model link- https://drive.google.com/drive/folders/1FO9PemDSuue2Ty7UGW5UYKox7kCs-k3A?usp=sharing

This repository is a part of Classification of Insincere Questions(CIQ) track at FIRE 2019. We have used combinations of different word embedding techniques namely Glove, FastText, and Paragram. Our model is based on a Bidirectional LSTM-GRU network coupled with checkpoint ensembling for the fine-grained classification of insincere questions into the following categories:

1. Rhetorical Questions
2. Hate Speech
3. Sexual Content
4. Hypothetical Questions
5. Other
6. Sincere Question

The final code/model used is Full Model.py

The link to the model is: https://github.com/diptamath/CIQ-Challenge/blob/master/CIQ_Diagram.jpg

Our approach was awarded the First Prize for this competition with an overall F1 score of 67.32%

The link to the working notes for this paper can be found here: http://ceur-ws.org/Vol-2517/

The link to the paper is: http://ceur-ws.org/Vol-2517/T5-5.pdf