Sentiment Analysis via Large Language Models for Real-Time Employee Engagement Monitoring: Opportunities and Ethical Considerations

ABSTRACT

Employee disengagement poses significant organizational costs, yet traditional assessment methods remain reactive and infrequent. This study investigates the application of large language models (LLMs) for real-time sentiment analysis of employee communications to enable proactive engagement monitoring. We fine-tuned GPT-3.5 and LLaMA-2 variants on 10,000 anonymized, crowdsourced workplace communication samples, evaluating performance against baseline sentiment classifiers. Results demonstrated 85.3% accuracy (F1-score: 0.83) in detecting disengagement signals, with false positive rates below 8%. The models identified burnout indicators 20% faster than lexicon-based approaches and exhibited 15% improvement in multilingual equity metrics. Findings suggest LLMs can augment HR analytics through nuanced emotional tone detection in asynchronous communications. However, deployment necessitates robust ethical frameworks addressing privacy preservation, algorithmic transparency, and consent mechanisms. This research contributes methodological insights for integrating LLMs into human capital management while establishing guardrails against surveillance overreach, proposing federated learning architectures and opt-in monitoring protocols as practical safeguards.

Keywords: sentiment analysis, large language models, employee engagement, workplace analytics, organizational ethics