Temporal data mining algorithms for metric temporal constraint networks discovery

  1. Rodríguez Álvarez, Miguel
Dirixida por:
  1. Paulo Félix Lamas Director
  2. Purificación Cariñena Amigo Co-director

Universidade de defensa: Universidade de Santiago de Compostela

Fecha de defensa: 22 de outubro de 2013

  1. José Luis Verdegay Galdeano Presidente/a
  2. Richard J. Duro Fernández Secretario/a
  3. José Hernandez Orallo Vogal
  4. Carlo Combi Vogal
  5. Alicia Troncoso Lara Vogal
  1. Departamento de Electrónica e Computación

Tipo: Tese


Temporal representation and reasoning play an important role in multiple areas of Artificial Intelligence, such as natural language processing, planning, scheduling or diagnostics. For long years, the main point in the research activity has been to put on formal grounds the representation of the multiple nuances of meaning for the different temporal notions, and to provide the corresponding inference mechanisms. These efforts have resulted in a set of formal tools that enable to elicit effective problem solving knowledge. The increasing availability of temporal data as a result of the activity of organizations has caused the emergence of the field of Temporal Data Mining, which aims to induce new and useful temporal knowledge from the computer data processing. Initially, the scientific community focused on designing efficient algorithms for mining frequent temporal patterns, where each temporal pattern emphasises a particular temporal order among a set of events. Since then, new algorithms have been proposed in order to induce more expressive information from data, specially focusing on coping with uncertainty, usually in a qualitative form. The present work provides a set of temporal data mining techniques grouped into two main algorithms: ASTPminer and HSTPminer. These algorithms find sets of frequent temporal patterns from a collection of sequences of time-stamped events and episodes. Patterns obtained by both algorithms are represented as temporal constraint networks that provide tools for the treatment of inference, consistency checking or scenario finding tasks, and also allow a data mining user to interact with the mining process by providing previous domain knowledge about the kind of patterns of interest.