Datasetpublic

Sonjoy/social-commerce-screenshot-claim-glance

Social Commerce Screenshot Claim Glance This is a 100-image public preview subset for research on catalog-grounded visual retrieval and exact-claim authorization in social-commerce screenshots. The dataset name for Hugging Face should be: social-commerce-screenshot-claim-glance What It Contains The preview contains synthetic/redacted app-style screenshots that resemble noisy buyer or social-commerce inputs. The examples include platform-like UI framing, cropped… See the full description on the dataset page: https://huggingface.co/datasets/Sonjoy/social-commerce-screenshot-claim-glance.

sourceHugging Faceotherupdated 3mo agoView on Hugging Face

0likes140downloads

Dataset Card

Social Commerce Screenshot Claim Glance

This is a 100-image public preview subset for research on catalog-grounded visual retrieval and exact-claim authorization in social-commerce screenshots.

The dataset name for Hugging Face should be:

social-commerce-screenshot-claim-glance

What It Contains

The preview contains synthetic/redacted app-style screenshots that resemble noisy buyer or social-commerce inputs. The examples include platform-like UI framing, cropped products, compression artifacts, captions, price/status overlays, partial products, and visually similar non-target products.

Label distribution:

label	count	meaning
`exact_in_catalog`	40	screenshot contains the exact catalog target
`visually_similar_wrong_product`	30	product is visually similar but not the exact target
`out_of_catalog`	20	product should not map to any catalog item
`ambiguous_partial_low_quality`	10	partial, low-quality, or uncertain visual evidence

Source-style distribution:

source_type	count
`facebook_post`	25
`instagram_post`	23
`messenger_chat`	27
`whatsapp_chat`	25

Folder Structure

text

.
├── README.md
├── metadata.csv
├── metadata.jsonl
├── summary.json
└── images/
    ├── glance_0001.jpg
    ├── glance_0002.jpg
    └── ...

The metadata.csv file follows the Hugging Face ImageFolder convention using the file_name column.

Metadata Fields

field	description
`file_name`	relative path to the image
`screenshot_id`	public preview identifier
`source_type`	app-style UI template
`label`	ground-truth claim label
`target_product_id`	pseudonymized exact target ID, or `none`
`displayed_product_id`	pseudonymized displayed/source product ID
`confusable_product_id`	pseudonymized visually confusable product ID, or `none`
`department`	broad catalog department
`garment_type`	broad product category
`color_families`	coarse color family
`render_profile`	synthetic rendering/noise recipe
`width`, `height`	image dimensions
`privacy_actions`	redaction and privacy transformations
`is_synthetic_ui`	always true for this preview
`sha256`	checksum of the released image file
`notes`	short label explanation

Privacy And Scope

This preview is synthetic/redacted. Usernames, avatars, comments, phone numbers, order IDs, and private chat details are removed or replaced. Product-like identifiers in the public metadata are pseudonymized.

This preview is not a claim that the benchmark contains real permissioned customer screenshots. It is intended as a small public glance of the evaluation setting.

Intended Use

Use this subset to inspect examples, test data loading, demonstrate social-commerce screenshot noise, and reproduce small qualitative examples for catalog-grounded retrieval or exact-claim authorization papers.

Do not use this preview for production identity matching, biometric analysis, or claims about real customer-chat behavior.