Graph-COM/Text-Attributed-Graphs

1---2license: apache-2.03---4 5## Overview6This dataset covers the encoder embeddings and prediction results of LLMs of paper 'Model Generalization on Text Attribute Graphs: Principles with Lagre Language Models', Haoyu Wang, Shikun Liu, Rongzhe Wei, Pan Li.7 8## Dataset Description9 10The dataset structure should be organized as follows:11 12```plaintext13/dataset/14│── [dataset_name]/15│   │── processed_data.pt    # Contains labels and graph information16│   │── [encoder]_x.pt       # Features extracted by different encoders17│   │── categories.csv       # label name raw texts18│   │── raw_texts.pt       # raw text of each node19```20 21### File Descriptions22- **`processed_data.pt`**: A PyTorch file storing the processed dataset, including graph structure and node labels. Note that in heterophilic datasets, thie is named as [Dataset].pt, where Dataset could be Cornell, etc, and should be opened with DGL.23- **`[encoder]_x.pt`**: Feature matrices extracted using different encoders, where `[encoder]` represents the encoder name.24- **`categories.csv`**: raw label names.25- **`raw_texts.pt`**: raw node texts. Note that in heterophilic datasets, this is named as [Dataset].csv, where Dataset can be Cornell, etc.26 27### Dataset Naming Convention28`[dataset_name]` should be one of the following:29- `cora`30- `citeseer`31- `pubmed`32- `bookhis`33- `bookchild`34- `sportsfit`35- `wikics`36- `cornell`37- `texas`38- `wisconsin`39- `washington`40 41### Encoder Naming Convention42`[encoder]` can be one of the following:43- `sbert` (the sentence-bert encoder)44- `roberta` (the Roberta encoder)45- `llmicl_primary` (the vanilla LLM2Vec)46- `llmicl_class_aware` (the task-adaptive encoder)47- `llmgpt_text-embedding-3-large` (the embedding api text-embedding-3-large by openai)48 49 50## Results Description51 52The ./results/ folder consists of prediction results of GPT-4o in node text classification and GPT-4o-mini in homophily ratio prediction.53 54```plaintext55./results/nc_[DATASET]/4o/llm_baseline       # node text prediction56./results/nc_[DATASET]/4o_mini/agenth        # homophily ratio prediction57```58 59## Reference60If you find the data useful, please consider citing our paper:61 62```63@inproceedings{wang2025generalization,64  title={Generalization Principles for Inference over Text-Attributed Graphs with Large Language Models},65  author={Wang, Haoyu and Liu, Shikun and Wei, Rongzhe and Li, Pan},66  booktitle={Forty-second International Conference on Machine Learning},67  year={2025}68}69```70