A Short Introduction to Sequential Data Mining

1
A Short Introduction to Sequential Data Mining

• Koji IWANUMA
• Hidetomo NABESHIMA
• University of Yamanashi
• The First Franco-Japanese Symposium on Knowledge
  Discovery in System Biology, September 17,
  Aix-en-Provence

2
Two Main Frameworks of Sequential Mining

• Sequential pattern mining for multiple data
  sequences
• Sequential pattern mining for a single data
  sequence

Sequence ID Purchase data record
1 ltbread, cheesegt
2 lt(wheat, milk), bread, (berry, sausage)gt
3 lt(bread, pumpkin, sausage)gt
4 ltbread, cheese, sausagegt
5 ltcheesegt
Data sequence
ltS1 S2 S3 S4 S5 S6 S7 Sngt
3
What Is Sequential Pattern Mining?
J. Han and M. Kamber. Data Mining Concepts and
Techniques, www.cs.uiuc.edu/hanji

• Given a set of sequences, find the complete set
  of frequent subsequences

A sequence lt (ef) (ab) (df) c b gt
A sequence database
An element may contain a set of items. Items
within an element are unordered and we list them
alphabetically.
SID sequence
10 lta(abc)(ac)d(cf)gt
20 lt(ad)c(bc)(ae)gt
30 lt(ef)(ab)(df)cbgt
40 lteg(af)cbcgt
lta(bc)dcgt is a subsequence of lta(abc)(ac)d(cf)gt
Given support threshold min_sup 2, lt(ab)cgt is a
sequential pattern
4
Challenges on Sequential Pattern Mining

• A huge number of possible sequential patterns are
  hidden in databases
• A mining algorithm should
• find the complete set of patterns, when possible,
  satisfying the minimum support (frequency)
  threshold
• be highly efficient, scalable, involving only a
  small number of database scans
• be able to incorporate various kinds of
  user-specific constraints

J. Han and M. Kamber. Data Mining Concepts and
Techniques, www.cs.uiuc.edu/hanji
5
Sequential Pattern Mining Algorithms for Multiple
Data Sequences

• Apriori-based method GSP (Generalized Sequential
  Patterns Srikant Agrawal _at_ EDBT96)
• Pattern-growth methods FreeSpan PrefixSpan
  (Han et al._at_KDD00 Pei, et al._at_ICDE01)
• Vertical format-based mining SPADE (Zaki_at_Machine
  Leanining00)
• Constraint-based sequential pattern mining
  (SPIRIT Garofalakis, Rastogi, Shim_at_VLDB99 Pei,
  Han, Wang _at_ CIKM02)
• Mining closed sequential patterns CloSpan (Yan,
  Han Afshar _at_SDM03)

J. Han and M. Kamber. Data Mining Concepts and
Techniques, www.cs.uiuc.edu/hanji
6
Mining Sequential Patterns from a Very-Long
Single Sequence
A series of daily news paper articles
lt
gt
typhoon
flood, landslide
typhoon
flood, landslide
lttyphoon (flood, landslide)gt
7
Sequential Pattern Mining Algorithms for a Single
data Sequence

• Discovery of frequent episodes in event
  sequences, based on a sliding window system
  Mannila 1998
• The frequency measure becomes anti-monotonic, but
  has a problem, i.e., a duplicate counting of an
  occurrence.
• Asynchronous periodic pattern mining Yang et.al
  2000, Huang 2004
• Any anti-monotonic frequency measures are not
  investigated.
• On-line approximation algorithm for mining
  frequent items, not for frequent subsequences
• Lossy counting algorithm Manku and Motwani,
  VLDB02

8
Research in Our Laboratory

• Sequential Data Mining from a very-large single
  data sequence.
• Main target sequential textual data,
  especially, newspaper-articles corpora
• Objectives to generate a robust and useful
  large-scale event-sequences corpus.
• Application 1 topic tracking/detection in
  information retrieval.
• Application 2 automated content-tracking in WEB.
• Application 3 scenario/story semi-automatic
  creation
• Ordinary temporal data analysis various log
  data in computer systems, genetic information,
  etc.

9
Technical Topics (1/2)

• A new framework for extracting frequent
  subsequences from a single long data sequence
  in IEEE Inter. Conf. on Data Mining 2005
  (ICDM2005)
• A new rational frequency measures, which
  satisfies the Apriori (anti-monotonic) property
  and has no duplicate counting.
• A fast on-line algorithm for a some limited case

10
Technical Topics (1/2)

• On-going current works and future work
• On-line rational filters based on confidence
  criteria and/or information-gain for eliminating
  redundant valueless sequences from system output
• Methods for finding meta-structures embedded in
  huge amount of frequent sequences generated by a
  system
• A method using compression based on context-free
  grammar-inference/learning
• More fast extraction algorithm based on a method
  for simultaneously searching multiple strings
  over compressed data.

11
References

• Jiawei Han and Micheline Kamber. Data Mining
  Concepts and Techniques (Chapter 8).
  www.cs.uiuc.edu/hanj

12
Thanks for your attention!!