Extracting Hierarchies of Closed Partially-Ordered Patterns Using Relational Concept Analysis

Abstract

Abstract. This paper presents a theoretical framework for exploring temporal data, using Relational Concept Analysis (RCA), in order to extract frequent sequential patterns that can be interpreted by domain experts. Our proposal is to transpose sequences within relational contexts, on which RCA can be applied. To help result analysis, we build closed partially-ordered patterns (cpo-patterns), that are synthetic and easy to read for experts. Each cpo-pattern is associated to a concept extent which is a set of temporal objects. Moreover, RCA allows to build hierarchies of cpo-patterns with two generalisation levels, regarding the structure of cpo-patterns and the items. The benefits of our approach are discussed with respect to pattern structures.

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

“…There is no main lattice and both relational (inter-lattices) and hierarchical (intra-lattices) links are included within the graphs. Nevertheless, it gives the same results as those presented in [8] when applied to sequential data.…”

“…We recall here some useful properties of the RCA result (proofs are in [8]), which rely on its aforementioned structure, and are used to help the extraction step of CPO-patterns. Briefly, sequential patterns that coexist in the same sequences in D S are revealed by navigating interrelated concept intents.…”

“…This algorithm is applicable to temporal concepts (in this case ipb 1 is replaced with ipb 2 ) as well. Lines [2][3][4][5][6][7][8]: delete all concepts in C next that are upper covers for other concepts in C next , i.e. delete concepts that are not the most specific ones in C next .…”

“…Furthermore, regarding RCA, users have to manually navigate the interrelated lattices in order to highlight the relationships between different categories of objects, which can be a complex task when there are several lattices. To deal with this problem, we proposed in a previous work [8] to synthesise the navigation paths into closed partially-ordered patterns (CPO-patterns [9]), i.e. directed acyclic graphs where vertices are labelled with information extracted from the concepts out of the family of lattices.…”

“…1. formalising an RCA-based approach, referred to as RCA-Seq, which relies on the principles given by [8], and presenting an algorithm CPOHrchy that directly extracts multilevel CPO-patterns by navigating the RCA result; 2. providing a complexity analysis of RCA-Seq based on the complexity of RCA;…”