S2MP: Similarity Measure for Sequential Patterns

Saneifar, H., Bringay, S., Laurent, A. and Teisseire, M.

In data mining, computing the similarity of objects is an essential task, for example to identify regularities or to build homogeneous clusters of objects. In the case of sequential data seen in various fields of application (e.g. series of customers purchases, Internet navigation) this problem (i.e. comparing the similarity of sequences) is very important. There are already some similarity measures as Edit distance and LCS suited to simple sequences, but these measures are not relevant in the case of complex sequences composed of sets of items, as is the case of sequential patterns. In this paper, we propose a new similarity measure taking the characteristics of sequential patterns into account. S2MP is an adjustable measure depending on the importance given to each characteristic of sequential patterns according to context, which is not the case of existing measures. We have experimented the accuracy and quality of S2MP against Edit distance by using them in a clustering of sequential patterns. The results show that the clusters obtained by S2MP are more homogeneous. Moreover these cluster are more precise and more complete according to the clusters obtained using Edit distance. The experiments show also that S2MP is efficient in term of calculation time and size of used memory.

Cite as: Saneifar, H., Bringay, S., Laurent, A. and Teisseire, M. (2008). S2MP: Similarity Measure for Sequential Patterns. In Proc. Seventh Australasian Data Mining Conference (AusDM 2008), Glenelg, South Australia. CRPIT, 87. Roddick, J. F., Li, J., Christen, P. and Kennedy, P. J., Eds. ACS. 95-104.

(from crpit.com) (local if available)