Negative Comments Multi-Label Classification

Abstract

It is a known fact that on the daily basis significant amount of information is produced due to large number of people being connected to social networking sites. Online interaction is now included in our lifestyle whether it is through a tweet, a message or through commenting on each other posts on different platforms. Online interaction contributes a significant part to our society but also contains several dangerous results. This paper mainly focuses on the cons of social interaction and a way through which it can be decreased. Aggression and related activities such as trolling peoples, harassing online involves hate comments in various forms. There are numerous such cases coming in present time and sites respond by closing down their remark areas. After the introduction of Machine Learning and having data in massive amount now its quite logical to build a tool which can tackle this situation. While there are algorithmic solution for these probelm but they are slow and expensive which make us curious about new approaches and frameworks. Four models are used for the purpose of classifying the comments, however the model which performed better than other three is Bi-LSTM + Bi-GRU model with accuracy of 97.89.