Relation-Aware Language-Graph Transformer for Question Answering

Authors: Jinyoung Park, Hyeong Kyu Choi, Juyeon Ko, Hyeonjin Park, Ji-Hoon Kim, Jisu Jeong, Kyungmin Kim, Hyunwoo Kim

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate the effectiveness of QAT on commonsense question answering datasets like Commonsense QA and Open Book QA, and on a medical question answering dataset, Med QA-USMLE. On all the datasets, our method achieves state-of-the-art performance.
Researcher Affiliation Collaboration 1Korea University 2 NAVER 3 NAVER Cloud 4 NAVER AI Lab
Pseudocode No The paper describes its proposed methods using mathematical formulations and textual explanations but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes Our code is available at http://github.com/mlvlab/QAT.
Open Datasets Yes We evaluate our method on three question-answering datasets: Commonsense QA (Talmor et al. 2019), Open Book QA (Mihaylov et al. 2018), and Med QA-USMLE (Jin et al. 2021).
Dataset Splits Yes Since the official test set labels are not publicly available, we mainly report performance on the in-house development (IHdev) and test (IHtest) sets following (Lin et al. 2019).; we experiment on the official data split from (Mihaylov and Frank 2018).; We use the same data split as (Jin et al. 2021).
Hardware Specification No The main paper does not explicitly describe the hardware used for experiments. It mentions 'The details on the backbone LMs and technical details are in the supplement' but this is not provided in the main text.
Software Dependencies No The paper mentions using specific Language Models like RoBERTa and SapBERT, but it does not specify version numbers for these LMs or other software dependencies (e.g., Python, PyTorch, TensorFlow versions) in the main text.
Experiment Setup No The paper states, 'The details on the backbone LMs and technical details are in the supplement.' However, no specific hyperparameters, training configurations, or system-level settings are provided in the main body of the paper.