OFFENSIVE LANGUAGE DETECTION USING DEEP LEARNING

Abstract

Since 2004, the use of social media has been developing exponentially. There is a very huge amount of users who are using social media daily. Anyone can use social media and there are number of users that spread hate speech, abusive content and offensive language to assault minorities or some individual user on platform like Twitter. Offensive language has become a major issue nowadays for users who use social media and reading online articles on the internet. There is lots of abusive content or derogatory terms on the internet because of which users have suffer from mental illness, cyber bullying, morale down etc. which impact on the society is very dangerous. This research focuses on automatically detecting the offensive content using deep learning technique from given dataset. We have detected the offensive content using Multilayer Long Short-term Memory model. Basically Long Short Term Memory (LSTM) model is an artificial intelligence Recurrent Neural Network (RNN) which is used in deep learning. The dataset we are using is Twitter tweets which contains Hate Speech, Offensive language and Neither both, this type of content is available from which we detect the offensive language text. First, we refined the tweets from dataset using in preprocessing section and then the refined tweets were entered in the first layer of LSTM model as inputs. After getting the model trained, the output was obtained, and this output is act as input for next layer LSTM and similarly this process was going till final layer of LSTM and the final output comes, from which the offensive content was predicted in the next step and the model’s accuracy was also evaluated. The comparison between single layer LSTM and Multilayer LSTM have been evaluated to shown which method is better in terms of accuracy to detect the offensive language content. The comparison results have been shown through as table and graphs in results section.