Machine Learning-Based Prediction Models for Sentiment Analysis on Online Customer Reviews: A Case Study on Airbnb

In the last decade, with the rise of sharing economy, in particular Airbnb, customers are not merely buyers but also actively share their thoughts and experiences toward goods and services.Sentiment analysis, a sophisticated technological approach, has emerged as a pivotal tool to extract people's opinions as well as sentiments from written language.On the other hand, assessing the price of a listing has always been a daunting task for hosts and guests.While numerous pricing models for Airbnb have been proposed, achieving precise accuracy remains a challenge.As a result, this paper aims to investigate whether incorporating the sentiment scores derived from online customer reviews could improve the accuracy of Airbnb price prediction or not.First, online customer reviews on Airbnb are examined using natural language processing techniques to seek the guest sentiment and its association with listings prices.Once sentiment scores are calculated, they are used as an additional attribute to forecast Airbnb listings price.Several machine learning models are employed, including Linear Regression, Ridge Regression, Support Vector Machine, XG-Boost and Random Forest.The experimental results show that the inclusion of sentiment scores slightly decreases model performance in the case of three Asian economies (Hong Kong, Japan and Taiwan).Overall, Random Forest without sentiment variable is the best-performing model among five models for Airbnb price prediction.