Reducing the Frequent Pattern Set

Abstract

One of the major problems in frequent pattern mining is the explosion of the number of results, making it difficult to identify the interesting frequent patterns. In a recent paper [7] we have shown that an MDL-based approach gives a dramatic reduction of the number of frequent item sets to consider. Here we show that MDL gives similarly good reductions for frequent patterns on other types of data, viz., on sequences and trees. Reductions of two to three orders of magnitude are easily attained on data sets from the web-mining field.