基於機器學習的台灣民宿訂價分析與預測模型－電子口碑的角色

A Pricing Analysis and Prediction Model for Taiwanese Bed and Breakfast Based On Machine Learning-The Role of Electronic Word-of-Mouth

蘇致瑩(Zhi-Ying Su) ； 陳俊智(Chun-Chih Chen) ； 包曉天(Hsiao-Tien Pao)

《觀光與休閒管理期刊》 8卷1期 (2020/06) Pp. 37-51

https://doi.org/10.6510/JTLM.202006_8(1).0004

摘要

政府南向政策帶動觀光產業及特色區域的民宿業蓬勃發展，線上訂房逐漸成為旅客訂房的主要管道。研究透過網路平台Booking.com，蒐集2018年宜蘭縣169家民宿之電子口碑及房型特徵資料，採用複迴歸模式及極限梯度提升（XGBoost）機器學習演算法，分析民宿訂價的重要影響變數，其模型配適能力分別達17.54%及0.43%的MAPE。XGBoost之結果顯示民宿訂價的重要影響變數依序為：是否具有景觀、房間大小、電子口碑之位置便利性及住宿感受度。XGBoost建構的民宿訂價預測模型，樣本外預測能力達22.82%之合理的MAPE。本研究建構的模型結果可供業者於投資民宿時，進行資源投放順序決策，以及擬定經營電子口碑之策略，進而提升民宿競爭力及促進我國觀光產業發展。

關鍵字

民宿訂價 ； 機器學習 ； XGBoos ； 電子口碑 ； 預測模型 ； 特徵重要性分析

並列摘要

The government's policy drives the new southbound tourism industry and domestic tourism subsidies to promote the booming tourism industry, and online booking has gradually become the main channel for passengers to make reservations. Through the online platform Booking.com, this paper collects the electronic word-of-mouth and characteristics of 169 Bed and Breakfast in Yilan County in 2018, and uses the multiple regression model and the extreme gradient boosting (XGBoost) machine learning algorithm to analyze the important influence factors of the Bed and Breakfast pricing. The model matching ability is 17.54% and 0.43% of MAPE, respectively. The results of XGBoost show that the important influence factors of the pricing are: whether it is the landscape room, the room size, the location convenience of electronic word-of-mouth and the accommodation feeling. The pricing prediction model constructed by XGBoost has a reasonable MAPE with an out-of-sample prediction capability of 22.82%. The model constructed in this study can be used by the industry to predict the price of the Bed and Breakfast and to develop a strategy of actively operating electronic word-of-mouth, thereby enhancing the competitiveness of the Bed and Breakfast and promoting the development of the tourism industry in Taiwan.

並列關鍵字

Bed and Breakfast ； Electronic Word-of-Mouth (EWOM) ； Prediction Model ； Feature Importance ； Machine Learning ； XGBoost