Link Prediction in Bipartite Graphs: A Kernel-based Approach for Recommendation
DescriptionRecommender system is widely used in e-commerce websites to provide personalized suggestions to consumers and promote sales. Previously, the collaborative filtering (CF) paradigm has successfully employed transaction information for recommendation. With the use of transaction information, the recommendation problem can be transformed to a link prediction problem in a consumer-product bi-partite graph, where consumers and products are connected by transactions. This graph has been employed in several heuristics algorithms and led to improve recommendation performances. However, there have been limited learning-based models utilized the consumer-product graph. In this research, I propose to develop a kernel-based framework that inspects the associative interaction graphs (AIGs) of consumer-product pairs to predict the possibilities that interactions may occur in the pairs. I plan to develop graph kernels that can effectively capture the information in AIGs for this framework to build recommendation models. I will test the effectiveness of this framework using multiple real-world datasets.
|Effective start/end date||1/04/10 → 16/10/12|