MWP-T5: Fine-tuning with Numeracy for Math Word Problem Generation

AI & ML interests

MWP generation, language modelling, representation learning

MWP-T5: Fine-tuning with Numeracy for Math Word Problem Generation

This repository contains the code for the paper Fine-tuning with Numeracy for Math Word Problem Generation [PDF] published at IEEE T4E 2023.

You can access our source code here. If you would like to cite our work, please use the following BiBTex entry:


Performance

MWP-T5 achieves state-of-the-art performance on the MAWPS and PEN datasets. The following tables show the performance of MWP-T5 and other models on the MAWPS and PEN datasets. For more information, please refer to the paper.

MAWPS

Model Name BLEU-4 ROUGE-L METEOR
seq2seq-rnn 0.153 0.362 0.175
seq2seq-rnn + GLoVe 0.592 0.705 0.412
seq2seq-tf 0.554 0.663 0.387
GPT 0.368 0.538 0.294
GPT-pre 0.504 0.664 0.391
GPT2-mwp2eq 0.596 0.715 0.427
MWP-T5 0.885 0.930 0.930

PEN

Model Name BLEU-4 ROUGE-L METEOR
MWP-T5 0.669 0.768 0.772

Contact

Please feel free to contact us by emailing us to report any issues or suggestions, or if you have any further questions.

Contact: - Yashi Chawla

You can also contact the other maintainers listed below.