GCN_for_Node_Classification_Task

Goal

Leverage GCN for node classification tasks on datasets Cora, Citeseer and PubMed.

Datasets

Dataset	#Nodes	#Edges	#Features	#Classes
Cora	2,708	10,556	1,433	7
Citeseer	3,327	9,104	3,703	6
PubMed	19,717	88,648	500	3

Data Source

https://github.com/kimiyoung/planetoid/tree/master/data

Datasets for Cora, Citeseet, and PubMed are available in the directory Cora, Citeseet, and PubMed, in a preprocessed format stored as numpy/scipy files.

• allx, the feature vectors of both labeled and unlabeled training instances (a superset of x),

• ally, the labels for instances in allx,

• x, the feature vectors of the labeled training instances,

• y, the one-hot labels of the labeled training instances,

• tx, the feature vectors of the test instances,

• ty, the one-hot labels of the test instances,

• test.index, the indices of test instances in graph, for the inductive setting,

• graph, a dict in the format {index: [index_of_neighbor_nodes]}, where the neighbor nodes are organized as a list. The current version only supports binary graphs.

Datasets Introduction

• Cora
  The Cora dataset consists of machine learning papers, which are classified into the following seven categories:
  • Case Based
  • Genetic Algorithms
  • Neural Networks
  • Probabilistic Methods
  • Reinforcement Learning
  • Rule Learning
  • Theory

The papers were selected in such a way that each paper in the final corpus cited or was cited by at least one other paper. There are 2708 papers in the entire corpus. After stemming and removing stop words, we were left with only 1433 unique words in size. All words with document frequency less than 10 were removed.

• Citeseer
  The Citeseer dataset is a selection of papers from the CiteSeer digital papers library, classified into the following six categories:
  • Agents
  • AI
  • DB
  • IR
  • ML
  • HCI

The papers were selected in such a way that each paper in the final corpus cited or was cited by at least one other paper. There are 3327 papers in the entire corpus.

• PubMed
  The PubMed dataset includes 19,717 scientific publications on diabetes from the Pubmed database, divided into three categories:
  • Diabetes Mellitus, Experimental
  • Diabetes Mellitus Type 1
  • Diabetes Mellitus Type 2

The citation network consists of 44338 links. Each publication in the dataset is described by a TF/IDF-weighted word vector from a dictionary of 500 unique words. TF-IDF (term frequency–inverse document frequency) is a commonly used weighting technique for information retrieval and data mining. TF is Term Frequency, and IDF is Inverse Document Frequency. TF-IDF is a statistical method for evaluating the importance of a word to a document set or a document in a corpus. The importance of a word increases proportionally to the number of times it appears in the document, but decreases inversely proportional to the frequency it appears in the corpus.

GCN Model Framework

Environment

python == 3.8.16

cuda == 11.3

pytorch == 1.12.1

scikit-learn == 1.2.2

matplotlib == 3.7.1

Usage

python gcn_cora.py
python gcn_citeseer.py
python gcn_pubmed.py

Note: The Parameter folder holds the GCN model training parameters that yield the results of the model runs. Files GCNCora.pkl, GCNCiteseer.pkl, GCNPubMed.pkl correspond to the datasets Cora, Citeseer, and PubMed, respectively. And see file gcn.ipynb for the detailed implementation.

Result

Measure model node classification using classification metrics:

Dataset	Accuracy	F1_score
Cora	0.8120	0.8051
Citeseer	0.6930	0.6670
PubMed	0.7940	0.7893

Reference

• Zhilin Yang, William Cohen, Ruslan Salakhudinov. Proceedings of Machine Learning Research. Proceedings of The 33rd International Conference on Machine Learning, PMLR 48:40-48, 2016.

• Graph Convolutional Networks.

• PyG: Learning Methods on Graphs.