Up and down: Mining multidimensional sequential patterns using hierarchies

Abstract

Data warehouses contain large volumes of time-variant data stored to help analysis. Despite the evolution of OLAP analysis tools and methods, it is still impossible for decision makers to find data mining tools taking the specificity of the data (e.g. multidimensionality, hierarchies, time-variant) into account. In this paper, we propose an original method to automatically extract sequential patterns with respect to hierarchies. This method extracts patterns that describe the inner trends by displaying patterns that either go from precise knowledge to general knowledge or go from general knowledge to precise knowledge. For instance, one rule exhibited could be data contain first many sales of coke in Paris and lemonade in London for the same date, followed by a large number of sales of soft drinks in Europe, which is said to be divergent (as precise results like coke precede general ones like soft drinks). On the opposite, rules like data contain first many sales of soft drinks in Europe and chips in London for the same date, followed by a large number of sales of coke in Paris are said to be convergent. In this paper, we define the concepts related to this original method as well as the associated algorithms. The experiments which we carried out show the interest of our proposal. © 2008 Springer-Verlag Berlin Heidelberg.