We compare and evaluate the performance of the following sequence to seuqence models on hindi text:
- Seq2Seq (RNN baseline)
- Seq2Seq + Attention
- Seq2Seq + fasttext embedding
- Seq2Seq + Attention + fasttext embedding
- BART_T5
The details of the files and the folders are as follows:
- NLP_project_hindi_text_summarization.ipynb : Code for baseline implementation and attention visualization
- NLP_BARTT5.ipynb : Code for transfer learning and BART_T5 implementation
- attention_visualizations: Folder containing attention visualization for few samples
- models: Folder containing pretrained models that can be directly used for evaluation