Skip to Main content Skip to Navigation
New interface

Contributions to pattern set mining : from complex datasets to significant and useful pattern sets

Tatiana Makhalova 1 
1 ORPAILLEUR - Knowledge representation, reasonning
Inria Nancy - Grand Est, LORIA - NLPKD - Department of Natural Language Processing & Knowledge Discovery
Abstract : In this thesis, we study different aspects of pattern mining in binary and numerical tabular datasets. The objective of pattern mining is to discover a small set of non-redundant patterns that may cover entirely a given dataset and be interpreted as useful and significant knowledge units. We focus on some key issues such as (i) formal definition of pattern interestingness, (ii) the minimization of pattern explosion, (iii) measure for evaluating the performance of pattern mining, and (iv) the discrepancy between interestingness and quality of the discovered pattern sets. Moreover, we go beyond the typical perspectives of pattern mining and investigate the intrinsic structure underlying a tabular dataset. The main contributions of this research work are theoretical, conceptual, and practical. Regarding the theoretical novelty, we propose a so-called closure structure and the GDPM algorithm for its computing. The closure structure allows us to estimate both the data and pattern complexity. Furthermore, practically the closure structure may be used to represent the data topology w.r.t. an interestingness measure. Conceptually, the closure structure allows an analyst to understand the intrinsic data configuration before selecting any interestingness measure rather than to understand the data by means of an arbitrarily selected interestingness measure. In this research work, we also discuss the difference between interestingness and quality of pattern sets. We propose to adopt the best practices of supervised learning in pattern mining. Based on that, we developed an algorithm for itemset mining, called KeepItSimple, which relates interestingness and the quality of pattern sets. In practice, KeepItSimple allows us to efficiently mine a set of interesting and good-quality patterns without any pattern explosion. In addition, we propose an algorithm for a greedy enumeration of likely-occurring itemsets that can be used when frequent closed itemset miners return too many itemsets. The last practical contribution consists in developing an MDL-based algorithm called Mint for mining pattern sets in numerical data. The Mint algorithm relies on a strong theoretical foundation and at the same time has a practical objective in returning a small set of numerical, non-redundant, and informative patterns. The experiments show that Mint has very good behavior in practice and usually outperforms its competitors.
Complete list of metadata
Contributor : Thèses UL Connect in order to contact the contributor
Submitted on : Monday, September 13, 2021 - 10:57:25 AM
Last modification on : Thursday, August 4, 2022 - 5:18:49 PM
Long-term archiving on: : Tuesday, December 14, 2021 - 6:39:55 PM


Files produced by the author(s)


  • HAL Id : tel-03342124, version 1


Tatiana Makhalova. Contributions to pattern set mining : from complex datasets to significant and useful pattern sets. Computer Science [cs]. Université de Lorraine, 2021. English. ⟨NNT : 2021LORR0124⟩. ⟨tel-03342124⟩



Record views


Files downloads