An Algorithm for Mining Approximate Frequent Itemsets Over Data Streams

摘要

It is much more difficult to mining frequent itemsets over data streams than traditional data model because data stream has the following characters: unbounded volume of data,rapid arriving rate of records,uncontrollability of records' arriving order,etc. A novel algorithm is devised based on Lossy Counting to mine frequent itemsets. Logarithmic tilted time window with an attenuation coefficient is adopted to emphasize the importance of new data. Multilayer count queue mode is designed to not only avoid the counter overflowing but also query top-K itemsets quickly using a index table.