Defense (PhD, Masters, Pre)
PhD DefenseComputational Methods to Understand the Association between Emojis and Emotions |
|
||
Monday, March 29, 2021, 01:00pm - 03:00pm |
|||
Speaker: Abu Awal Md Shoeb
Location : Remote via Zoom
Committee:
Prof. Gerard de Melo (Advisor)
Prof. Matthew Stone
Prof. Yongfeng Zhang
Daniel Preotiuc-Pietro (Bloomberg)
Vita G. Markman (LinkedIn)
Event Type: PhD Defense
Abstract: Emojis have become ubiquitous in digital communication due to their visual appeal as well as their ability to vividly express human emotion, among other factors. They are also heavily used in customer surveys and feedback forms. Hence, there is a need for methods and resources that shed light on their meaning and communicative role. In this work, we seek to explore the connection between emojis and emotions by employing new resources and methodologies. First, we compile a unique corpus of ~20.8 million emoji-centric tweets, such that we can capture rich emoji semantics using a comparably small dataset. We then train a model to generate interpretable word-vectors and show how domain-specific emoji embedding gives better emotion prediction than other vanilla embeddings like Glove and Word2Vec. Second, we conduct annotation experiments for a set of 150 popular emojis. This gives 1,200 emoji-emotion pairs of human ratings of association concerning 8 basic human emotions such as anger, anticipation, disgust, fear, joy, sadness, surprise, and trust. This gold standard emoji-emotion score is the first of its kind. We additionally conduct experiments to assess to what extent such associations can be inferred from existing data. Our experiments show that this succeeds when high-quality word-level information is available. Lastly, we consider a set of popular NLP tools to perform additional experiments on texts containing emojis to analyze how they operate on emojis.Keywords: EmoTag, Emoji--Emotion, Emoji--Embedding, Twitter--Emoji, Emotion--Lexicons
: