Shape Your Feed: LLM Agentic System for Real-Time Conversational Recommendation
A recent study published on arXiv (2608.06632) presents Shape Your Feed (SYF), a recommendation framework utilizing LLM technology for the real-time, multimodal co-curation of content. SYF overcomes the drawbacks of conventional systems that depend on passive ranking based on implicit behavioral signals, which frequently do not align with users' explicit preferences. It comprises a three-tier structure: a Perception Flow that captures user intent through text prompts, voice commands, and UI interactions; a Serving Flow dedicated to real-time re-ranking and pruning; and a likely third tier for generation or feedback. SYF seeks to improve user agency and personalization in recommendations, marking a transition towards user-centric approaches in digital content platforms. This research contributes to the integration of large language models into recommendation systems.
Key facts
- Paper titled 'Shape Your Feed: An LLM-based Agentic System for Conversational Recommendation' on arXiv.
- arXiv ID: 2608.06632.
- Announcement type: new.
- SYF is an LLM-based agentic recommendation framework.
- Enables real-time, multimodal co-curation of content.
- Three-tier architecture: Perception Flow, Serving Flow, and a third tier.
- Perception Flow captures user intent from text, voice, and UI interactions.
- Serving Flow performs real-time agentic re-ranking and pruning of candidate items.
- Addresses gap between passive behavioral algorithms and explicit user interests.
- Traditional systems infer preferences from clicks and dwell time.
Entities
Institutions
- arXiv