Seminar on Computational Learning and Adaptation


  Golden Path Analyzer: Using Divide-and-conquer to cluster Web Clickstreams

Kamal Ali

Yahoo

This talk presents a novel algorithm and deployment that analyzes clickstreams during a web browsing session with respect to a success criterion such as ability to easily navigate a web site to purchase a product. It finds the shortest 'golden' paths taken by users (panelists) who succeeded at the task. The paths taken by the rest of the users are then analyzed with respect to each golden path. GPA determines whether a given user took a golden path or not, where she dropped off that golden path, and whether or not she rejoined that golden path or joined another path. These analyses allow one to find which web pages are problematic, i.e. those on which a substantial percentage of the visitors drop off. They also allow one to identify links that are problematic. A link is deemed problematic if it distracts users from proceeding on a golden path. The system also provides a mechanism that allows one to determine what percentage of the panelists used one path (eg: search) versus another (eg: browse) to get to a target page. The system has been used in 20 client engagements, has been implemented in Perl and Visual Basic, runs in a Win2k/Intel environment and outputs a bundle of interlinked HTML pages.



Date: Wednesday, January 21

Time: 4:15-5:30PM

Place: Cordura 100


Return to the seminar schedule