In this project we try to leverage natural language procesing to help victims of sexual harrasment in the city of Maastricht. This project is a part of my master course work and was done along with 4 other students and professor as our guide. We aim to create a medium through which people could share their misfortunate events in a non judgemental environment. The idea is to create a chatbot that can understand user input related to the event and classify what type of event the user underwent and reply with appropriate information. The bot will not be self contained meaning it can only understand certain thing related to events since the training data was limited.
The training data for positive harrasement cases were kinly provided to use by these folks. They are a NGO operating in India and provide help for women in their community. For negative class, random unrelated text were used from websites like IMDB and good reads.
The pipeline has two different classifications, one for identifying if the input is a case or not which is done by a SVM classifier. The second classifier makes uses of deep learning, specifically BERT architecture which is hosted on top of a chatbot layer. For the original implementation the chatbot was written using the telegram API. The structure for the dialog is shown in the image below.
Further details can be reffered in depth from the report in this repository.
Happy to say the team won the best paper award in ECMLPKDD data science workshop. The chatbot is also being hosted and can be used with this link.