Tourist expression heterogeneity in multimodal sentiment

Multimodal sentiment analysis has captivated substantial interest in tourism and hospitality. However, extant research overlooks that travelers exhibit diverse preferences in sentiment expression across image, text, and rating modalities. To surmount the limitation, we innovatively present an indicator-based multimodal interactive fusion network. It extracts indicators from online reviews to precisely gauge heterogeneity in sentiment expression. Moreover, we formulate an indicator-based attention mechanism that dynamically assigns weights to modalities in sentiment prediction according to the relative contributions of different modalities to sentiment expression. Additionally, we explore sentiment expression differences across tourism scenarios by integrating travel type heterogeneity and destination category heterogeneity. Experimental findings show that the proposed model outperforms state-of-the-art methods, offering deeper insights into multimodal sentiment analysis in tourism.