MINING RELAXED GRAPH PROPERTIES IN INTERNET.pdfVIP

MINING RELAXED GRAPH PROPERTIES IN INTERNET Wilhelmiina H m l inen Depa rtment of Comp uter Science, University of J oensuu, P.O. Box 111, FIN-80101 Joensuu, Finland whamalai@cs.j oensuu.fi Hannu Toivonen Depa rtment of Comp uter Science, University of Helsinki P.O. Box 26, FIN-00014 University of Helsinki, Finland htoivone@cs.helsinki.fi Vladimir Poroshin Depa rtment of Comp uter Science, University of J oensuu, P.O. Box 111, FIN-80101 Joensuu, Finland vp oro@cs.jo ensuu.fi ABSTRACT Many real world datasets are represented in the form of graphs. The classical graph properties found in the data, like cliques or independent sets, can reveal new interesting information in the data. However, such properties can be either too rare or too trivial in the given context . By relaxing the criteria of the classical properties, we can find more and to- tally new patterns in the data. In this paper, we define relaxed graph properties and study their use in analyzing and processing graph-based data. Especially, we consider the problem of finding self-referring groups in WWW, and give a general algorithm for mining all such patterns from a collection of WWW pages. We suggest that such self-referring groups can reveal web communities or other clustering in WWW and also facilitate in compression of graph-formed data. KEYWORDS graph, data mining, WWW. 1.