A Case Study in Multi-Emotion Classification via Twitter

Abstract

Social media platforms generate continuously tremendous quantities of valuable knowledge for users’ perspectives towards our global societies for example, Twitter. Sentiment analysis reveals its vital role to take the advantage of these different perspectives for different applications like, political votes, business domains, financial risks, and etc. Most traditional approaches in sentiment analysis predict a single attitude from the users’ tweets. This is not considered a quiet correct approach, due to multiple of implied feelings in the users’ tweets towards a specific topic, person, or event. This research presents hybrid machine learning approach, that can predict multiple feelings in the same tweet. It applies two methods, which are Binary relevance based on four machine learning algorithms in addition to Convolutional neural networks. The tweets preprocessed and converted into feature vectors. Word embedding, emotion lexicons, and frequency distribution probability are used to extract features from the input tweets. The paper finally presents a case study of two experiments to show the multi emotion prediction classifiers workflow on real tweets. The applied dataset is on SemEval2018 Task E-c. Binary relevance method has hamming score 0.53, and Convolutional neural network method has score 0.54.