A Computational Geometry Approach to Web Personalization Maria Rigou 1,3 1 Research Academic Computer Technology Institute, 61 Riga Feraiou Str. 26110 Patras, Greece Spiros Sirmakessis 1,2 2 Technological Educational Institution of Messolongi Nea Ktiria, Messolongi, Greece {rigou, syrma, tsak}@cti.gr Athanasios Tsakalidis 1, 3 3 Department of Computer Engineering and Informatics University of Patras, 26500 Patras, Greece Abstract In this paper we present an algorithm for efficient personalized clustering. The algorithm combines the orthogonal range search with the k-windows algorithm. It offers a real-time solution for the delivery of personalized services in online shopping environments, since it allows on-line consumers to model their preferences along multiple dimensions, search for product information, and then use the clustered list of products and services retrieved for making their purchase decisions. 1. Introduction Nowadays, Internet has become one of the largest data repositories and the increasing number of people that use it as information source face the problem of information overload. In this difficult to manage volume of data, web users are trying to identify explicit information that satisfies their needs and suits their preferences. When the results matching a user’s query, are of a size that does not allow for easy manipulation and hinder –instead of favoring- decision-making, further processing must be applied, so that results are presented in a way that can help users evaluate all available alternatives and perform relative comparisons before proceeding with the online purchase. Applications that operate in such an assistive manner are already available for online shopping (usually mentioned as shopping aids) and appear to have strong favorable effects on both the quality and the efficiency of purchase decisions [10]. Clustering is a data mining technique [16] that has been extensively studied and used in numerous cases where big volumes of information must be handled in a way that allows for knowledge extraction. Cluster analysis aims to discover objects that have representative behavior in the collection. The basic idea is that if a rule is valid for one object, it is very possible that the rule also applies to all the objects that are very similar to it. Algorithms for clustering data have been widely studied in various fields including Machine Learning, Neural Networks, Databases and Statistics. Based on the characteristics or attributes used [1], the available clustering algorithms can be categorized into text-based [9], [11], link-based [4], [5], and hybrid like [15], [12]. The utility of clustering in the web personalization domain and more especially in e-Commerce lies in the so- called cluster hypothesis; given a ‘suitable’ clustering of a product collection, if the user is interested in product p, he is also likely to be interested in other members of the cluster to which p belongs. As with products, we can set up a bipartite relation for people liking or being interested in products, and use this to cluster both people and products, with the premise that similar people like similar products, and vice versa. This important ramification of clustering is called collaborative filtering [8] and is the basis for the majority of recommender systems we meet in today’s e-stores. Personalization on the other hand, is an approach that has already spread widely in the web and is used -in some form- by all well-established web shopping environments. Tracing back its roots one ends up at the introduction of adaptive hypermedia applications in Brusilovsky’s work of 1996 [6] and its updated version of 2001[7]. Adaptive hypermedia was introduced as an alternative to the traditional “one-size-fits-all” approach with the purpose of addressing the specific needs of individual users. Personalization on the web today covers a broad area, ranging from check-box customisation to recommender systems, and adaptive web sites. In this paper we are focusing on personalization for electronic commerce. With the enormous and ongoing growth of products and services available from different sources for online transactions, many online customers face the problem of putting together the appropriate list of products based on their needs. This is where personalization may come into the picture and transparently ‘deliver’ to users tailored products and services. In our context, personalization takes up the form of providing the online purchaser the ability to specify the range of individual preferences (in terms of product features such as price, functionalities, appearance, size, etc.). Using this range as input, e-stores can process online user requests and return products that reside inside the preference spectrum of the individual customer. The idea behind this is to accomplish effective (in terms of Proceedings of the IEEE International Conference on E-Commerce Technology 0-7695-2098-7/04 $20.00 © 2004 IEEE