R2-D2: A Modular Baseline for Open-Domain Question Answering
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 system's 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.
PDF Abstract Findings (EMNLP) 2021 PDF Findings (EMNLP) 2021 AbstractCode
Datasets
Natural Questions
TriviaQA
Results from the Paper
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Benchmark |
|---|---|---|---|---|---|---|
| Open-Domain Question Answering | Natural Questions | R2-D2 \w HN-DPR | Exact Match | 55.9 | # 2 | |
| Question Answering | Natural Questions | R2-D2 (full) | EM | 55.9 | # 5 | |
| Passage Retrieval | Natural Questions | DPR+ELECTRA-large-extreader-reranker | Precision@20 | 85.26 | # 2 | |
| Precision@100 | 88.25 | # 5 | ||||
| Passage Retrieval | Natural Questions | DPR+RoBERTa-base-crossencoder-reranker | Precision@20 | 84.46 | # 4 | |
| Precision@100 | 88.03 | # 6 | ||||
| Question Answering | Natural Questions (long) | R2-D2 \w HN-DPR | EM | 55.9 | # 3 |
