Integration of BERTopic and IndoBERTweet for Aspect-Based Sentiment Analysis (ABSA) on Short Text Data: A Case Study of Responses to Government Policies in 2025

Abstract

The implementation of various government policies in 2025 has triggered massive public opinion on social media platform X; however, traditional sentiment analysis often fails to provide details on specific topics, necessitating an Aspect-Based Sentiment Analysis (ABSA) approach. This research integrates the BERTopic model for aspect extraction and IndoBERTweet for sentiment classification to address the challenges associated with the characteristics of short and unstructured text. By preserving the data without a stemming process to maintain semantic context integrity, the BERTopic model demonstrates optimal performance with a Coherence score (C_v) of 0.7539 and a Topic Diversity of 0.9285. The synergy between BERTopic and IndoBERTweet proves effective in generating coherent topic representations and accurate sentiment classification for informal language on social media. Consequently, this integration provides a more profound and superior solution for mapping public responses to the dynamics of government policy.