TutorQA Benchmark

This dataset is part of the benchmark introduced in the paper Graphusion: Leveraging Large Language Models for Scientific Knowledge Graph Fusion and Construction in NLP Education. We also release more data in our GitHub page. It contains 6 tasks designed for evaluating various aspects of reasoning, graph understanding, and language generation.

Dataset Structure

Each task is a separate split:

• task1: Relation Judgment
• task2: Prerequisite Prediction
• task3: Path Searching
• task4: Subgraph Completion
• task5: Clustering
• task6: Idea Hamster (no answers, open ended)

Split	Fields
task1	question, answer
task2	question, answer
task3	question, answer
task4	question, answer
task5	question, answer
task6	question

Usage Example

from datasets import load_dataset

dataset = load_dataset("li-lab/tutorqa")

# Access individual tasks
task1 = dataset["task1"]
task6 = dataset["task6"]

Citation

@inproceedings{yang2025graphusion,
  title={Graphusion: A RAG Framework for Knowledge Graph Construction with a Global Perspective},
  author={Yang, Rui and Yang, Boming and Feng, Aosong and Ouyang, Sixun and Blum, Moritz and She, Tianwei and Jiang, Yuang and Lecue, Freddy and Lu, Jinghui and Li, Irene},
  booktitle={Proceedings of the NLP4KGC Workshop at The Web Conference 2025 (WWW'25)},
  year={2025},
  url={https://arxiv.org/abs/2410.17600}
}