隨著Web2.0的普及，線上評論已成為消費者購買決策的重要依據。本研究針對旅宿業線上評論文本與評分不一致的現象進行探討，此問題可能對消費者選擇、業者和第三方平台信譽產生負面影響。文獻回顧顯示飯店星級(HSC)和評論者貢獻度(RC)可能存在交互作用，即評論評分並非受到單一因素影響，而是二個變數共同作用的非簡單線性關係。研究使用網頁爬蟲從 TripAdvisor 收集飯店評論，並通過比較不同詞嵌入方法和迴歸模型來探討在交互作用下找出解決評論不可靠之最有效的模型組合。 研究結果顯示，評論者貢獻度和飯店星級之間的交互作用對於評論評分有顯著影響，尤其在城市旅遊類型的資料集中。這種交互作用顯示不同星級的飯店對評論者期望和評價標準的影響不同，導致評分差異。在實驗一中，比較不同詞嵌入方法，Longformer 在處理長文本評論的特徵提取上表現優越，證明了基於 Transformer 的模型在處理評論文本方面的優勢。實驗二中的結果顯示，比較一系列迴歸模型後，Longformer-LSTM 最能夠有效預測評論評分。 綜合上述，本研究在理論上提供了新方法 Longformer-LSTM，用於解決飯店評論的不可靠問題，在實務上提供業者改善評論系統的策略，特別是較為依賴客戶評論來吸引新顧客的旅遊業者，可透過提高評論的可靠性，提升消費者信任度，間接影響其購買決策。 ;With the proliferation of Web 2.0, online reviews have become a crucial part of consumer purchasing decisions. This study investigates the issue of inconsistency between the text and ratings of online reviews in the hospitality industry, a problem that can negatively affect consumer choices, business reputations, and the credibility of third-party platforms. A literature review indicates that the interaction between hotel star classification (HSC) and reviewer contributions (RC) may affect the ratings, suggesting that the influence on ratings is not simply due to one factor but a complex interaction between these variables. The study collected hotel reviews from TripAdvisor using a web scraping system and compared different word embedding methods and regression models to identify the most effective combination for addressing the unreliability of reviews under the influence of these interactions.
The results show that the interaction between RC and HSC significantly impacts ratings, especially in the datasets related to city tourism. This interaction demonstrates that different levels of hotel stars may influence reviewers′ expectations and standards differently, leading to variations in ratings. In Experiment One, Longformer excelled in feature extraction for long text reviews, proving the superiority of Transformer-based models in processing review texts. Experiment Two′s findings indicate that the Longformer-LSTM combination is the most effective at predicting review ratings after comparing a series of regression models.
In summary, this study provides a novel Longformer-LSTM method theoretically for addressing the unreliability of hotel reviews and offers practical strategies for businesses to improve their review systems. This is particularly beneficial for tourism operators who rely heavily on customer reviews to attract new clients. By enhancing the reliability of reviews, businesses can significantly increase consumer trust, indirectly influencing their purchasing decisions.
