New LLM approach predicts user satisfaction with AI search overviews

Researchers have developed STaRS, a new LLM-based approach for predicting user satisfaction with AI overviews in search results. Unlike previous methods that focused primarily on response quality, STaRS incorporates user behavior signals into its prediction process. The system uses a structured reasoning procedure where individual behavioral signals are interpreted to inform satisfaction predictions, with reinforcement learning further optimizing this process. Evaluations on real-world datasets show that STaRS outperforms existing methods, offering search providers a practical tool for estimating user satisfaction. AI

IMPACT This research offers a more nuanced way to evaluate AI search features by incorporating user behavior, potentially leading to better-designed and more user-friendly AI overviews.

RANK_REASON The cluster describes a new research paper detailing a novel LLM-based approach for a specific AI application. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.IR (Information Retrieval) →

• AI Overviews
• alphaXiv
• arXiv
• CatalyzeX
• CORE Recommender
• DagsHub
• Gotit.pub
• Hugging Face
• information retrieval
• reinforcement learning
• ScienceCast

• paper
• product

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