In the context of elderly-oriented cultural tourism, understanding and enhancing the emotional experience of elderly tourists is crucial for improving tourism service quality and engagement. This paper proposes a DeepSeek-based multimodal emotion recognition algorithm tailored for elderly-oriented cultural tourism scenarios. The model integrates facial expressions, speech, and physiological signals to achieve a more accurate and comprehensive emotion recognition system. By leveraging DeepSeek’s advanced deep learning capabilities, the proposed algorithm effectively fuses multimodal data through a cross-modal attention mechanism and a hierarchical feature extraction framework. To address domain-specific challenges, such as age-related variations in emotional expression and noisy real-world environments, the model incorporates adaptive feature alignment and domain adaptation techniques. Experimental results on a specially curated elderly tourism emotion dataset demonstrate that the proposed method outperforms existing approaches in accuracy and robustness. The findings provide valuable insights into personalized tourism service recommendations, intelligent tour guide interactions, and affective computing applications in the cultural tourism industry.