A0951
Title: Preference matrix completion with multiple network views based on graph neural networks
Authors: Yipeng Zhuang - The Education University of Hong Kong (Hong Kong)
Chenlu Wang - The Education University of Hong Kong (Hong Kong)
Philip Yu - The Education University of Hong Kong (Hong Kong) [presenting]
Abstract: In our digital age, people are connected to many networks. Their preferences (such as ratings and rankings) for all the items (such as movies and products) may not be complete and be affected by some of these networks. To predict the missing preferences, a novel graph neural network model will be proposed for preference matrix completion with multiple network views and side information. An attention mechanism was applied to measure the influences from multi-view networks. New objective functions are proposed to evaluate the preference prediction performance. Finally, the proposed model will be applied to some real-world movie recommendation datasets. Empirical results demonstrate that this model significantly improves preference prediction over other existing models.
