Abstract
This study investigates the often-dismissed potential of emojis and hashtags in enhancing sentiment analysis with social media data. Utilising data collected from an emerging decentralised platform Mastodon, we examine the roles of these text-element
features and assess their impact on sentiment analysis performance. Employing three prevalent machine learning models (Support Vector Machine, Naive Bayes, and Random Forest), we design four experiments and compare the influence of text, hashtags, emojis, and their combination on polarity classification. The results consistently show that incorporating hashtags and emojis alongside text improves sentiment analysis accuracy. These findings emphasise the importance of recognizing the value of these text features in social media text analysis, challenging their common dismissal as inconsequential noise.
features and assess their impact on sentiment analysis performance. Employing three prevalent machine learning models (Support Vector Machine, Naive Bayes, and Random Forest), we design four experiments and compare the influence of text, hashtags, emojis, and their combination on polarity classification. The results consistently show that incorporating hashtags and emojis alongside text improves sentiment analysis accuracy. These findings emphasise the importance of recognizing the value of these text features in social media text analysis, challenging their common dismissal as inconsequential noise.
| Original language | English |
|---|---|
| Title of host publication | 2023 Annual Meeting of ASIS&T Asia Pacific Chapter |
| Publication status | Published - 17 Nov 2023 |