Temporal Sentiment Trajectory Analysis of Customer Support Escalations on Twitter Using Recurrent Neural Networks and Attention Mechanisms

• Abey Litty

  Texas University

  Author

The proliferation of social media as a customer service channel has created an urgent need for automated systems capable of monitoring and predicting sentiment dynamics during support interactions. While existing sentiment analysis approaches effectively classify isolated messages, they fail to capture the temporal evolution of customer emotions across multi-turn conversations—a critical capability for early escalation detection and proactive intervention. This study addresses this gap by developing a hybrid deep learning framework that integrates Bidirectional Long Short-Term Memory networks with attention mechanisms to model sentiment trajectories in customer support dialogues on Twitter. Using a corpus of 6,500 customer service conversations extracted from the @AmazonHelp Twitter account, the proposed framework achieves an accuracy of 89.4% in predicting sentiment trajectory patterns and demonstrates a 12.3% improvement in early escalation detection lead time compared to baseline static classification models. The attention mechanism enhances model interpretability by identifying key linguistic markers associated with sentiment deterioration across conversation turns. This research contributes a replicable framework for temporal sentiment analysis in social media customer service contexts, with practical implications for real-time monitoring systems and proactive customer engagement strategies.

• pdf

Copyright (c) 2026 Abey Litty (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.