# Sentiment Expression via Emoticons on Social Media

**Authors:** Hao Wang, Jorge A. Castanon  
**arXiv:** [1511.02556](https://arxiv.org/abs/1511.02556)  
**Date:** 2015-11-09  
**Summary page:** https://www.summarizepaper.com/en/paper/1511.02556v1/

## Key points

- Emoticons are commonly used as features in machine learning algorithms or sentiment lexicons
- Relationship between emoticons and sentiment polarity is not always clear
- Caution should be exercised when relying on emoticons for sentiment analysis
- Emoticons are prevalent on social media, as demonstrated by the analysis of a large Twitter dataset
- Surveying participants helped identify strong and reliable signals of sentiment polarity among emoticons
- Clustering words and emoticons together provided insights into the meaning conveyed by different combinations
- Removing emoticons from microblog posts affected sentiment classification accuracy, indicating their importance in sentiment analysis
- Two machine learning models were trained to compare the impact of removing emoticons on sentiment classification performance
- Some emoticons were reliable indicators of sentiment polarity, but others conveyed complicated sentiments that required caution when interpreting them

## Summary

The paper titled "Sentiment Expression via Emoticons on Social Media" by Hao Wang and Jorge A. Castanon explores the use of emoticons in sentiment analysis and other natural language processing (NLP) tasks. While emoticons like :) and :( are commonly used as features in machine learning algorithms or sentiment lexicons, the authors argue that the relationship between emoticons and sentiment polarity is not always clear. Therefore, caution should be exercised when relying on emoticons for sentiment analysis. To demonstrate the prevalence of emoticons on social media, the authors analyzed the frequency of emoticons in a large Twitter dataset. They then conducted four analyses to examine the relationship between emoticons and sentiment polarity, as well as the contexts in which they are used. The first analysis involved surveying a group of participants to determine their perceived sentiment polarity of the most frequent emoticons. This helped identify which emoticons were strong and reliable signals of sentiment polarity. The second analysis focused on clustering words and emoticons together to better understand the meaning conveyed by different combinations. This provided insights into how specific emotions are expressed through certain word-emoticon pairs. In the third analysis, the authors compared the sentiment polarity of microblog posts before and after removing emoticons from the text. This allowed them to assess whether removing emoticons affected sentiment classification accuracy. Finally, they tested the hypothesis that removing emoticons from text would negatively impact sentiment classification by training two machine learning models—one with and one without emoticons in the text—and comparing their performance. The results confirmed that while some emoticons were reliable indicators of sentiment polarity, a large group conveyed complicated sentiments that required caution when interpreting them. Overall, this paper highlights that although emoticons can provide valuable signals for sentiment analysis on social media, careful consideration is necessary when determining which ones to rely on due to their varying degrees of clarity in expressing sentiment polarity.