Thesis

Code for Conversation modeling in offensive language detection. There are 4 types of models:

1. stl_regression: is a simple BERT-base regression model for offensiveness degree prediction
2. 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