Court Judgement Prediction, Explanation on Indian Supreme Court Cases
- Created a corpus of 32,000 legal jugdements and tackled the open research problem of Court Judgement Prediction.
- Achieved the best acc. of 78% (using XLNet) after extensive experiments on 14 SOTA document classification methods.
- Used different explainability methods such as IG, Occlusion, Attention etc. to address explainability.
- Evaluated our explainability results using gold annotations by legal experts using ROUGE, BLEU, METEOR etc.