THE PREDICTION OF WEB USER TASKS BY ANALYZING CLIENT LOGS
Author(s):
Anne Gutschmidt
Paper abstract: Not only a web user's long-term preferences play a role in offering personalized content and interfaces.
Above all, it is the user's information need at the very moment which is of essential importance. This
paper deals with the recognition of a person's current task by the surfing behavior. The surfing behavior
is represented by several attributes derived from an event log which contains user actions, such as
opening Web pages, mouse and scroll moves etc. The challenge is to select those attributes which best
describe the user behavior such that a conclusion on the user's task is possible. A pilot study was
conducted where users performed exercises, each corresponding to one of the predefined user tasks Fact
Finding, Information Gathering and Just Browsing. The particularly extensive event logs we gained
from the experiments allowed the consideration of stronger behavioral features which, moreover, formed
a novel composition of attributes. Decision trees and Naive Bayes classification were applied on this
composition, leading to accuracy values of up to 95% of correctly identified user tasks.
Keywords:
User tasks, user behavior, event logs, exploratory study, empirical research
