Paper Reading #23 - Automatic Generation of Research Trails in Web History

Comments:

Reference:

Automatic Generation of Research Trails in Web History

Elin Pedersen, Karl Gyllstrom, Shengyin Gu, Peter Jin Hong

IUI '10

Summary:

After conducting an ethnographic study, the authors of this paper determined that while people are using the internet to conduct extensive research, they are not following traditional scolarly and investigative methods.  The research conducted usually falls into one of the following categories: personal consumption, fragmented process, topic sliding, or premature structure.  The authors believe that tools designed to track research sessions and provide context to the researcher upon subsequent research would be beneficial.  Research trails may be used to do just that by grouping events that the user perceives belong in the same category and listing them as temporally ordered lists of segments.

Research trails helps the researcher by showing them where they are in the current activity.  This will help answer possible questions like "what did I leave unfinished?" and "where did I leave of last time I worked on this?"  It also extracts activity-based and semantic information from user activity.  Enough ambiguity is allowed to allow users to switch between topics based on timely affinity.  Events are temporally grouped into specific periods of activity called segments.  A segment is bounded when more than five minutes pass between two consecutive events.  Each event is provided with a topic vector, a list of its coverage of automatically determined set of topics.  These vectors allow for semantic analysis of the events.  In their preliminary assessment, the authors found that their segment definition naturally captured the concept of a work session.  They also found that segments typically were related and coherent locally, though there were occasionally both unrelated segments in trails and related segments excluded from trails.  In the future, they plan to capture a richer set of activity data, such as user activity on a visited page.

Discussion: This concept certainly seems like it would be a boon to researchers.  Aside from tracking the sites visited while doing research, it provides context and relations between the visited resources.  The automated aspect is somewhat of a double-edged sword, for while it makes things easier for the researcher overall, in doing so it removes the responsibilities of the researcher to keep track of their work and increases their overall expectations that everything should be done for them by default.  Still, this seems like a valuable tool that I wouldn't mind using myself.