基於機器學習的台灣觀光酒店經營績效的分析及預測模型—電子口碑的角色

Analysis and Prediction Model of Taiwanese Tourist Hotel Performance Based on Machine Learning—the Role of Electronic Word-of-Mouth

蘇麗心(Li-Shin Su)

指導教授 ： 包曉天

國立交通大學/管理學院/管理科學系所/碩士(2019年)

摘要

本研究針對台灣觀光酒店的績效，利用非線性以及線性模型，分析重要影
響因子並建立預測模型。本研究分為兩部分，一為重要性模式，二為預測模式。重要性模式中使用了兩種方法分析，分別是機器學習演算法極限梯度提升與迴歸分析，探討台灣觀光酒店旅館績效—入住率的重要影響因子，並且比較在線性與非線性的兩個方法下結論的差異。預測模式則是針對入住率建立預測模式，並且利用建好的模型設計情境模式，模擬旅館經理人為了提升旅館績效制定策略之真實情況，提供本研究在實務上的貢獻。
本研究之重要性模式中，探討影響旅館績效的因子分為三個構面，分別為旅館特徵、區域、電子口碑，而研究中發現，入住率之重要影響因子為旅館特徵構面中的自由行旅客比例、服務品質、旅館規模以及平均房價，還有電子口碑構面中的評論數量以及對於位置評分，也就是說，隨著消費者在網路上訂房與評論愈加盛行，重要性模式之結論鼓勵旅館經理人除了考慮旅館本身的變數以外，也要將精力聚焦在管理網路上的評論與評分上，才能提升旅館績效、永續經營。
預測模式則是針對旅館入住率，建立了樣本外平均絕對誤差百分比為11.2931 %的預測模型，準確度良好，而使用者可將任意一家旅館的實際資料投入本研究模型並建立情境模式，分析在不同情況下，不同變數的變動對於績效的影響，在旅館經理人擬定經營策略時，此模式可以給予實質的資訊與建議。

關鍵字

旅館績效 ； 電子口碑 ； 預測模型 ； 情境分析 ； 機器學習 ； XGBoost ； 入住率

並列摘要

Based on the performance of Taiwan Tourist Hotel, this study uses nonlinear and linear models to analyze the important impact factors and establish a prediction model.
This study is divided into two parts, one is the importance mode and the other is the prediction mode. Two methods are used in the importance mode, the machine learning algorithm Extreme Gradient Boosting(XGBoost) and multiple regression respectively, to explore the important influence factors of Taiwan tourism hotel performance, and to compare the differences between the two methods of linear and nonlinear. The forecasting mode is to establish a forecasting model for the hotel occupancy rate, and
use the built model to design the situation model to simulate the real situation of the hotel manager in order to improve the hotel performance and provide the practical contribution of the research.The importance mode of this study found that the important factors affecting hotel
performance are the proportion of free independent travelers, service quality, hotel size and average room price in the hotel's characteristic facet, as well as the number of comments and the rating of the location in the electronic word-of-mouth facet. That is to say, as consumers make more choices and comments on the Internet, the conclusion of the importance mode encourages hotel managers to focus on managing the comments and ratings on the Internet and considering the variables of the hotel itself in order to improve hotel performance. The prediction model is for predicting the occupancy rate of the hotel, and mean absolute percentage error (MAPE) is 11.8231% when used to predict the data outside the training data set. The accuracy is very good, users can put the actual data of any hotel into the research model and establish a situational model, analyzing the impact of changes in different variables on performance under different conditions. When the hotel manager formulates the business strategy, this mode can give some substantial information and advice.

並列關鍵字

hotel performance ； occupancy rate ； EWOM ； predictive model ； situational analysis ； machine learning ； XGBoost

參考文獻

一、 英文部分

1. Assaf, A. G., Josiassen, A., Woo, L., Agbola, F. W., & Tsionas, M. (2017). Destination characteristics that drive hotel performance: A state-of-the-art global analysis. Tourism Management, 60, 270-279.

2. Caicedo-Torres, W., & Payares, F. (2016, November). A machine learning model for occupancy rates and demand forecasting in the hospitality industry. In Ibero-American Conference on Artificial Intelligence (pp. 201-211). Springer, Cham.

3. Chattopadhyay, M., & Mitra, S. K. (2019). Determinants of revenue per available room: Influential roles of average daily rate, demand, seasonality and yearly trend. International Journal of Hospitality Management, 77, 573-582.

4. Chawla, N. (2013, November). Online information monitoring for utilize hotel occupancy rate. In 2013 15th IEEE International Conference on Communication Technology (pp. 754-758). IEEE.