Table 4 Comparaison of Distil-BERT-MRC performance with State-of-the-Art Methods.
Year | Ref | Model | Dataset | Exact match | F1 score | Key observations |
|---|---|---|---|---|---|---|
2022 | DistilBERT | Covid data | 80.6 | 87.3 | Achieved high accuracy on domain-specific data with efficiency. | |
2023 | BERT | SQuAD 2.0 | 72.34 | 60.45 | Larger model; resource-intensive and slower inference. | |
2023 | RoBERTa | MultiRC | 68.78 | 72.12 | Demonstrates strong contextual understanding in multi-sentence reasoning tasks. | |
2024 | XLNet | NewsQA | 65.14 | 70.89 | Excels in paragraph-level QA but require significant computational resources. | |
2024 | GPT-3 | TriviaQA | 85.23 | 88.67 | Outstanding performance demands high computational power. | |
2024 | Parallel-DistilBERT | SQuAD 1.1 | 86.20 | 88.36 | Enhanced exact match and F1-score with 31% reduced parameters. | |
2025 | Few-shot-Neural-MRC | Chinese comprehension corpus | 86.42 | 74.65 | Prepared Chinese MRC dataset and set the baseline for further MRC research studies. | |
- | Proposed | Distil-BERT-MRC | WikiQA | 90.23 | 91.42 | Achieved state-of-the-art performance with minimal computational cost. |
