HOSSemEval-EB23: a robust dataset for aspect-based sentiment analysis of hospitality reviews

Aspect-Based Sentiment Analysis (ABSA), also known as fine-grained opinion mining, is a crucial task focused on understanding the sentiment expressed in a text with respect to specific aspects. This paper tackles the challenge of aspect-based sentiment analysis in the hospitality domain by proposing HOSSemEval-EB23, a novel dataset designed to address several limitations in existing resources. The new dataset not only fills the gaps found in other datasets but also broadens the range of domains within aspect-based sentiment analysis. The effectiveness of HOSSemEval-EB23 is evaluated by applying state-of-the-art aspect-based sentiment analysis models, such as variants of TAS Transformers and generative models like T5. Experimental results show that the proposed method achieves the best performance with TAS- \(BERT_{\tiny MEDIUM}\) $BER{T}_{\mathit{MEDIUM}}$ , reaching an F1-score of 79.74. Moreover, HOSSemEval-EB23 shows superior performance in capturing sentiment towards different aspects within sub-sentences of reviews compared to existing datasets. Consequently, HOSSemEval-EB23 represents a unique and valuable resource for researchers and practitioners aiming to improve the accuracy and scope of aspect-based sentiment analysis in this field.