旅遊業是規模龐大且成長速度仍不斷增加的產業之一，其中旅宿業與旅遊業之間緊密相關，因此如何帶動旅宿業的績效成長是當今熱切關注的議題。隨著社交媒體和線上平台的普及，越來越多消費者在網路上表達自己的意見和情緒，因此線上評論的分析對於業者至關重要。透過線上旅館評論之情感分析能有效地量化消費者在各面向的顧客體驗和輿論，讓企業能夠更加深入地了解消費者的需求和心理，從而提升服務品質。過去研究較少針對中文旅館評論進行細粒度情感分析。因此，本研究旨在利用基於建構輔助句的Bert句對分類技術Bert-pair來對中文旅館評論進行面向級情感分析。此外，本研究進一步提出一種新的自動建構輔助句之方法Bert-pair-AA，其透過主題關鍵字技術及文本相似度等技術來自動捕捉句子之隱含意義，以提升模型之預測效能。研究結果顯示，在實驗1中，Bert-pair方法優於傳統的區分子任務方法，並可大幅提升模型預測的準確性，整體Macro-F1進步約為41.1%。而在實驗2中，本研究所設計的Bert-pair-AA方法略優於Bert-pair方法，整體Macro-F1進步約2.1%。這些結果表明採用Bert-pair方法可以提高中文旅館評論情感分析的準確性，而Bert-pair-AA方法進一步優化了該方法的效果。以上發現能為旅宿業提供更準確和實用的情感分析工具，有助於提升客戶體驗和業務效益。;The tourism industry is one of the largest and continuously growing industries, with a close connection between the accommodation sector and the overall tourism sector. Thus, driving performance growth in the accommodation industry is a pressing concern. With the popularity of social media and online platforms, an increasing number of consumers express their opinions and sentiments online, so analyzing online reviews is crucial for businesses. Utilizing sentiment analysis of online hotel reviews quantifies customer experiences and opinions, enabling enterprises to better understand consumer needs and preferences, and improve service quality. Previous studies have focused less on aspect-based sentiment analysis of Chinese hotel reviews. Therefore, this study aims to use the Bert-pair technique based on constructing auxiliary sentences to perform aspect-based sentiment analysis on Chinese hotel reviews. Additionally, a novel method called Bert-pair-AA is proposed, which automatically constructs auxiliary sentences to enhance the model’s predictive performance. This method captures the implied meanings of sentences using techniques such as topic keywords and text similarity. Experimental results indicate that in Experiment 1, the Bert-pair method outperforms traditional classification methods, significantly improving the accuracy of the model’s predictions with an overall Macro-F1 improvement of approximately 41.1%. In Experiment 2, the proposed Bert-pair-AA method slightly outperforms the Bert-pair method, achieving an overall Macro-F1 improvement of approximately 2.1%. These findings demonstrate that adopting the Bert-pair method can enhance the accuracy of sentiment analysis for Chinese hotel reviews, and the Bert-pair-AA method further optimizes its effectiveness. These findings offer more accurate sentiment analysis tools for the accommodation industry, leading to improved customer experiences and business efficiency.
