Personalized Advertising Services through Hybrid Recommendation Systems: The Case of Digital Interactive Television
PhD Thesis Description
The research presented in this dissertation is concerned with the development of an efficient approach for the personalized delivery of advertisements in a digital interactive television environment. Technological advances in the television domain offer marketers an unprecedented opportunity to target their advertising messages to those consumers/viewers that are most likely to be interested in them, while at the same time decreasing the information overload caused to the viewers by messages that are obtrusive and irrelevant to their information needs. The main objective underlying this dissertation is to improve the predictive accuracy of current personalization approaches through the efficient management of the factors affecting their performance. To this end, marketing and consumer behavior theory and practice are employed. The collaborative filtering personalization method, based upon the nearest neighbor algorithmic implementation, is initially identified as a suitable and representative approach that fits our context. In this approach, user ratings upon observed items (advertisements) are utilized to trace behaviorally similar users, whose ratings are then aggregated and directed to appropriate recipients. We then propose the exploitation of the notion of lifestyle as a user characteristic that can be utilized to trace similar viewers independently from the availability of data, thus overcoming the most common collaborative filtering limitation. The lifestyle-based approach, combined with on-line behavioral data under a collaborative filtering reasoning, is shown to yield improved personalization accuracy. We then follow an incremental improvement process in which an initial approach (segmentation-based approach) is formulated upon a novel behavioral classification mechanism that classifies viewers into lifestyle segments. Subsequently, this approach is refined at the user-level (integrated approach) by hybridizing collaborative filtering and lifestyle-based personalization techniques. The predictive performance is then further improved by developing an item-level approach (best-item approach), which is sensitive to differences in viewer behavior upon the observed items. Finally, the proposed approaches are combined into an integrated personalization strategy that holistically addresses the major limitations of extant personalization methods. The outcomes of our research can be extended to provide insights, not only to the domain of advertising personalization, but also to other product-related personalization systems, thus contributing more widely to adaptive and recommender systems knowledge.
Dr. Georgios Lekakos
|Office phone:||+30 2108203921|
|Address:||Evelpidon 47-A & Lefkados 33 Room 907, GR-11362, Athens, Greece|
Professor Georgios I. Doukidis