Clustering Users by Information-Seeking Style: An Empirical Study on an Academic Search Engine

Nowadays, academic search engines have become indispensable tools for getting important online scholarly information. User differences are important factors that influence the use of information systems. The way people use academic search engines to find information varies depending on their information-seeking style. Therefore, finding and understanding different information-seeking behaviors has become an important line of research. User behavior patterns can be discovered by examining user interaction logs to determine who the users are and what they intend to do. These insights can be useful in designing more optimized academic search engines. In this paper, we analyze the user interaction logs collected from the Iranian scientific information database. The Ganj database is the official repository for collecting and organizing theses and dissertations in Iran. Many researchers search scientific and research resources from the Ganj database daily. We use a sequential pattern mining approach to extract frequent sequential behavior patterns on user interaction logs and to cluster users into three groups based on their frequent behavior patterns, using the K-means clustering algorithm. Cluster analysis shows that users with similar frequent behavior patterns have similar information-seeking styles. Finally, we found three clusters and named them: fast surfers, deep divers, and broad scanners. Our findings can help developers of academic search engines and policymakers to identify users' needs and priorities to make better decisions to design a reasonable page layout and well-organized website for all users based on their search styles.