Mining top-k frequent items in a data stream with flexible sliding windows