Detail výsledku
R2-D2: A Modular Baseline for Open-Domain Question Answering
Dočekal Martin, Ing., UPGM (FIT)
Ondřej Karel, Ing., UPGM (FIT)
Smrž Pavel, doc. RNDr., Ph.D., UPGM (FIT)
This work presents a novel four-stage open-domain QA pipeline R2-D2 (Rank twice, reaD twice). The pipeline is composed of a retriever, passage reranker, extractive reader, generative reader and a mechanism that aggregates the final prediction from all systems components. We demonstrate its strength across three open-domain QA datasets: NaturalQuestions, TriviaQA and EfficientQA, surpassing state-of-the-art on the first two. Our analysis demonstrates that: (i) combining extractive and generative reader yields absolute improvements up to 5 exact match and it is at least twice as effective as the posterior averaging ensemble of the same models with different parameters, (ii) the extractive reader with fewer parameters can match the performance of the generative reader on extractive QA datasets.
question answering, QA, ODQA, ensemble modeling, retrieval corpora
@inproceedings{BUT175855,
author="Martin {Fajčík} and Martin {Dočekal} and Karel {Ondřej} and Pavel {Smrž}",
title="R2-D2: A Modular Baseline for Open-Domain Question Answering",
booktitle="Findings of the Association for Computational Linguistics: EMNLP 2021",
year="2021",
series="Findings of the Association for Computational Linguistics",
pages="854--870",
publisher="Association for Computational Linguistics",
address="Punta Cana",
doi="10.18653/v1/2021.findings-emnlp.73",
isbn="978-1-955917-10-0",
url="https://aclanthology.org/2021.findings-emnlp.73.pdf"
}