Authors:Hongwei Wang, Fuzheng Zhang, Jialin Wang, Miao Zhao, WenjieLi,
Xing Xie, Minyi Guo
Venue : CIKM 2018
Presenter : JIARUI CHEN (201700301042)
Outline
• Introduction
• Motivation
• Method
• Dataset
• Experiment
• Conclusion
2
Introduction
· Recommendation System
· Information overload.
· Brother of Search Engine.
· Categories.
3
Introduction
· Knowledge Graph
· Heterogeneous network.
· Node -> Entity,Edge -> Relation.
· Example.
4
Introduction
· Knowledge Graph based Recommendation
· Heterogeneous network.
· Node -> Entity,Edge -> Relation.
· Example.
5
Introduction
· Knowledge Graph based Recommendation
· Heterogeneous network.
· Node -> Entity,Edge -> Relation.
Knowledge Graph Layer
User Layer
6
Motivation
· Existing Works
· Embedding-based methods :
· Use knowledge graph embedding to incorporates the learned entity
embeddings into a recommendation framework.
· These models more suitable for a graph tasks (e.g. link prediction) than
recommendation.
· Path-based methods :
· Explore the various pattern of connections among items in KG to provide additional guidance
for recommendations, e.g.through Meta-path.
· Rely heavily on manually designed meta-paths,which is hard to optimize in practice.
7
Motivation
· Key Idea : Preference propagation as ripples on the water
· Combine above two categories together.
· Use the items in the ripple sets as side information.
· Example.
8