Sentiment Analysis of AI-Driven Job Automation in Indonesia: Integrating LSTM, BERTopic, and Time-Series Forecasting for Job Displacement Risk Prediction

Abstract

The rapid advancement of artificial intelligence (AI) has intensified public concerns regarding job automation and potential workforce displacement. While previous studies often examine sentiment, topic modeling, or temporal trends separately, this study integrates sentiment classification, topic modeling, and time-series forecasting to provide a comprehensive analysis of public perceptions in Indonesia. From an initial collection of 1.026 tweets, 966 validated Indonesian-language tweets from Twitter/X collected between 2024 and 2026 were retained for analysis. Sentiment labeling was conducted using a weakly supervised lexicon-based approach, followed by classification with a Long Short-Term Memory (LSTM) model. The model achieved 72% accuracy with a weighted F1-score of 0.72 and demonstrated strong performance in detecting negative sentiment, achieving a recall of 0.83. BERTopic modeling revealed dominant concerns related to job replacement, automation risks, vulnerable professions, and skill adaptation in the digital era. Furthermore, ARIMA-based time-series forecasting indicates that negative sentiment trends are likely to persist in the near future. This study contributes methodologically by integrating deep learning, topic modeling, and temporal forecasting within a unified framework to better understand evolving public risk perceptions of AI-driven job automation in Indonesia.

Ähnliche Arbeiten