| 摘要: | 隨著網際網路與數位平台的快速發展，電子口碑(Electronic word-of-mouth, eWOM)已成為消費者選擇餐廳時的關鍵參考依據，過去的研究亦廣泛證實eWOM對消費者行為及產品銷售具有顯著影響。其中，評論評分作為一項直觀且易於理解的指標，不僅有助於消費者快速掌握餐廳的整體評價以做出消費決策，亦能協助餐飲業者追蹤eWOM的變化趨勢。既有的餐廳評論評分預測研究主要聚焦於食物、服務、價格與氛圍等傳統用餐體驗主題對於評論評分的影響，然而隨著「外帶」、「外送」、「得來速」與「移動餐飲設備」等新興用餐模式的興起，這些主題在不同用餐模式下的重要性與影響力可能產生變化，進而增加消費者評論評分行為的複雜性，但相關文獻尚未深入探討此議題。為填補此研究缺口，本研究結合OpenAI的GPT-4o mini模型對餐廳線上評論文本進行用餐模式分類與基於面向之情感分析，並結合機器學習技術建立各類用餐模式的評論評分預測模型，以系統性分析不同用餐模式與用餐體驗主題對評論評分的影響。研究結果顯示，XGBoost於「內用」、「外帶」、「外送」、「得來速」中的表現最為穩定且優異；隨機森林在「移動餐飲設備」中的預測能力最佳，顯示模型選擇應根據具體用餐模式進行調整。另一方面，37項用餐體驗主題自變數的重要性亦因用餐模式而異，「食物味道」於多數情境中皆最具影響力，但在「得來速」模式中，「員工服務態度」的重要性更為顯著，顯示影響評論評分的關鍵因素會隨用餐模式而發生變化。;With the rapid advancement of the internet and digital platforms, electronic word-of-mouth (eWOM) has become a critical reference point for consumers when selecting restaurants. Prior research has widely confirmed the significant impact of eWOM on consumer behavior and product sales. Among various eWOM indicators, review ratings serve as an intuitive and easily interpretable metric, enabling consumers to assess overall restaurant quality and assisting busi-nesses in tracking changes in public perception. Existing studies on restaurant review rating prediction have primarily focused on traditional dining experience themes such as food, service, price, and atmosphere. However, the emergence of new dining modes—including takeout, de-livery, drive-thru, and mobile units—has introduced greater complexity, as the importance of these themes may vary across different consumption contexts. Despite this shift, the influence of dining mode has received limited attention in the current literature. To address this research gap, this study applies OpenAI’s GPT-4o mini model to classify dining modes from online restaurant reviews and to conduct aspect-based sentiment analysis. The sentiment data were then com-bined with machine learning techniques to develop review rating prediction models across five dining modes. Experimental results show that XGBoost consistently outperformed other models in dine-in, takeout, delivery, and drive-thru contexts, while Random Forest achieved the best performance in mobile unit scenarios, suggesting that model selection should be tailored to spe-cific dining modes. Additionally, the relative importance of 37 dining experience aspects varied by context. While "food taste" remained the most influential factor in most modes, "staff service attitude" was more prominent in the drive-thru setting, indicating that key factors influencing review ratings shift depending on the dining mode. |
