Methods For Mining Frequent Items Over Data Stream Base On Time Windows

Mining frequent items is a kind of technique in stream data mining area. It takes an important role on both research and application. Frequency counting algorithms can be divided into two class:algorithms based on counting, which are used to find frequent items, and based on sketch, which are used to get data distribution. There are two problems of classical frequent-items mining algorithms, one is frequency of items is not accurate and the other one is combination of query time window and counting time window. Users cannot specify query time window freely. We study on both counting and sketch algorithms and propose a frequent-items mining algorithm. We propose SReEC and RFreq algorithms to get more exact frequency of items, and then solve the query time window problem by divide and merge of counting time window. Finally we get the algorithm to query Top-K frequent items in any query time window, which is one time period of counting time window.