Analysis of Underlying Emotions in Textual Data Using Sentiment Analysis Which Classifies Text In to Positive, Negative or Neutral Sentiments

Analysis of underlying emotions in textual data can provide important insights and has been the focus of much research. Sentiment analysis is a popular technique which classifies text into positive, negative or neutral sentiments based on polarity and subjectivity. But this fails to extract actual emotions represented by text. This can be achieved by analyzing the occurrence of particular words, symbols and emojis in text. For emotion classification, machine-learning and deep-learning techniques Peng et al.(2022) are also used. These models classify textual data into only a small subset of the broad spectrum of emotions. Subtle variations such as "disappointment" and "grief" are lost and text is simply classified into "sadness". Extracting such subtle emotions from texts present some unique problems as one single sentence may represent multiple related emotions. Thus we have explored various Deep learning models based on convolution Neural NetworkO’Shea and Nash (2015) and Long Short Term Memory Staudemeyer and Morris (2019) for classifying textual data into one or more of 27 different fine grained emotions.