Multiclass Trip Type Prediction from Online Reviews Using Various Document Representations

This paper proposes a methodology for predicting people's travel purposes (trip types) from online reviews, which is essential for businesses to understand customer needs. It addresses a multiclass classification problem with five trip type labels: Business, Couple, Family, Friend, and Solo. Textual reviews are represented using frequency count, tf-idf, doc2vec, and LDA. Various Machine Learning models, including AdaBoost, Decision Tree, Logistic Regression, Random Forest, and Support Vector Machine (SVM), are employed for prediction. A 10-fold cross-validation selects optimal model parameters. The dataset comprises customer reviews from TripAdvisor for 20 hotels in Bali. Both undersampling and oversampling techniques balance the labels (trip types). The best performance is achieved using tfidf representation and SVM, with a test F1-score of 70.17. The comparison indicates that tf-idf outperforms other representations, while Random Forest and SVM models generally perform best. Results suggest grouping trip types into Solo, Business, and Couple-Family-Friends categories. Additionally, prediction models imply that different customer segments express experiences differently in reviews, and it is predictable to some extent.