Multi-Category Classification Confidence for Ad Contextualization

This project studies Machine Learning algorithms for multi-category document classification. The purpose is to effectively predict user’s behavior based on the contextualization of the advertising and the associated document and therefore, to increase the click rate and the success of a dynamical advertising campaign. Due to the nature of the World Wide Web, the feature sets for the classification is extremely large. However, many learning algorithms don’t perform well with large number of features or attributes. Feature selection technique is necessary to avoid overfitting and provide faster and more cost-effective learning models. The focus of our study will be on the process and methods of feature generation and selection. We will also study the effectiveness of different learning algorithms such as the Naïve Bayes classification, Support Vector Machines and Neural Networks in this particular application area. The work will directly contribute to the business of the industry partner and also generate research achievement academically.

Hasham Burhani
Faculty Supervisor: 
Dr. Wenying Feng