Prediction based on social media dataset using CNN-LSTM to classify the accurate Aggression level
Rahul Sai Ganesh, T Devi.
Abstract
The content generated by user available on social media is huge in size where the aggression in content need to be identified. The issue of aggression has to be detected when it comes to social media. Expression of aggression can either be done directly, overtly or in hidden manner. Social media content is mostly non-aggressive when considered about their nature. The proposed system works on the methods in deep learning such as CNN (Convolutional Neural Network) as well as LSTM (Long Short Term Memory Network). The first method CNN works based on further subdivided layers which are five in number including the input as well as output layer. Along with the above mentioned classifiers, ensemble method based on voting is utilized. The comments from Facebook are being used for training the model which are considered as comments from in domain. Also posts related to other social media are also being used for training that belong to cross domain. F1-score weightage for the posts related to Facebook are found to be 0.702 compared to the 0.603 for the posts related to other social media.