Heterogeneous Graph Representation for Recommendation

Chuan Shi,Xiao Wang,Philip S. Yu
DOI: https://doi.org/10.1007/978-981-16-6166-2_7
2021-01-01
Abstract:With the rapid development of web services, various kinds of useful auxiliary data (a.k.a., side information) become available in recommender systems. To characterize these complex and heterogeneous auxiliary data, heterogeneous graph (HG) representation methods have been widely adopted due to the flexibility in modeling data heterogeneity. In this chapter, we introduce three HG representation based recommendation systems solving the unique challenges existing in diverse real-world scenarios, including Top-N recommendation (MCRec), cold-start recommendation (MetaHIN), and bibliographic recommendation (ASI). In the field of HG representation for recommendation, methods mainly contain three key components: HG constructions, HG representation learning and recommendation based on the HG representation.
What problem does this paper attempt to address?