Robust Hotel Booking Dialogues with Hybrid Intent and DeepSeek-V3

The hotel booking industry is rapidly digitizing, yet traditional solutions often fail to meet users’ diverse needs. We present an intelligent hotel booking dialogue system based on DeepSeek-V3, integrating hybrid intent recognition, hierarchical dialogue management, and a fallback mechanism with template-based responses. A semantic similarity model (Paraphrase-MiniLM-L6-v2) strengthens API fallback and retrieval robustness. Experiments on a cleaned MultiWOZ 2.4 hotel-domain subset show 95.0% intent recognition, 92.0% slot-filling accuracy, and a 90.0% dialogue success rate—reducing human workload by 30–40%. Our framework is scalable and adaptable for the hospitality industry.