Skip to content

hiroaki222/aesthetic-prediction-data-collector

Repository files navigation

License: MIT Privacy Policy (EN) Privacy Policy (JA)

English | 日本語

Note

Deployment is currently suspended. The previously available demo at aesthetic-prediction-data-collector.vercel.app is no longer accessible. Please run it locally if you want to try the app.

現在デプロイは停止しています. 以前公開していた aesthetic-prediction-data-collector.vercel.app のデモは利用できません. 試したい場合はローカルで起動してください.


English

Overview

This web application was developed for the data collection phase (Section 2) of the following study:

Hayashi T., Takahara H., Mawalim C. O., Narimatsu H., Kumano S., Okada S. "Cross-Domain Personalized Aesthetic Assessment: Data Collection and Analysis" HCG Symposium 2025.

The study constructs a dataset that captures the aesthetic preferences of the same individuals across three domains — artworks, fashion, and landscape — to investigate the consistency and transferability of personalized aesthetic preferences across domains.

What this app does

The application served as the annotation platform for the data collection. Each rater was assigned one annotation set and asked, for every stimulus, to report:

  • Overall aesthetic judgment on a 7-point scale (1 = not aesthetic at all, 7 = extremely aesthetic)
  • Nine aesthetic emotions on a 5-point scale, based on AESTHEMOS (Schindler et al., 2017): Liked it, Found it beautiful, Found it distasteful, Was impressed, Challenged intellectually, Motivated to act, Felt nostalgic, Made me sad, Amused me

Stimuli from the three domains were interleaved (not presented in long single-domain blocks) so that raters maintained a consistent aesthetic standard across domains.

Data collection results

Item Value
Raters 142 Japanese speakers
Total samples 6,528
Artwork images 2,345 (from LAPIS)
Fashion images 2,083 (from Clothing Co-Parsing)
Landscape videos 2,100 (from Sekai-Real-Walking-HQ, 101 countries)
User–item interactions ~98,000
Collection period Within 1 month

Each rater also provided demographic information (age, gender, nationality, education), Big Five personality traits, and interest level for each domain.

Screenshots

Dashboard

Dashboard

Annotation pages

Image annotation: Annotation page for images

Video annotation: Annotation page for videos

Tech stack

  • Framework: Next.js 15 (App Router) + React 19 + TypeScript
  • Backend: Supabase (Auth + DB), Vercel Blob (media storage)
  • UI: Tailwind CSS v4, Radix UI, shadcn/ui
  • i18n: next-intl
  • Email: Resend

Local setup

bun install
bun run dev

Environment variables for Supabase, Vercel Blob, and Resend are required. See the source for details.

License

MIT


日本語

概要

このWebアプリケーションは, 以下の研究の データ収集フェーズ (2章) のために開発されました.

林貴斗, 高原大明, Mawalim C. O., 成松宏美, 熊野史朗, 岡田将吾. 「ドメインを横断したパーソナライズ美的評価のためのデータ収集と分析」 HCGシンポジウム 2025.

本研究は, 同一個人の美的嗜好を美術作品・ファッション・風景の3ドメインにわたって収集 し, ドメインを跨いだ美的嗜好の一貫性と転移可能性を分析するためのデータセットを構築するものです.

このアプリの役割

データ収集におけるアノテーションプラットフォームとして機能しました. 各評定者は割り当てられた1セットを評価し, 各サンプルに対して以下を報告しました.

  • 総合的な美的判断 を 7件法 (1 = 全く美的でない, 7 = 非常に美的) で評価
  • 9種類の美的感情 を 5件法で評価. 項目は AESTHEMOS (Schindler et al., 2017) に基づく: 好ましい / 美しい / 不快 / 感銘 / 知的刺激 / 動機づけ / 懐かしい / 悲しい / 面白い

3ドメインの刺激は単一ドメインを連続提示するのではなく 交互に提示 することで, ドメイン間で一貫した美的判断基準を促しました.

データ収集結果

項目
評定者 日本語話者 142名
総サンプル数 6,528
美術作品画像 2,345枚 (LAPIS より)
ファッション画像 2,083枚 (Clothing Co-Parsing より)
風景動画 2,100本 (Sekai-Real-Walking-HQ, 101カ国)
ユーザ–アイテム相互作用 約 98,000 件
収集期間 1ヶ月以内

評定者からは人口統計情報 (年齢, 性別, 国籍, 学歴), Big Five 性格特性, 各ドメインへの興味も収集しました.

スクリーンショット

ダッシュボード

ダッシュボード

アノテーション画面

画像アノテーション: 画像のアノテーション画面

動画アノテーション: 動画のアノテーション画面

技術スタック

  • フレームワーク: Next.js 15 (App Router) + React 19 + TypeScript
  • バックエンド: Supabase (認証 + DB), Vercel Blob (メディア保存)
  • UI: Tailwind CSS v4, Radix UI, shadcn/ui
  • 多言語対応: next-intl
  • メール: Resend

ローカル起動

bun install
bun run dev

Supabase, Vercel Blob, Resend の環境変数が必要です. 詳細はソースを参照してください.

ライセンス

MIT