MoraleTrack: A Dual-Phase Transformer-Based Framework for Sentiment-Aware Team Morale Forecasting in Agile Project Environments

Abstract

Effective monitoring of team morale is crucial for maintaining productivity and preventing burnout in agile project environments, yet current project management tools lack robust mechanisms for capturing emotional dynamics from routine communication. This paper presents MoraleTrack, a novel dual-phase transformer-based framework for sentimentaware team morale forecasting. The framework first performs fine-grained sentiment classification at the message level using a BERT-based encoder enhanced with role-aware embeddings and, then models temporal sentiment trends via a transformer architecture to predict team morale trajectories. Evaluated on realistic synthetic datasets, MoraleTrack demonstrated superior performance over the baseline models in terms of classification accuracy ($85.4 \% \mathrm{~F} 1$-score), robustness to class imbalance (0.92 F1 for minority classes), and forecasting precision (0.076 RMSE). Ablation studies validate the contributions of each component, particularly role embeddings and temporal attention mechanisms, while efficiency analyses confirm practical deployability. MoraleTrack bridges affective computing and agile management with a scalable solution for passive morale monitoring, thereby enabling data-driven interventions through real-time analysis.

Ähnliche Arbeiten