Mining TopK Frequent Items in a Data Stream with Flexible Sliding 挖掘topK频繁的项目在一个灵活的滑动数据流.pptVIP

/ name of department PAGE * * Memory Usage Sligro Data / name of department PAGE * * Conclusions We prove that the exact solution for the top-k frequent item problem requires a prohibited amount of memory However it is possible to summarize the stream with very small memory usage to solve the problem with high accuracy / name of department PAGE * * Questions Thank you very much ! * * * * * Assume that we now the important items, we can do a lot of useful things for examples: we can do resource allocation better of prediction burst early. For example in a stream of products sold out by a supper market we can measure how likely the items would be sold out in the very near future and consider this probability as a factor of item importance. * So the crucial task is to measure the importance. There are a lot of such measures in practice. For example the most natural way to do this is considering the relative frequency of item in the data stream. The relative frequency of an item is fraction between the number of occurrences of the given item and the stream size . Or in the application that consider the recent occurrence of items being more important we have two measures that use in practice. The last measure called Max frequency is a recently proposed measure. We will discuss this measure in more detailed because of the following reasons. * This measure is totally independent from the parameters setting while the others are not. The second property is that is consider the recent occurrence of item more important than the occurrence in the past but it can also capture the historical occurrence of item in the data stream * So now I am introducing you the Max-Frequency measure. The main idea is to consider a window such that relative of a given item in that window is maximized. In this example we will show you how to calculate the max-frequency of an item a at the current time. Well we just consider every possible window, (by the term window I mean the suffix