Coherence 값을 활용한 감성 분석 기반 토픽모델링 비교 연구: Airbnb 리뷰 데이터를 활용하여

A Comparative Study of Sentiment Analysis-Based Topic Modeling Using Coherence Values: Leveraging Airbnb Reviewdata

Customer reviews play a crucial role in enhancing service quality and satisfaction on shared accommodation platforms such as Airbnb. Many studies employ sentiment analysis to classify reviews and detect customer sentiments, while also applying topic modeling to uncover the main themes and issues embedded in the textual data. Prior studies have performed emotion classification by applying threshold-based scoring schemes, with model effectiveness commonly assessed using standard evaluation metrics such as accuracy, precision, and recall. However, such methods are not well-suited for unlabeled data. To address this, we propose using Coherence scores as an alternative evaluation metric. This study classified sentiment using four sentiment analysis models (VADER, BERT, RoBERTa, and LLaMA2) and applied Topic Modeling (LDA), evaluating the results using Coherence scores. Our findings indicate that BERT achieved the highest Coherence scores for both positive and negative reviews. In addition, we observed significant differences between the sentiment analysis models for negative reviews. This research suggests the potential of Coherence scores as a novel method for evaluating sentiment analysis models when combined with topic modeling on unlabeled, real-world review data, leading to the formation of more consistent topic structures.