An abstractive text summarization model based on the seq2seq transformer architecture, that beats PreTrained BERT Summarizer with respect to Rogue Scores.
It uses the CNN_DailyMail Dataset for training.
Pre-Trained BERT
| **** | Rogue-1 | Rogue-2 | Rogue-L |
|---|---|---|---|
| r | 0.253672 | 0.088063 | 0.230165 |
| p | 0.398178 | 0.142948 | 0.361394 |
| f | 0.297954 | 0.103405 | 0.27031 |
Our Seq2Seq Transformer
| **** | Rogue-1 | Rogue-2 | Rogue-L |
|---|---|---|---|
| r | 0.3818 | 0.161194 | 0.362263 |
| p | 0.378692 | 0.17094 | 0.359935 |
| f | 0.370184 | 0.160745 | 0.351657 |