WebColor dataset

A dataset for web page coloring task consisting of e-commerce mobile web pages, adapted from the Klarna Product Page Dataset.

Training examples	Data structure

Download dataset

./download.sh {main,cache,text,image}

• "main" (1.5 GB)

  • Main dataset
  • Split file
  • Note: Processing to DGL format takes about 50 minutes.

• "cache" (693 MB, recommended)

  • Processed DGL cache
  • Split file

• "text" (356 MB, optional)

  • Text data for reference

• "image" (6.3 GB, optional)

  • Image data for reference

Data format

The dataset is organized in HDF5 format, and its contents can be accessed using the h5py library as follows:

import h5py

f = h5py.File('data/webcolor_v1.1.hdf5')
for data_name, grp in f.items():
    for node_id, dset in grp.items():
        node_id = int(node_id)
        parent_id = int(dset[()])
        node_attrs = dict(dset.attrs)

Variable	Type	Description
data_name	str	Name to identify the web page in the original datasets
node_id	int	Node ID
parent_id	int	Parent node ID (-1 indicates no parent, i.e., root)
node_attrs	dict	Metadata corresponding to the node (element)

The contents of node_attrs are listed below.

Key	Value type	Description
"text_color"	str	Text color (computed value of the CSS property color, text elements only)
"background_color"	str	Background color (computed value of the CSS property background-color)
"sibling_order"	int	Order between sibling elements
"html_tag"	str	HTML tag
"text_feat"	numpy.ndarray (dtype=float32, shape=(13,))	Low-level features of text (text elements only)
"img_feat"	numpy.ndarray (dtype=float32, shape=(13,))	Low-level features of image (image elements only)
"bgimg_feat"	numpy.ndarray (dtype=float32, shape=(12,))	Low-level features of background image (only for elements with a background image)

Details of the text and image features can be found in the supplemental material.

Data split

Train: 27,630 pages
Val: 3,190 pages
Test: 13,228 pages

License

This dataset is based on the Klarna Product Page Dataset and is licensed under CC BY-NC-SA 4.0.

Versions

1.1: fix bug on text color (Dec 29, 2022)

1.0: v1 release (Dec 23, 2022)