Bidirectional LSTM Model in Predicting Satisfaction Level of Passengers on Airline Service

Abstract

With the advancement of aerospace technology, flying has become one of the most common and popular modes of travel in the world. Passengers' opinions and feedback on U.S. aerospace services are of great reference value for improving flight services. Based on over 120 thousand rows of data with more than 20 indicators related to passenger satisfaction in recent years, this paper employs multiple machine learning models such as Logistic Regression, Random Forest, Adaboost, Xgboost, Catboost, etc. to classify customers' satisfaction level. It is found that Random Forest classifier reaches the highest AUC value of 99.3%, and selects out several important indicators that significantly affect passenger satisfaction, including Online Boarding, Inflight Wifi Service, Inflight Entertainment and Seat Comfort. In addition, this experiment applies Bidirectional LSTM model to classify three types of emotion based on over 10 thousand tweets, the accuracy of which on test set attains 91.27%. Pivotal reasons for negative feedback are extracted from these tweets such as flight service quality, flight cancellation, flight delay, inconvenient booking and loss and damage of luggage, in which aerospace companies should make targeted improvements. So Bidirectional LSTM model could reliably identify emotional tendency of tweets hence guiding aerospace companies to timely adjust their service items and better satisfy customers' demand.