← Back to the feed
AI & softwareResearch findings

HARP: Training-Free Agentic Communication for LLM-Based Recommendation

Original published 2026.11.04 03:24 KSTAdded to Blackdeck 2026.11.05 04:30 KST

What has changed?

Both users and items generate self-descriptions, enabling symmetric communication and resolving information asymmetry.

Key points

  1. Reformulates LLM-based recommendation as an agentic communication process with four stages: build, transmit, understand, and sync.
  2. Introduces symmetric self-descriptions for both users and items, enabling balanced information exchange.
  3. Achieves up to 24.0% improvement in Hit@1 over the strongest baseline across four public datasets without training.

Connect it to your work

The user profile is no longer frozen after warmup; item profiles are now dynamically constructed via rule extraction.

What remains unproven

This research note is based on an abstract; check the original for test conditions and application scope.

Originals used for this note

Links open the original publisher in a new tab.

  1. arxivHARP: Training-Free Dual-Profile Agentic Communication for LLM-Based Recommendation