An improved model for sentiment analysis on luxury hotel review

This article proposes a heuristic model for sentiment analysis on luxury hotel reviews to analyse and explore marketing insights from attitudes and emotions expressed in reviews. We make several significant contributions to visual and multimedia analytics. This research will develop the practical application of visual and multimedia analytics as the research foundation is based on information analytics, geospatial analytics, statistical analytics and data management. Large amounts of data are generated by hotel customers on the Internet, which provides a good opportunity for managers and analysts to explore the hidden information. The analysis of luxury hotels involves different types of data, including real-world scale data, high-dimensional data and geospatial data. The diversity of data increases the difficulty of processing computational visual analytics. It leads to that some classical classification methods, which cost too much time and have high requirements for hardware, are excluded. The goal is to achieve a compromise between performance and cost. An experiment of this model is operated using data extracted from Booking.com. The entire framework of this experiment includes data collection, data preprocessing, feature engineering consisting of term frequency-inverse document frequency and Doc2Vec based feature generation and feature selection, Random Forest classification, data analysis and data visualization. The whole process combines statistical analysis, review sentiment analysis and visual analysis to make full use of this dataset and gain more decision-making information to improve luxury hotels' service quality. Compared with simple sentiment analysis, this integrated analytics in social media is expected to be used in practice to gain more insights. The result shows that luxury hotels should focus on staff training, cleanness of rooms and location choice to improve customer satisfaction. The sentiment distribution shows that scores are consistent with the emotion they show in reviews. Hotels in Spain have a much better average score than hotels in the other five countries. In the experiment, the sentiment analysis model is evaluated by receiver operating characteristic and precision-recall curve. It is proved that this model performs well. Twenty most essential features have been listed for future adjustments to the model