Empowering Airline Route Decisions with LLM-Generated Pseudo-labels and Zero-Shot Review Prediction

Abstract

The airline industry has suffered a severe impact due to the COVID-19 pandemic. It resulted in significant financial losses. Strategic route planning is now an urgent need to mitigate the ongoing crisis. Motivated by the importance of customer sentiment in informing airline route decisions, this paper presents EAGLE (Enhancing Airline Groundtruth Labels and rEview rating prediction), a novel two-stage framework that leverages the power of Large Language Models (LLMs) to address the limitations of current works, which often rely on manual labeling and traditional machine learning models. In the first phase, EAGLE introduces a pseudo-labeling approach using LLMs to automatically label customer reviews to reduce the need for manual annotation and mitigate potential biases that exist in human labeling. The second phase employs a zero-shot LLM-based text classification method to predict customer sentiment and preferences from online reviews to provide a more accurate and context-aware analysis of customer feedback. Through extensive experiments, we demonstrate the effectiveness and robustness of EAGLE to demonstrate its superior performance compared to existing techniques. The proposed framework empowers airline companies to make data-driven decisions about route expansions, considering customer preferences and sentiments. Our contribution fibs in enhancing the objectivity of sentiment analysis and providing a comprehensive and scalable solution for airline route planning in the post-pandemic era, eventually leading to improved customer satisfaction and optimized operations.