Скачать или смотреть Does AI Really Need Formal Language? | Sentiment Analysis on Indonesian Social Media

Can artificial intelligence understand informal language the way humans do?

In this research trailer, we explore how modern AI models interpret slang-rich social media text, focusing on sentiment analysis of Indonesian X (Twitter) posts. While text normalization is commonly used to convert informal language into standard forms, our study asks a critical question: does normalization always improve AI performance?

We conducted a controlled experiment using real Indonesian social media data and compared three preprocessing strategies: raw text without normalization, dictionary-based normalization, and AI-based normalization using a large language model. These approaches were evaluated across multiple transformer models, including IndoBERT, IndoBERT-large, and IndoBERTweet.

Surprisingly, our results show that models trained on raw, informal text often perform as well as or better than those trained on normalized text. This suggests that slang and informal expressions are not just noise, but important signals that carry emotional meaning.

Acknowledgments:
We would like to express our sincere gratitude to Mr. Henry Lucky and Mr. Rilo Chandra Pradana for their guidance and support throughout the Research Methodology in Computer Science course. We also thank Ms. Noviyanti Tri Maretta Sagala for her valuable guidance in the Text Mining course, which greatly contributed to the development of this research.

Research Team | Group 24:
1. Adrian Marc Fedier Purnama (2702247216)
2. Farren Angelica Dinzani (2702243546)