Navigation Pattern Discovery from Internet Data

Electronic commerce sites need to learn as much as possible about their customers and those browsing their virtual premises, in order to maximize their marketing effort. The discovery of marketing related navigation patterns requires the development of data mining algorithms capable of discovering sequential access patterns from web logs. This paper introduces a new algorithm called MiDAS that extends traditional sequence discovery with a wide range of web-specific features. Domain knowledge is described as flexible navigation templates that can specify navigational behavior, as network structures for the capture of web site topologies, in addition to concept hierarchies and syntactic constraints. Unlike existing approaches, field dependency has been implemented, which allows the detection of sequences across monitored attributes, such as URLs and http referrers. Three different types of contained-in relationships are supported, which express different types of browsing behavior. The carried out experimental evaluation have shown promising results in terms of functionality as well as scalability.