Artificial Intelligence-Driven Hybrid Task Allocation Using NLP and Machine Learning for Next-Gen Client Support Systems

Abstract

Efficient task allocation between human agents and AI systems is critical for optimizing customer support operations. This research presents an advanced hybrid classification framework that accurately determines the optimal assignment of support tasks to humans or AI by integrating rich semantic embeddings with meticulously engineered textual features. Using a publicly available customer support tweets dataset, texts were preprocessed and embedded with Sentence-BERT to capture semantic context. Sophisticated yet computationally efficient features such as precise measurements of text length, interrogative complexity, and uppercase character distribution were incorporated to enhance predictive power and model robustness. A synthetic labeling approach simulated human and AI assignments based on task complexity heuristics. An ensemble model combining Random Forest and tuned XGBoost classifiers was trained on these features. Evaluation on a held-out test set achieved 86% accuracy, balanced precision and recall (F1 scores of 0.86 and 0.85), an ROC AUC of 0.92, and a Precision-Recall AUC of 0.89. These results demonstrate the effectiveness of combining semantic and engineered features for hybrid task allocation in customer support.

Ähnliche Arbeiten