Skip to content

Latest commit

 

History

History
84 lines (60 loc) · 3.03 KB

README.md

File metadata and controls

84 lines (60 loc) · 3.03 KB

GNN Node Regression

Comparative Analysis of Graph Neural Networks for Node Regression on Wiki-Squirrel dataset (bachelor's Research Project).

Project Overview

Bachelor's Research Project focused on node regression within graph structures.

Dataset & Data preprocessing

Wikipedia article networks dataset named “Wiki-Squirrel dataset” containing three different topics “Chameleon”, “Crocodile”, and “Squirrel”. Conducted different data preprocessing tasks including normalization, outlier handling and one-hot encoding to ensure data quality and consistency.

GNN Models

Implemented and fine-tuned four distinct Graph Neural Network (GNN) models—GCN, GAT, GATv2, and GraphSAGE—exploring diverse architectural approaches for node regression tasks. Additionally, designed and applied four unique loss functions—MSE, RMSE, MAE, and MAPE—to compare their effectiveness.

Analysis

Executed a comparative analysis to evaluate and compare the performance of GNN models and loss functions across multiple datasets.

Project Information

  • Supervisor: Prof. Mostafa H. Chehreghani
  • University: Amirkabir University of Technology (Tehran Polytechnic)
  • Semester: Spring 2023

Contact

Amirmehdi Zarrinnezhad - [email protected]

Project Link: https://github.com/zamirmehdi/GNN-Node-Regression

(back to top)