PATTERN MINING OVER NOMINAL EVENT SEQUENCES USING CONSTRAINT RELAXATIONS

January 5, 2005

One of the main unresolved problems that arises in the data mining process is treating data that contains temporal information. A complete understanding of this phenomenon requires that the data should be viewed as a sequence of events. Sequential pattern mining is the approach most commonly used to explore nominal sequences, enabling the discovery of frequent sequential patterns. The main drawback of algorithms for this task has been their lack of focus on user expectations and the high number of discovered patterns. However, in some cases, the solution most commonly adopted, based on the use of constraints, can transform the mining process into a hypothesis-testing task. This risk is even stronger when mining sequential data, where more restrictive constraints, like regular languages, have been used. In this dissertation, we argue that it is possible to use constrained sequential pattern mining algorithms, over nominal data, to discover unknown information, keeping the process centered on the user. In order to demonstrate the validity of this thesis, we propose a new methodology based on the use of Ω-constraint relaxations. A constraint relaxation establishes a weaker condition than the original constraint, making possible the discovery of patterns that are not completely accepted by the original constraint. We propose a new hierarchy of relaxations that ranges from conservative ones, to approximately accepted and non-accepted relaxations. Additionally, we propose several extensions to existing sequential pattern mining algorithms, in order to deal efficiently with constraints and, in particular with gap and Ω-constraints

Authors in the community:

• 

  C

  Claudia Martins Antunes

  ist14046

  Affiliations and sources

  IST > DEI

  Departamento de Engenharia Informática

  In: cienciavitae

  ID: 4934079

PhD in Computer Science and Engineering

computer-and-information-sciences - Computer and information sciences

eng - English

http://web.ist.utl.pt/claudia.antunes/artigos/antunes05phd.pdf