iampom/WEAR-PD

Fine-Grained Fashion Feature Embedding for Global, Local, and Contextual Analysis in Fashion Social Media Popularity Prediction

This repository provides the dataset used in the paper "Fine-Grained Fashion Feature Embedding for Global, Local, and Contextual Analysis in Fashion Social Media Popularity Prediction" (IEEE Access 2026).

Paper Link: https://ieeexplore.ieee.org/document/11522804

Project Link: https://github.com/minotake-match-life/FFGLC

WEAR-PD

WEAR-PD is a fashion social media popularity prediction dataset constructed from WEAR, a Japanese fashion coordination platform. WEAR-PD covers 313,532 posts from 17,187 users, collected from December 2012 to May 2024. Additional details are listed in the supplementary material.

This release includes anonymized metadata, image URLs, generated captions, and predefined train, validation, and test splits. Raw WEAR images and original user IDs are not redistributed.

Dataset Contents

Each sample contains the following fields:

• —sample_id: internal sample identifier
• —split : train or val or test
• —image_url: URL to the publicly available image
• —post_date: date of the post
• —hashtags: hashtags included in the post
• —description: description text included in the post
• —itemtag_count: number of tagged fashion items
• —comment_status: whether comments are allowed
• —view_count_log2: base-2 logarithm of the view count
• —anon_user_index: internal anonymized user index
• —official_status: whether the user is official
• —post_count: number of posts by the user
• —following_count: number of users followed by the user
• —follower_count: number of followers of the user
• —caption_c1: general description caption
• —caption_c2: item analysis caption
• —caption_c3: color analysis caption
• —caption_c4: seasonal context caption
• —caption_c5: composition analysis caption

Data Splits

The dataset provides predefined train, validation, and test splits:

• —train.csv
• —val.csv
• —test.csv

Image Policy

This dataset does not redistribute raw WEAR images. It only provides image URLs for reproducibility. Image URLs may become unavailable if the original content is removed or changed.

Privacy and Terms

Original user IDs are not included. They are replaced with internal anonymized user indices.

Users must comply with the terms of the original platform. This dataset must not be used for re-identification, user profiling, commercial targeting, redistribution of raw images, or any use that may harm the privacy or rights of the original users.

Citation

If you find this dataset useful, please cite:

@ARTICLE{FFGLC2026,
  author={Okamoto, Shun and Iizuka, Satoshi and Fukui, Kazuhiro},
  journal={IEEE Access}, 
  title={Fine-Grained Fashion Feature Embedding for Global, Local, and Contextual Analysis in Fashion Social Media Popularity Prediction}, 
  year={2026},
  volume={14},
  number={},
  pages={75674-75689},
  doi={10.1109/ACCESS.2026.3694039}
}