|
Abstract:
|
In this work, we propose a novel multilayer recommender system to integrate multi-source multilayer data. Specifically, we utilize tensor structure to combine multilayer information embedded by latent features shared by grouping features arising from users, items and locations. To incorporate the dependency among latent features, we introduce dependent tensor decomposition based on multilayer features from multi-source data. One major advantage is that the proposed method is able to simultaneously capture the complex dependency structure among latent features across different data sources, hence enables us to borrow multi-source information to provide more effective recommender system.
|
