觀光旅館訂價評估：基於AHP先驗機率的貝氏機率網路法

Pricing of Tourist Hotels in Taiwan: A Bayesian Classification Approach with AHP priors

觀光旅館是觀光產業主要的核心產業，因為旅館具有住宿、休閒、餐飲、社交、會議、娛樂等多元功能。透過市場機制，旅館的定位最終必須反映在其價格上。研究已指出房價會影響諸如觀光旅館的選擇、住房率與旅客滿意度等。因此，掌握影響觀光旅館訂價之關鍵屬性因素，以建構一套評估觀光旅館合理訂價的方法，對於觀光旅館的投資決策或營運，都有重大的參考意義。過去這方面的研究，大多透過迴歸方法來建立預測模型。本研究則採取新的途徑，透過機器學習演算法，以分類的方式來處理，並提出結合層級分析法的貝氏機率網路，以處理小樣本數據案例。將貝氏機率網路結合AHP先驗機率的分類方法應用於台灣觀光旅館的訂價預測，實證結果表明運用本法能大幅提升分類的正確率，由58.1%提升至87.1%。對於小樣本數據的案例，本研究方法能提供有效的解決方案。

Hotels have played a central role and nowadays been regarded as a basic and functional business within the tourism industry. It has been suggested that room rates affect, among others, hotel occupancy rate, customer satisfaction, and customer’s decision to select a hotel. Room rate pricing thus becomes a very important decision to make in the hotel business management. The present study attempts to develop a method that takes advantage of the existing database so that room rates may reflect hotel market position, perceived value, but are still within a reasonable and competitive price range. Hedonic pricing method has been employed by most of the previous studies to tackle this problem. A novel approach in which a machine learning technique (PNN) based on the Naïve Bayes classifier was used to predict hotel room rates. A total of 103 tourist hotels in Taiwan were used, and the PNN rendered a correct classification rate of 58.1% with the validation dataset. The correct rate increased to 87.1% when a prior probability computed with AHP was incorporated. Findings from this study suggest that this approach may assist hotel managers in adequately pricing room rates, and, in addition, it is well suited for cases whereas only small datasets are made available. Possible applications are discussed.