LOGIN

Implementation of LOGIN: Large Language Model-Graph Oriented Interaction for Node Classification (WSDM 2025).

Setup

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Place graph data, node prompts, LM embeddings, and LLM outputs under data/ (see data/README.md), or point to an existing tree:

export LOGIN_DATA_ROOT=/path/to/your/assets

Training

Run from the repository root:

python3 -m core.train_lcgnn_with_ucscore dataset cora gnn.model.name GCN gnn.train.feature_type TA seed 0

Common options (YACS opts):

Option	Default	Description
dataset	cora	Dataset name
gnn.model.name	GCN	GNN backbone
gnn.train.feature_type	TA	Node features (ogb, TA, llm, …)
gnn.train.pl_rate	0.1	Fraction of train nodes for LLM pseudo-labels
gnn.train.dropout	0.6	Dropout (MC ensemble uses fixed ratio)
llm.name	vicuna	LLM for consultation
seed	random	Random seed

Layout

LOGIN/
├── core/                 # training code
│   ├── train_lcgnn_with_ucscore.py
│   ├── pipelines/
│   ├── GNNs/
│   ├── LLMs/
│   └── data_utils/
├── data/                 # assets (not shipped; see data/README.md)
└── requirements.txt

Citation

@inproceedings{qiao2025login,
  title={Login: A large language model consulted graph neural network training framework},
  author={Qiao, Yiran and Ao, Xiang and Liu, Yang and Xu, Jiarong and Sun, Xiaoqian and He, Qing},
  booktitle={Proceedings of the Eighteenth ACM International Conference on Web Search and Data Mining},
  pages={232--241},
  year={2025}
}