Emotion Detection from Text Using ML Framework

With technology improving at astonishing strides, most people have access to Internet for communication via text, image, audio and video. Emotions are expressed through words, gestures, expressions and in today's world through tweets and comments. With the applications tending to be more user personalised this data needs to be processed in real time. Their behaviour needs to be analysed by their emotions. The dataset is sourced from Kaggle where the data has been streamed from X (formerly known as Twitter). This data is first converted into a vectorised form by the TF-IDF algorithm and then supervised classification techniques such as such as K-means, Naive Bayes, SVM and Ensemble Learning Techniques like Gradient Boosting, Random Forest, Decision Tree, XG Boost, Cat Boost and deep learning network named Multi-Layer Perceptron is used to determine and correctly classify the data into one of the 6 labels: sad, joy, love, anger, fear and surprise. The data balancing technique SMOTE is applied to enhance the performance of classifiers. MLP classifier was able to achieve the best results followed by the tree-based classifiers Random Forest, Decision Tree, and SVM with 0.99 F-measure.