Abstract
Nowadays, social data analysis is widely used in different businesses. Machine learning has proved its effectiveness in big data analysis for customers’ segmentation and decision making. In the context of tourism and hospitality, there have been several studies for customers’ decision making through social data analysis. However, the use of machine learning techniques for data-driven analysis by data obtained from social networking services is still in the early stage of development. This paper aims to develop a new hybrid method using machine learning techniques for social data analysis in hotel selection. The author develops the method by the combination of feature selection, supervised learning and unsupervised learning techniques. The author uses fuzzy
k
-means for data clustering, correlation-based feature selection for feature selection, and the ANFIS (adaptive neuro-fuzzy inference system) method for predicting customers’ overall satisfaction in each cluster. The method is applied to a dataset for hotels in Saudi Arabia obtained from TripAdvisor. The prediction models of different clusters are evaluated using a set of evaluation metrics and compared with the other prediction techniques. The method was evaluated by mean-square error (MSE) and correlation coefficient (R
2
). The author compared the results of this study with other methods, which are ANFIS, support vector regression (SVR), artificial neural network (ANN) and decision trees. In this comparison, ANN was trained for 200 epochs and SVR was implemented by radial basis function kernel. The results revealed that the method that uses fuzzy
k-
means and neuro-fuzzy system (MSE = 0.0734;
R
2
= 0.9834) has better predicted overall ratings with a lower mean-squared error.