Harassment by cyberbullies is a significant phenomenon on the social media which is quite disturbing and troubling. It appears in various forms and is commonly seen in a textual format in most social networks.
Intelligent and advanced systems are necessary for automated detection of these incidents. Existing works for cyberbullying detection have quite a few bottlenecks.
They address just one topic of cyberbullying. Also they rely on the use of swear words. We show that deep learning based models can overcome these bottlenecks.
We plan to use publicly available datasets to efficiently train our models where each dataset addresses a different topic of cyberbullying. For example, Twitter dataset contains examples of racism. There are also datasets which contain examples of sexism. We are using Deep Neural Network (DNN) Based Models like BLSTM for training our model.
In order to capture the semantics and similarities between words, word embedding is used to model words. For the sake of word embedding, Global Vector (GloVe) was chosen because it has outperformed other models in word similarity and named entity recognition.
The users who post cyberbullying content will be reported to the helpline via mail. The system also makes an analysis of the cyberbullying posts which are visible to the user.