RCA-Seq: An original approach for enhancing the analysis of sequential data based on hierarchies of multilevel closed partially-ordered patterns

Abstract

Methods for analysing sequential data generally produce a huge number of sequential patterns that have then to be evaluated and interpreted by domain experts. To diminish this number and thus the difficulty of the interpretation task, methods that directly extract a more compact representation of sequential patterns, namely closed partially-ordered patterns (CPO-patterns), were introduced. In spite of the fewer number of obtained CPO-patterns, their analysis is still a challenging task for experts since they are unorganised and besides, do not provide a global view of the discovered regularities. To address these problems, we present and formalise an original approach within the framework of Relational Concept Analysis (RCA), referred to as RCA-Seq, that focuses on facilitating the interpretation task of experts. The hierarchical RCA result allows to directly obtain and organise the relationships between the extracted CPO-patterns. Moreover, a generalisation order on items is also revealed, and multilevel CPO-patterns are obtained. Therefore, a hierarchy of such CPO-patterns guides the interpretation task, helps experts in better understanding the extracted patterns, and minimises the chance of overlooking interesting CPO-patterns. RCA-Seq is compared with another approach that relies on pattern structures. In addition, we highlight the adaptability of RCA-Seq by integrating a user-defined taxonomy over the items, and by considering user-specified constraints on the order relations on itemsets.