 2 of 2 Paper Titles