The dataset is currently empty. Upload or create new data files. Then, you will be able to explore them in the Dataset Viewer.
InspiredHub Aesthetic Attention Dataset
Status: Data accumulation in progress. Estimated public release: Q4 2026.
Research access requests accepted now — see Contact below.
Collected by: InspiredHub — A Sanctuary for Artists, Creators, and AI
Organization: InspiredHubAI
What This Dataset Will Contain
Most existing aesthetic preference datasets are built on explicit ratings — crowdsourced workers asked to score artworks on a 1–5 scale in a 10-second task. This dataset is different in kind, not just degree.
InspiredHub is a platform where aesthetically motivated users — artists, designers, and creatives — voluntarily explore classical fine art at their own pace, with no task, no reward, and no time pressure. The behavioral signals we collect are therefore a direct window into genuine human aesthetic attention: what people choose to look at, how long they stay, and where their attention travels next.
When released, this dataset will include three components:
1. Aesthetic Attention Events
Timestamped interaction records (view, save, like) enriched with artwork metadata: title, artist, historical era, medium, geographic origin. The key signal is dwell time — measured in milliseconds from page load to navigation away — which serves as a proxy for aesthetic absorption. A user who spends 17 minutes with a Rembrandt self-portrait is expressing something that no rating scale can capture.
2. Artwork Engagement Profiles
Per-artwork aggregated statistics: total view count, mean and maximum dwell time, deep engagement rate (fraction of views lasting ≥ 1 minute), and navigation centrality (how often an artwork serves as a hub that draws users from other works). These profiles form the foundation for aesthetic signal vectors — dense numerical representations of each artwork's attentional gravity.
3. Aesthetic Navigation Graph
A directed graph of artwork-to-artwork transitions, capturing the paths users take as their aesthetic attention flows across paintings. Each edge represents a user who was viewing artwork A and then chose to view artwork B — an implicit aesthetic judgment that A and B share something worth following. This graph structure is, to our knowledge, not available in any existing public dataset.
Why This Matters for AI Research
The field of aesthetic modeling faces a fundamental data problem. RLHF for visual aesthetics has largely relied on either:
- Explicit preference datasets (WikiArt ratings, OAEI, AVA) — collected under artificial conditions, often by non-expert raters
- Social media engagement (likes, shares) — confounded by social dynamics, algorithmic amplification, and novelty bias
Neither captures the kind of slow, deliberate, intrinsically motivated aesthetic attention that characterizes how artists and designers actually engage with art.
InspiredHub's dataset is designed to fill this gap. Our users come to the platform specifically to engage with classical art. Their dwell times, navigation paths, and save behaviors are signals from people who have already self-selected for aesthetic seriousness — the closest available proxy to the aesthetic judgment of a trained human eye.
This makes the dataset particularly relevant for:
- Training or fine-tuning models for artwork recommendation and aesthetic similarity
- Studying the relationship between art-historical context (era, medium, origin) and human engagement depth
- Building taste models that reflect genuine aesthetic preference rather than social popularity
- Providing behavioral grounding for AI systems designed to support creative work
Data Collection and Privacy
All behavioral data is collected with explicit user consent. InspiredHub's data philosophy is built on transparency: users are told precisely what is collected, why it is collected, and what it will be used for. Users who do not consent to data contribution can use the platform freely without their behavior being recorded.
When released, the dataset will contain no user-level identifiers. All data will be aggregated at the artwork level or anonymized at the event level such that no individual's browsing history can be reconstructed.
Access Model
This dataset will be released under a gated access model. Researchers affiliated with academic institutions or non-profit research organizations may apply for access. Commercial use requires a separate licensing agreement with InspiredHub.
This approach reflects our belief that aesthetic attention data is a shared cultural resource that should serve research and creativity — not be extracted for commercial advantage without acknowledgment of its origins.
Roadmap
| Version | Target | Contents |
|---|---|---|
| v0.1 (current) | Placeholder | This README; no data files yet |
| v0.2 | Q4 2026 | First data release: 5,000+ events, 500+ artworks, gated access |
| v0.3 | Q1 2027 | Cross-domain extension: music and literary engagement alongside visual art |
| v1.0 | 2027 | Full aesthetic trajectory dataset with longitudinal taste evolution signals |
Contact
For research partnership inquiries, early access requests, or questions about the dataset:
Email: research@inspiredhub.ai
Website: inspiredhub.ai
HuggingFace: InspiredHubAI
We are particularly interested in connecting with researchers working on:
- Computational aesthetics and visual preference modeling
- RLHF for creative and aesthetic tasks
- Human-AI collaboration in artistic practice
- Art history and digital humanities
Citation
If you reference this dataset or the InspiredHub research program in your work, please cite:
@dataset{inspiredhub_aesthetic_attention,
title = {InspiredHub Aesthetic Attention Dataset},
author = {InspiredHub},
year = {2026},
url = {https://huggingface.co/datasets/InspiredHubAI/aesthetic-attention},
note = {Dataset in preparation. Behavioral signals from self-directed human engagement with classical fine art.}
}
About InspiredHub
InspiredHub is a platform built for artists, designers, and creatives who believe that deep engagement with classical art is the foundation of original creative work. Unlike algorithmic content feeds optimized for engagement metrics, InspiredHub is curated — every artwork, book, and piece of music on the platform has been selected for its aesthetic depth and historical significance.
InspiredHub's long-term research mission is to build the world's most carefully collected human aesthetic attention dataset — a record of how aesthetically motivated humans engage with great art over time, at scale, and across cultures. This HuggingFace repository is the public home of that ongoing effort.
- Downloads last month
- 30