Dataset: Fashion-related Objects on Instagram
To analyze fashion trends on Instagram, a dataset comprising fashion-related objects from the platform is created through a multi-step process:
- Account Selection: A list of prominent fashion influencers on Instagram is compiled.
- Data Crawling: A limited number of recent posts from these influencers are scraped.
- Open-set Object Detection: An open-set object detection model is employed to identify all objects in the post images.
- Object Grouping and Filtering: Tag groups are formed by grouping similar object tags, and those tag groups that are likely related to fashion are then selected for further analysis.
Dataset Summary
The dataset construction process can be summarized in the following stages:
| Stage | Result |
|---|---|
| Account Selection | 724 accounts |
| Data Crawling | 647 accounts, 30,337 posts, 80% of posts from 2023, median post count of 46 |
| Open-set Object Detection | 30,179 images, 29,937 with detected objects, 182,123 objects identified, 4,561 unique tags |
| Object Grouping and Filtering | 76,485 objects classified into 13 selected groups, 61,751 objects from 2023 |
As a result, the final dataset comprises 61,751 fashion-related objects sourced from 647 accounts and 30,337 posts, exclusively from the year 2023.