Code for Conversation modeling in offensive language detection. There are 4 types of models:
- stl_regression: is a simple BERT-base regression model for offensiveness degree prediction
- Conversation based models, where the previous 'n' comments in the conversation are modeled using an attention mechanism. It has following variants:
a. Attention at word level
b. Attention at sentence level
c. Hierarchical attention model
d. An experimental model where subsequent comments are taken as a part of the conversation