Sentiment Analysis and Customer Satisfaction Factors Based on LSTM and Topic Modeling

Abstract

Objective: To predict sentiment of the Airbnb text reviews using Long Short Term Memory (LSTM). To improve the accuracy and performance metrics. To identify customer satisfaction and dissatisfaction factors of the Airbnb customers using Sentiment Analysis and Topic Modeling. Method: The study is divided into two parts after performing necessary pre-processing steps. First part focuses on sentiment analysis using LSTM. Dataset is created by combining review data of 3 cities, then, operations like pre-processing, sentiment analysis, label column creation, under sampling etc. were conducted. After this, data was trained on the configured LSTM Model. The second part of the study was Topic Modeling after applying Sentiment Analysis, on an Airbnb dataset, to derive and understand customer satisfaction and dissatisfaction factors.Findings: Sentiment Analysis using LSTM showed training accuracy of 96.37%.and testing accuracy of 93.89%. The performance metrics showed promising results. The topics found for negative and positive sentiment portraying the customer satisfaction and dissatisfaction factors after Topic Modeling align with the existing literature findings and are important to generalize the existing literature as well. Novelty: Improved performance metrics like Accuracy, F1score and Recall for sentiment analysis using LSTM. Results stating customer satisfaction and dissatisfaction factors add value to the existing literature and help to generalize findings.