Loading...
Paper Number
1284
Paper Type
Completed
Description
Social media content has been widely used for financial forecasting and sentiment analysis. However, emojis as a new “lingua franca” on social media are often omitted during standard data pre-processing processes, we thus speculate that they may carry additional useful information. In this research, we study the effect of emojis in facilitating financial sentiment analysis and explore the most effective way to handle them during model training. Experiments are conducted on two datasets from stock and crypto markets. Various machine learning models, deep learning models, and the state-of-the-art GPT-based model are used, and we compare their performances across different emoji encodings. Results show a consistent increase in model performances when emojis are converted to their descriptive phrases, and significant enhancements after refining the descriptive terms of the most important emojis before fitting them into the models. Our research shows that emojis are a valuable source for better understanding financial social media texts that cannot be omitted.
Recommended Citation
Chen, Siyi and Xing, Frank, "Understanding Emojis for Financial Sentiment Analysis" (2023). ICIS 2023 Proceedings. 3.
https://aisel.aisnet.org/icis2023/socmedia_digcollab/socmedia_digcollab/3
Understanding Emojis for Financial Sentiment Analysis
Social media content has been widely used for financial forecasting and sentiment analysis. However, emojis as a new “lingua franca” on social media are often omitted during standard data pre-processing processes, we thus speculate that they may carry additional useful information. In this research, we study the effect of emojis in facilitating financial sentiment analysis and explore the most effective way to handle them during model training. Experiments are conducted on two datasets from stock and crypto markets. Various machine learning models, deep learning models, and the state-of-the-art GPT-based model are used, and we compare their performances across different emoji encodings. Results show a consistent increase in model performances when emojis are converted to their descriptive phrases, and significant enhancements after refining the descriptive terms of the most important emojis before fitting them into the models. Our research shows that emojis are a valuable source for better understanding financial social media texts that cannot be omitted.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
15-SocialMedia