LLM Agent Generates World Knowledge for POI Recommendation
A recent research publication presents AWARE (Agent-based World knowledge Augmented REcommendation), a novel generative model for point-of-interest recommendations. This model utilizes a large language model (LLM) agent to generate narratives that are sensitive to both location and time. These narratives reflect local cultural traits, seasonal patterns, and current events pertinent to individual users. By grounding external knowledge in tailored spatial-temporal frameworks, AWARE avoids the pitfalls of generic or irrelevant information. It overcomes the limitations of current LLM-driven recommendation systems, which are static post-training and fail to adapt to changing real-world dynamics. Comprehensive testing on three real-world datasets validates the model's effectiveness.
Key facts
- AWARE uses an LLM agent to generate contextual narratives for POI recommendation
- Narratives capture regional cultural characteristics, seasonal trends, and ongoing events
- Knowledge is anchored in each user's behavioral context
- Addresses fixed knowledge limitation of existing LLM-based recommenders
- Tested on three real-world datasets
- Paper published on arXiv with ID 2605.11807
- Generative POI recommendation formulated as sequence generation task
Entities
Institutions
- arXiv