Similarity assessment in a cbr application for clickstream data mining plans selection

Abstract

We implemented a mining plans selection system founded on the Case Based Reasoning paradigm, in order to assist the development of Web usage mining processes. The system's main goal is to suggest the most suited methods to apply on a data analysis problem. Our approach builds upon the reuse of the experience gained from prior successfully mining processes, to solve current and future similar problems. The knowledge acquired after successfully solving such problems is organized and stored in a relational case base, giving rise to a (multi-) relational cases representation. In this paper we describe the similitude assessment devised within the retrieval of similar cases, to cope with the adopted representation. Structured representation and similarity assessment over complex data are issues relevant to a growing variety of application domains, being considered in multiple related lines of active research. We explore a number of different similarity measures proposed in the literature and we extend one of them to better fit our purposes.