Skip to content

Code for PAKDD 2024: Are Graph Embeddings the Panacea? – an Empirical Survey from the Data Fitness Perspective

Notifications You must be signed in to change notification settings

PascalSun/PAKDD-2024

Repository files navigation

PAKDD

This is the Code and Datasets for the paper

"Are Graph Embeddings the Panacea? – an Empirical Survey from the Data Fitness Perspective"

accepted by PAKDD 2024.

Datasets

We have used the following datasets in our experiments:

"Actor"
"AMAZON_COMPUTERS"
"AMAZON_PHOTO"
"AttributedGraphDataset_BlogCatalog"
"AttributedGraphDataset_CiteSeer"
"AttributedGraphDataset_Cora"
"AttributedGraphDataset_Flickr"
"AttributedGraphDataset_Pubmed"
"AttributedGraphDataset_Wiki"
"CitationsFull_CiteSeer"
"CitationsFull_Cora"
"CitationsFull_Cora_ML"
"CitationsFull_DBLP"
"CitationsFull_PubMed"
"CiteSeer"
"Coauther_CS"
"Coauther_Physics"
"Cora"
"Cora_Full"
"GitHub"
"HeterophilousGraphDataset_Amazon_ratings"
"HeterophilousGraphDataset_Minesweeper"
"HeterophilousGraphDataset_Questions"
"HeterophilousGraphDataset_Roman_empire"
"HeterophilousGraphDataset_Tolokers"
"PubMed"
"TWITCH_DE"
"TWITCH_EN"
"TWITCH_ES"
"TWITCH_FR"
"TWITCH_PT"
"TWITCH_RU"
"WEBKB_Cornell"
"WEBKB_Texas"
"WEBKB_Wisconsin"

Experiment Design

Results

Similarity matrices for: F1 score (left), Accuracy score (middle), Network Parameters (|E|, K, L) (right).

Similarity matrices for F1 score Similarity matrices for Accuracy score Similarity matrices for Network Parameters (|E|, K, L)

Reproduce the Results

Setup

  1. Clone the repository

  2. Setup the conda environment

    conda create -n PAKDD-481 python=3.8
    conda activate PAKDD-481
    pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118 # install pytorch
    pip install git+https://github.com/pyg-team/pytorch_geometric.git  # install pytorch geometric
    pip install pyg_lib torch_scatter==1.4.0 torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.2.0+cpu.html  
    pip install -r requirements.txt
  3. Calculate the network characteristics First start the docker container with the following command:

    docker-compose up

    Then run the following command to calculate the network characteristics:

    python3 -m graph_metrics.dataset_network_metrics
  4. Run the experiments

    bash ./scripts/all_datasets.sh

Citation

@InProceedings{10.1007/978-981-97-2253-2_32,
   author="Sun, Qiang
   and Huynh, Du Q.
   and Reynolds, Mark
   and Liu, Wei",
   editor="Yang, De-Nian
   and Xie, Xing
   and Tseng, Vincent S.
   and Pei, Jian
   and Huang, Jen-Wei
   and Lin, Jerry Chun-Wei",
   title="Are Graph Embeddings the Panacea?",
   booktitle="Advances in Knowledge Discovery and Data Mining",
   year="2024",
   publisher="Springer Nature Singapore",
   address="Singapore",
   pages="405--417",
   abstract="Graph representation learning has emerged as a machine learning go-to technique, outperforming traditional tabular view of data across many domains. Current surveys on graph representation learning predominantly have an algorithmic focus with the primary goal of explaining foundational principles and comparing performances, yet the natural and practical question ``Are graph embeddings the panacea?'' has been so far neglected. In this paper, we propose to examine graph embedding algorithms from a data fitness perspective by offering a methodical analysis that aligns network characteristics of data with appropriate embedding algorithms. The overarching objective is to provide researchers and practitioners with comprehensive and methodical investigations, enabling them to confidently answer pivotal questions confronting node classification problems: 1) Is there a potential benefit of applying graph representation learning? 2) Is structural information alone sufficient? 3) Which embedding technique would best suit my dataset? Through 1400 experiments across 35 datasets, we have evaluated four network embedding algorithms -- three popular GNN-based algorithms (GraphSage, GCN, GAE) and node2vec -- over traditional classification methods, namely SVM, KNN, and Random Forest (RF). Our results indicate that the cohesiveness of the network, the representation of relation information, and the number of classes in a classification problem play significant roles in algorithm selection.",
   isbn="978-981-97-2253-2"
}

About

Code for PAKDD 2024: Are Graph Embeddings the Panacea? – an Empirical Survey from the Data Fitness Perspective

Resources

Stars

Watchers

Forks