LLM-Based Interviews Improve AI Aesthetic Judgment

Researchers have developed an integrated deep learning and large language model system that outperforms both conventional AI and human predictors in personalized image aesthetics assessment. The system uses LLM-based semi-structured interviews to actively elicit individual aesthetic preferences, then combines low-level image features with high-level semantic features extracted via the LLM. In experiments, the proposed system achieved higher accuracy than standard deep learning models and human raters. The work addresses the fundamental challenge that aesthetic preferences are inherently subjective and individual-dependent, which traditional objective feature extraction fails to capture. The study is published on arXiv under identifier 2605.14761.