DKG-MTI: A Dual Knowledge Graph Framework for User Intent Inference from Travel Reviews
A groundbreaking framework, named DKG-MTI, has been introduced by researchers to enhance the understanding of user intent in online travel reviews. This model surpasses current hierarchical and retrieval methods by using an inference-only strategy to improve knowledge acquisition. It constructs a personalized User-Specific Intent Knowledge Graph for each review, which is synchronized with a Global Hotel Knowledge Graph through advanced semantic techniques. When tested on TripAdvisor reviews, DKG-MTI achieved superior performance in classification and intent prediction compared to established benchmarks. The findings are available for review on arXiv, reference number 2608.06752.
Key facts
- DKG-MTI is a dual knowledge graph framework for unified multi-task user intent inference.
- It dynamically constructs a User-Specific Intent Knowledge Graph from each review.
- It aligns the user-specific graph with a Global Hotel Knowledge Graph via structure-aware semantic smoothing.
- The aligned knowledge is combined with the original review and processed by a large language model.
- The model simultaneously predicts aspect ratings and generates reverse user intent statements.
- Experiments were conducted on TripAdvisor reviews.
- DKG-MTI outperforms strong LLM and retrieval-based baselines in classification and intent generation.
- The paper is available on arXiv (2608.06752).
Entities
Institutions
- arXiv
- TripAdvisor