This is a experimentation work to build an open domain chatbot for hindi Language.
- The original data is from Daily-Dialog Dataset
- We have preprocessed and modified the data, post that translated the data using AI4Bharat Indic Trans Model. You can find the translation script in the repo.
- Download the en-indic model
- You can find the Modified data in the /data directory.
We have compared 4 different architecture in incremental fashion.
- Sequence-Sequence(Bi-GRU)
- Sequence-Sequence(Bi-GRU) + Attention
- Sequence-Sequence(Bi-GRU) + TCN + Attention
- Sequence-sequence(Transformer)
Fastext Pretrained Hindi Embeddings were also used to see if it can contribute to enhance the model's performance.
Yet to be added.
It's a part of my Mtech research work at NIT Silchar.
The code and experiment is highly influenced by