AI Outperforms Humans in Personalized Image Aesthetics Assessment via LLM-Based Interviews and Semantic Feature Extraction

2026-05-14 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Human-Computer Interaction · Depth: Expert, quick

Summary

A novel integrated Deep Learning (DL) and Large Language Model (LLM) system has been developed to accurately predict individual aesthetic evaluations for images. This system actively elicits aesthetic preferences through LLM-based semi-structured interviews and leverages both low-level image features extracted by DL models and high-level semantic features. Experiments compared this system against conventional AI models, human predictors, and an individual's own re-evaluations over time. The proposed system demonstrated superior performance, particularly for highly-rated images, achieving a prediction error smaller than within-person variability. Human predictors exhibited the largest error, suggesting AI's potential to better capture individual aesthetic preferences than humans or even one's future self.

Key takeaway

For AI Scientists developing personalized content recommendation systems, this research indicates that integrating LLMs for preference elicitation with DL for feature extraction can yield superior aesthetic prediction. You should explore hybrid DL-LLM architectures to capture nuanced individual tastes, potentially outperforming human evaluators and improving user satisfaction in applications like image curation or design.

Key insights

An integrated DL-LLM system accurately predicts individual image aesthetics by combining low-level and high-level feature extraction.

Principles

• Aesthetic preferences are inherently subjective.
• LLMs can elicit individual aesthetic preferences.
• AI can outperform human aesthetic prediction.

Method

The system uses LLM-based semi-structured interviews to gather preference data, then predicts aesthetic evaluation by integrating low-level DL features with high-level semantic features.

In practice

• Use LLMs for personalized preference elicitation.
• Combine DL and LLM for subjective assessment.
• Benchmark against within-person variability.

Topics

• Personalized Image Aesthetics
• LLM-based Interviews
• Semantic Feature Extraction
• DL-LLM System
• Human-AI Performance

Best for: AI Scientist, Research Scientist

Related on AIssential

• Personalized Benchmarking: Evaluating LLMs by Individual Preferences — Aggregate LLM benchmarks fail to capture individual user preferences, necessitating personalized evaluation methods.
• Last Week in AI #339 - DLSS 5, OpenAI Superapp, MiniMax M2.7 — DLSS 5 uses generative AI to enhance real-time game photorealism by analyzing scene semantics and lighting.
• LLM-Based Visual Explanation Evaluation Framework for Assessing the Explainability of Facial Skin Disease Classification Models — An LLM-based framework evaluates Grad-CAM explanations for facial skin disease models, ensuring clinical relevance and trustworthiness.
• From Fallback to Frontline: When Can LLMs be Superior Annotators of Human Perspectives? — LLMs can statistically outperform human annotators in estimating aggregate subgroup opinions under common conditions.
• FSPO: Few-Shot Optimization of Synthetic Preferences Personalizes to Real Users — FSPO enables LLMs to personalize responses by meta-learning user preferences from few-shot synthetic data.
• Semantic Search At LinkedIn, LLM-Driven Autonomous Optimization for Industrial-Scale Recommendation Systems, and More! — Recent advancements in information retrieval focus on LLM-driven optimization, efficient embeddings, and scalable recommendation systems.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.