<p>The increasing availability of user-generated content on review platforms and tourism websites has made sentiment analysis a vital tool for understanding customer experiences. In the hospitality and tourism sector, reviews often contain diverse opinions about specific aspects of service, such as room quality, staff behavior, and amenities. Aspect-Based Sentiment Analysis (ABSA) addresses this by identifying sentiment polarity related to distinct aspects within a review. However, challenges such as implicit aspect references, opinion ambiguity, and context-dependence hinder the effectiveness of existing models. To overcome these limitations, this study proposes a novel ABSA framework based on SpanBERT–Cross-Attention–BiGRU. SpanBERT is employed to extract aspect and opinion terms at the span level, capturing richer contextual dependencies than token-level models. A cross-attention mechanism is then applied to model fine-grained interactions between aspect and opinion spans. The fused representations are passed to a Bidirectional Gated Recurrent Unit (BiGRU) to capture sequential dependencies and predict sentiment polarity. The proposed framework is evaluated on four benchmark tourism and hospitality review datasets, achieving an average accuracy of 94.21%, precision of 92.95%, and recall of 93.54%. The proposed model is compared with baseline methods, demonstrating superior performance in terms of precision, recall, and F1-score. The results confirm that integrating span-level extraction and attention-based alignment significantly enhances the granularity and accuracy of sentiment analysis in multi-aspect hotel reviews.</p>