This project is focused on generating headlines for Nepali news using a well-trained deep neural network model.
Transformer model based on encoder-decoder architecture was used for generating the headlines. It was trained
on Nepali language datasets collected from different online news portals. The model was validated on a test
dataset with rogue metrics. For training data, the average ROUGE scores were 11.4, 3.65, 11.4 for rogue-1,
rogue-2, and rogue-L respectively. For the testing set, the corresponding ROUGE scores were 7.85, 1.47, and
7.39 respectively.
The model was based on this paper Attention is all you need.
Python 3
Tensorflow
pip install -r requirements.txt
Access this drive for model.
- Achyut Burlakoti
- Sijal Baral
- Subodh Baral
- Tapendra Pandey
