Title:	Heterogeneous Information Network Transformers
DNr:	Berzelius-2022-93
Project Type:	LiU Berzelius
Principal Investigator:	Ahmed Emad <aesy@kth.se>
Affiliation:	Kungliga Tekniska högskolan
Duration:	2022-04-19 – 2022-11-01
Classification:	10299
Homepage:	http://rais-itn.eu
Keywords:

Abstract

This work focus on graph representation learning on Heterogeneous Information Networks(HINs). Heterogeneous networks are graphs of nodes and edges that connect between the nodes; where there are more than one type of nodes in the graph; For example, an academic graph can be nodes that represent authors, papers, and venues with edges such as author-paper that indicates authorship relation. Learning dense vector representations of these nodes capture all the semantic knowledge in the graph which is highly important for tasks such as recommender systems and node classification. The goal of this work is to devise Transformer-based learning paradigm on graph. Being based on Transformers encoders, the model requires heavy processing that I don't have. Having resources is essential to completing my research work for my doctoral studies.