Hierarchies of Weighted Closed Partially-Ordered Patterns for Enhancing Sequential Data Analysis

Cristina Nica (1) , Agnès Braud (1) , Florence Le Ber (1)

1 ICube - Laboratoire des sciences de l'ingénieur, de l'informatique et de l'imagerie

Discovering sequential patterns in sequence databases is an important data mining task. Recently, hierarchies of closed partially-ordered patterns (cpo-patterns), built directly using Relational Concept Analysis (RCA), have been proposed to simplify the interpretation step by highlighting how cpo-patterns relate to each other. However, there are practical cases (e.g. choosing interesting navigation paths in the obtained hierarchies) when these hierarchies are still insufficient for the expert. To address these cases, we propose to extract hierarchies of more informative cpo-patterns, namely weighted cpo-patterns (wcpo-patterns), by extending the RCA-based approach. These wcpo-patterns capture and explicitly show not only the order on itemsets but also their different influence on the analysed sequences. We illustrate how the proposed wcpo-patterns can enhance sequential data analysis on a toy example.

Domaines: Intelligence artificielle [cs.AI]

Cristina Nica, Agnès Braud, Florence Le Ber. Hierarchies of Weighted Closed Partially-Ordered Patterns for Enhancing Sequential Data Analysis. Int. Conference on Formal Concept Analysis, Jun 2017, Rennes, France. ⟨hal-01521562⟩