Representation learning on heterogenous information networks
In real world, most of the information networks are heterogeneous in nature, which contains different types of nodes and relationships. Representation learning or feature learning techniques are needed to extract features of these Heterogeneous Information Networks (HIN) and convert them to low dime...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2021
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/150188 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | In real world, most of the information networks are heterogeneous in nature, which contains different types of nodes and relationships. Representation learning or feature learning techniques are needed to extract features of these Heterogeneous Information Networks (HIN) and convert them to low dimensional vectors such that they can be used as input to machine learning models to perform machine learning tasks. Current graph embedding methods have the limitations on either considering singular types of nodes and relationships or losing important node features due to ignoring node contents or relationships between nodes. In this project, an advanced graph embedding technique, Metapath Aggregated Graph Neural Network (MAGNN), is studied. With the idea of metapath, which captures the relationships between node types, and graph neural network, a powerful graph embedding model based on deep learning, MAGNN aims to address these problems and generate node embedding with more structural and semantic information of HIN. Empirical studies with more benchmark datasets are conducted to investigate the effectiveness of MAGNN model. The results are useful for comparison with the state-of-the art baselines. |
|---|