Data Annotation and Multi-Emotion Classification for Social Media Text

Abstract

In recent years, sentiment or emotion analysis has become a key research area due to its vast potential applications in getting insights from social media comments, marketing, political science, psychology, human-computer interaction, and artificial intelligence. Emotion analysis deals with identifying the emotions in any given data such as text, speech, or image. The current work proposes to identify and associate social media text to multiple emotions with varying degrees. The data collection and annotation process employed in the proposed work is a combination of manual and semi-supervised annotation method where each tweet is mapped to a six dimensional emotion vector. Totally six human emotions such as happy, sad, anger, disgust, surprise, and fear are considered for emotion-tagging. Word mover‘s distance (WMD) based on twitter word embeddings (word2vec) is proposed to develop a labelled dataset in the current work. A set of classifiers is developed on the labelled dataset to identify emotions at the tweet-level in any given text data. In the current work, KNN, tree-based, and neural network classifiers are developed.