A method for discovering clusters of e-commerce interest patterns using click-stream data

• User’s browsing behavior is a comprehensive reflection of their interests.
• A novel method is proposed for discovering clusters of interest patterns.
• A novel rough leader clustering algorithm is developed for large dataset clustering.
• A case study is conducted based on near 3,000,000 clickstream data.

Having a good understanding of users’ interests has become increasingly important for online retailers hoping to create a personalized service for a target market. Generally speaking, user’s browsing behaviors (when looking at websites) represent a comprehensive reflection of their interests. Users with various interests will visit multiple categories and research various items. Their browsing paths, the frequency of page visits and the time spent on each category all vary widely. Based on these considerations, a novel approach to discovering consumers’ interests is proposed and is systematically studied in this paper. The browsing behavior of a number of consumers – including their visiting sequence, frequency and time spent on each category – are mined via the click-stream data recorded on an e-commerce website. Given this behavioral data, we construct an improved leader clustering algorithm and leverage it with a rough set theory in order to generate users’ interest patterns. Furthermore, a case study is conducted based on nearly three million click-stream data, which was collected from one of the largest Chinese e-commerce websites. Using this data, the parameters of the algorithm are tested and optimized to make the algorithm more effective in terms of large data analysis and to make it more suitable for discovering users’ multiple interests. Using this algorithm, three typical user interest patterns are derived based on a real click-stream dataset. More importantly, further calculations based on different click-stream datasets verify that these three interest patterns are consistent and stable. This study demonstrates that the proposed algorithm and the derived interest patterns can provide significant assistances on webpage optimization and personalized recommendation.