Database Reference
In-Depth Information
Chapter 4
Mining Long Patterns
Feida Zhu
Abstract The value and importance of long patterns are gaining increasing recog-
nition in a wide range of domains including bioinformatics, social network analysis,
software engineering and business intelligence. Yet the task of mining long patterns
has remained a challenge due to the prohibitively large number of smaller patterns
which often need to be generated first. In this chapter, we first use a pattern lat-
tice model to illustrate and compare various mining paradigms. Then we present
recent studies for mining long patterns according to their respective pattern min-
ing paradigms. For each category, we discuss the representative algorithms and the
state-of-the-art development.
Keywords Frequent pattern
·
Long pattern
·
Colossal pattern
·
Large pattern
1
Introduction
Pattern mining has been a central theme for data mining since its inception. Among
the various constraints often imposed on the patterns to be mined, patterns of large
sizes are of growing interest for a number of reasons. Firstly, long patterns are
a natural result of ever larger data sets. For example, social network analysis on a
network like that of Facebook or Twitter has been shown that functional communities
could reach size up to 150, much larger than what most algorithms can typically mine.
Similarly, for web structure mining in today's Internet, one should expect the real web
structures mined for any domain to be fairly complicated. Secondly, long patterns
are more informative in characterizing large data sets and in many cases (e. g.,
bioinformatics) give more meaningful insights than shorter patterns. For example, in
DBLP co-authorship network, small patterns, e.g., several authors collaborate on a
paper, are almost ubiquitous. It is shown in [ 33 ] that only long patterns would reveal
interesting common collaborative patterns, or distinguish distinct patterns, across
different research communities. In software engineering, long patterns uncovered
Search WWH ::




Custom Search