Multi-facet in heterogeneous information network representation learning

Most of the networks we encounter in practice are Heterogeneous Information Networks (HINs), where it contains nodes of different classes connected by edges. Due to its nature, HIN contains more information than Homogeneous Information Networks and are therefore more complex and cumbersome to analyze or study. The process of converting raw HIN datasets into dense matrixes of lower dimensions while still preserving the network structure as much as possible is called graph embedding or representation learning, which is the very first step to be carried out before any algorithm could be applied on the network to study its structure or node relationships. Graph embedding for HINs often faces more restriction and obstacles due to the existence of multiple node classes, and therefore demands more time and effort in constructing an ideal algorithm. In this project, we examined a multi-facet approach in carrying out graph embedding for HINs by dissecting its algorithms, testing with real world datasets like DBLP2 and movielens, and finally build a frontend project to visualize the embedding.Bachelor of Engineering (Electrical and Electronic Engineering