HARP: Training-Free Dual-Profile Agentic Communication for LLM-Based Recommendation
arXiv cs.AI · 4 Nov 2026, 04:24 AEST
HARP introduces a training-free approach to LLM-based recommendation by addressing communication challenges in user and item profiles. The method enhances recommendation performance significantly across multiple benchmarks.
Key facts
- In LLM-based recommendation, memory acts as a user-level agent in a memory-augmented backbone.
- We recast LLM-based recommendation as an agentic-communication process in four stages: BUILD a self-description, TRANSMIT it along the collaborative graph, UNDERSTAND the other side, and SYNC it from the next interaction.
- On four public benchmarks (Yelp, Books, MovieTV, Goodreads), HARP lifts Hit@1 over the strongest published baseline by 6.9%-24.0%, with family-wise Holm-corrected p < 1e-9 on every dataset.
Read aloud
HARP introduces a training-free approach to L L M-based recommendation by addressing communication challenges in user and item profiles. The method enhances recommendation performance significantly across multiple benchmarks. In L L M-based recommendation, memory acts as a user-level agent in a memory-augmented backbone. We recast L L M-based recommendation as an agentic-communication process in four stages: BUILD a self-description, TRANSMIT it along the collaborative graph, UNDERSTAND the other side, and SYNC it from the next interaction. On four public benchmarks Yelp, Books, MovieTV, Goodreads, HARP lifts Hit@one over the strongest published baseline by six point nine percent-twenty four point zero percent, with family-wise Holm-corrected p < 1e-nine on every dataset.
Sources in this cluster
- HARP: Training-Free Dual-Profile Agentic Communication for LLM-Based Recommendation
arXiv cs.AI · 4 Nov 2026, 04:24 AEST