Math Word Problem solving on MathQA (test)
81.5AccuracyExpression Tree Decoding Strategy (Ours)
Evaluation Results
| Method | Links | |
|---|---|---|
| Expression Tree Decoding Strategy (Ours)Architecture Paradigm=Expression-level decoder, Layer Sharing=None2023.10 | 81.5 | |
| Expression Tree Decoding Strategy (Ours)Architecture Paradigm=Expression-level decoder, Layer Sharing=Layer Shared2023.10 | 81.1 | |
| Multi-viewPre-trained=true2022.10 | 80.6 | |
| ElasticArchitecture Paradigm=Seq2Exp, pre-processing=different method and operators2023.10 | 80.3 | |
| Ana-CLArchitecture Paradigm=Seq2Seq / Tree2023.10 | 79.6 | |
| M-ViewArchitecture Paradigm=Seq2Exp, reproduction_mode=standard dataset without data Augmentation2023.10 | 79.5 | |
| MWP-NASArchitecture Paradigm=Seq2Exp, reproduced=true2023.10 | 79.2 | |
| RE-DeductionParadigm=I-RE, Pre-trained=true2022.10 | 78.6 | |
| RE-ExtArchitecture Paradigm=Seq2Exp2023.10 | 78.6 | |
| Textual-CLArchitecture Paradigm=Seq2Seq / Tree, reproduced=true2023.10 | 78 | |
| mBERTParadigm=Seq2Seq, Pre-trained=true2022.10 | 77.1 | |
| mBERTArchitecture Paradigm=Seq2Seq / Tree2023.10 | 77.1 | |
| MWP-RoBERTaPre-training architecture=Encoder pre-train, Pre-training=MWP pre-training and fine-tuning2021.07 | 76.6 | |
| M-TreeArchitecture Paradigm=Seq2Exp, reproduced=true2023.10 | 76.5 | |
| BERT-CLPre-training architecture=Seq2Seq pre-train2021.07 | 76.3 | |
| CL-PrototypeParadigm=CL-Gen, Pre-trained=true2022.10 | 76.3 | |
| PrototypeArchitecture Paradigm=Seq2Seq / Tree2023.10 | 76.3 | |
| MWP-BERTPre-training architecture=Encoder pre-train, Pre-training=MWP pre-training and fine-tuning2021.07 | 76.2 | |
| RoBERTaPre-training architecture=Encoder pre-train, Pre-training=None2021.07 | 75.3 | |
| BERTPre-training architecture=Encoder pre-train, Pre-training=None2021.07 | 75.1 | |
| BERT-TreeParadigm=Structure-Gen, Pre-trained=true2022.10 | 73.8 | |
| BERT-TArchitecture Paradigm=Seq2Seq / Tree2023.10 | 73.8 | |
| E-pointerArchitecture Paradigm=Seq2Exp, reproduced=true2023.10 | 73.5 | |
| T-DisArchitecture Paradigm=Seq2Seq / Tree, reproduced=true2023.10 | 73.1 | |
| Graph2Tree2021.07 | 72 | |
| Graph2TreeParadigm=Structure-Gen, Pre-trained=false2022.10 | 72 | |
| G2TArchitecture Paradigm=Seq2Seq / Tree2023.10 | 72 | |
| GTS2021.07 | 71.3 | |
| GTSParadigm=Structure-Gen, Pre-trained=false2022.10 | 71.3 | |
| GTSArchitecture Paradigm=Seq2Seq / Tree2023.10 | 71.3 | |
| GroupAttnParadigm=Seq2Seq, Pre-trained=false2022.10 | 70.4 | |
| GroupAttnArchitecture Paradigm=Seq2Seq / Tree2023.10 | 70.4 | |
| Self-ConsistencyArchitecture Paradigm=LLM, reproduced=true2023.10 | 50.7 | |
| gpt-3.5-turboArchitecture Paradigm=LLM, reproduced=true2023.10 | 42.6 |