LLM-based agentic system enhances conversational recommendations

Researchers have developed Shape Your Feed (SYF), an LLM-based agentic system designed to enhance conversational recommendation by allowing users to co-curate content in real-time. Unlike traditional passive ranking systems, SYF captures nuanced user intent through multimodal inputs like text prompts and voice commands. The system's Serving Flow re-ranks candidate items based on a persistent Semantic Profile of user preferences, while a Self-Evolution Flow uses Direct Preference Optimization and an LLM-as-a-Judge ensemble to align with human feedback. SYF has demonstrated high accuracy in alignment scoring and improved feed relevance and user sentiment in large-scale online experiments. AI

IMPACT This system could enable more interactive and personalized content discovery, moving beyond passive recommendation models.

RANK_REASON The item is a research paper detailing a new LLM-based system for conversational recommendations. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• Direct Preference Optimization
• Hugging Face
• LLM
• LLM-as-a-Judge
• Perception Flow
• Self-Evolution Flow
• semantic profiles
• Serving Flow
• Shape Your Feed

• product
• paper

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →