Emotion Recognition in Informal Text

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Final Year Thesis Oral Defense

Title: "Emotion Recognition in Informal Text"

by

HUANG Yu-ning

Abstract:

Emotion identification has long been an important and interesting field in 
Natural Language Processing (NLP). Previous works usually focused only on 
classifying emotion categories. However, when we talk in daily lives, we 
not only express different emotions such as anger, joy, fear or sadness 
but also imply the intensity of the emotion. Therefore, knowing both the 
emotion category and emotion intensity enables us to understand the bigger 
picture of how exactly the author of a piece of text felt when the words 
were being written. Another drawback of previous works is that they often 
rely on emotion lexicons which require a huge amount of experts' effort to 
annotate them at first and to maintain them constantly.

In the thesis, we present EmotionNet, a machine learning based model that 
can predict emotion category and emotion intensity simultaneously from 
tweets without the help of emotion lexicons. We also offer two versions of 
EmotionNet, namely, One-Stage EmotionNet Model and Two-Stage EmotionNet 
Model. Our result indicates that while Two-Stage EmotionNet Model gets 
higher accuracy, One-Stage EmotionNet Model is more efficient during both 
training and predicting phases.

With EmotionNet, we further examine the influence of emojis on emotion 
intensity. The result shows that emojis indeed enhance the emotion 
intensity which is aligned with the intuition. Finally, we introduce a 
word embedding dedicated to emotion identification. Trained in supervised 
learning with an emotion dataset, this emotion word embedding can be 
viewed as an emotion space which is capable of representing the emotion 
differences in words.


Date            : 3 May 2019 (Friday)

Time            : 10:30 - 11:10

Venue           : Room 5566 (near lifts 27/28), HKUST

Advisor         : Prof. YEUNG Dit-Yan

2nd Reader      : Prof. ZHANG Nevin Lianwen