Abstract

This talk describes research carried out at Microsoft Bing, regarding user behavior when interacting with search engines, and more generally, browsing the Web. Analysis is based on proprietary large-scale interaction log user data collected in mid 2009. A new action-topic task breakdown of user activity is presented, and compared to those of Broder 2002 and Russell et al 2009. Tasks are characterized by popularity, pivoted on search-specific and general Web activity. The Compare task is explored in more detail, and we detail a sub-task breakdown. Key takeaways about user behavior include: (1) information needs manifest as user tasks which often require multiple queries; (2) such tasks are complex, and may involve sub-tasks; (3) significant time is dedicated to these tasks; (4) user modeling and user task modeling is critical to true understanding of information interaction.

‚Äč