Discovery of tree structured patterns using Markov chain Monte Carlo method

Yasuhiro Okamoto, Kensuke Koyanagi, Takayoshi Shoudai, Osamu Maruyama

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

A tree contraction pattern (TC-pattern) is an unordered tree-structured pattern which can express a tree-structure common to given unordered trees. A TC-pattern has some special vertices, called contractible vertex, into which every uncommon connected substructure is merged by edge contractions. In this paper, we propose a probabilistic method for computing a binary classification problem on tree-structured data. Given a positive set P and a negative set N of unordered trees with vertex labels on a finite alphabet, the problem is to find meaningful and optimal TC-patterns that classify P and N with high statistical measures. We formalize this problem as a multiple optimization problem, and propose a probabilistic method for computing it by employing enumeration algorithms for TC-patterns and Markov chain Monte Carlo method. In addition, as a theoretical aspect of this problem, we show the hardness of approximability of it. Finally, we show the experimental results of our method on glycan structure data.

Original languageEnglish
Title of host publicationProceedings of the 7th IADIS International Conference Information Systems 2014, IS 2014
PublisherIADIS
Pages95-102
Number of pages8
ISBN (Electronic)9789898704047
Publication statusPublished - Jan 1 2014
Event7th IADIS International Conference on Information Systems, IS 2014 - Madrid, Spain
Duration: Feb 28 2014Mar 2 2014

Other

Other7th IADIS International Conference on Information Systems, IS 2014
CountrySpain
CityMadrid
Period2/28/143/2/14

All Science Journal Classification (ASJC) codes

  • Hardware and Architecture
  • Information Systems
  • Software
  • Computer Science Applications

Fingerprint Dive into the research topics of 'Discovery of tree structured patterns using Markov chain Monte Carlo method'. Together they form a unique fingerprint.

Cite this