New Framework Uses RoBERTa to Detect Fake Review Campaigns and Customer Satisfaction Drops by Analyzing Sentiment Trends Over Time

Researchers developed a temporal sentiment aggregation framework that addresses limitations of traditional single-text sentiment analysis by tracking collective behavioral shifts

The system uses pretrained transformer-based language models, specifically RoBERTa, to extract sentiment signals from individual comments and aggregate them into time-window-level scores

Significant downward shifts in aggregated sentiment scores are flagged as potential anomalies, enabling early detection of malicious review campaigns and sudden user satisfaction declines

The approach tackles real-world challenges including inherent noise and class imbalance in short user comments for applications like customer feedback monitoring and brand reputation management