Targeted Online Advertisement Based on User Trails
Recommendation System 12.2010
ABSTRACT. In this project we address the problem of finding the most suitable banner display advertisement option for a user given his/her current browsing session. Using the historical browsing session information for an advertisement campaign, we mine the association of different ad views, engagements and clicks. Using a probabilistic model, we find the likelihood of an ad to be clicked given a specific set of events that describe the user session. The major challenge in training the model for optimum precision is the sparsity of data (~0:5% Click through rate) and we propose the use of ad engagement as a success event like ad click to train the model more effectively. Our results show a good click through conversion probability on test data.

>> final report (PDF)
(c) 2011 Tak Yeon Lee