Hyperlinks Dataset

This dataset contains subpage links, their features, and corresponding search engine rankings from Google and Bing. The data was collected as part of the research project: "Accessible Hyperlinks and Search Engine Rankings: An Empirical Investigation".

Dataset Description

This dataset is designed to facilitate research on the relationship between website accessibility, specifically hyperlink accessibility, and search engine rankings. It consists of subpage links extracted from various websites, along with their associated features and ranking positions on Google and Bing SERPs.

Data Sources

The dataset was constructed using the following sources and processes:

1. Trending Keywords: Initial keywords were obtained from Google Trends ("Shopping" category) over a 7-day period. These keywords are provided in raw/keywords/trending_US_7d_20141203-1845.csv.
2. Google SERPs: The top 5 Search Engine Results Pages (SERPs) for each trending keyword were collected from Google.
3. Website Extraction: Unique websites were extracted from the collected Google SERPs, forming the basis for further data collection.
4. Website Scraping: Each website from processed/websites.csv was scraped to collect all of its subpage links and their features. These are stored in processed/links.csv.
5. Subpage Link Collection from Search Engines: The site: operator was used on both Google and Bing to collect subpages associated with each website. Results are in processed/google.csv and processed/bing.csv.
6. Matching and Labeling: Subpage links from the website scraping step were matched with the search engine results and labeled with their corresponding ranking positions (Google and Bing). This labeled data is stored in processed/labeled_links.csv.
7. Dataset Balancing: Balanced datasets were created based on position numbers (grouped into pages) and the presence or absence of position numbers. These are in the datasets/ directory.
8. Feature Extraction: Accessibility and other relevant features were extracted from the labeled datasets. The extracted features are stored in the extracted/ directory.

Data Fields

processed/links.csv (Raw Link Features - scraped from websites):

• order: The order of the link in the HTML document.
• href: The URL the link points to.
• innerText: The visible text of the link.
• innerHTML: The HTML content within the link tag.
• title: The title attribute of the link.
• ariaLabel: The aria-label attribute of the link.
• ariaLabeledText: Text of the element referenced by aria-labelledby.
• isVisual: Whether the link is visually rendered.
• ariaHidden: The aria-hidden attribute of the link.
• isVisible: Whether the link is visible on the page.
• target: The target attribute of the link (e.g., "_blank").
• role: The ARIA role of the link.
• tabIndex: The tabindex attribute of the link.
• isKeyboardAccessible: Whether the link is accessible via keyboard.
• fontSize: The font size of the link text.
• fontWeight: The font weight of the link text.
• underlined: Whether the link text is underlined.
• color: The text color of the link.
• backgroundColor: The background color behind the link element.
• actualBackgroundColor: The actual rendered background color.
• x: The x-coordinate of the link on the page.
• y: The y-coordinate of the link on the page.
• width: The width of the link element.
• height: The height of the link element.
• documentWidth: The width of the document.
• documentHeight: The height of the document.
• deviceScaleFactor: The device scale factor.
• viewportWidth: The width of the viewport.
• viewportHeight: The height of the viewport.
• hasAltText: Whether the link has alternative text (relevant for image links).

processed/labeled_links.csv:

• This file contains the same features as links.csv plus:
• google_position: The ranking position of the link on Google SERPs.
• bing_position: The ranking position of the link on Bing SERPs.

datasets/ (Balanced datasets with raw features):

• The balanced datasets in this folder contain a subset of the raw features from labeled_links.csv. The exact features included may vary depending on the specific balancing criteria used. Refer to the Data Collection and Processing repository for details.

extracted/ (Extracted Accessibility Features):

• label: The class label derived from the search engine ranking position. (e.g., page number of the link on SERPs).
• has_text: Whether the link has innerText, title, ariaLabeledText.
• has_html: Whether the link has innerHTML.
• is_visual: Whether the link is visually rendered (same as in links.csv).
• is_bold: Whether the link text is bold.
• is_underlined: Whether the link text is underlined (same as in links.csv).
• is_visible: Whether the link is visible on the page (same as in links.csv).
• is_keyboard_accessible: Whether the link is accessible via keyboard (same as in links.csv).
• is_unique: Whether the link is unique on the page based on its href attribute.
• contrast_ratio: The contrast ratio of the link text against its background.
• text_clarity: a custom metric based on text length, font-size and color.
• in_viewport: Whether the link is within the viewport.

Dataset Structure

├── raw/                  
│   ├── keywords/
│   │   └── trending_US_7d_20141203-1845.csv 
│   ├── screenshots/ [NOT INCLUDED]
│   └── search_results/ [NOT INCLUDED]
│       ├── google/
│       └── bing/
│       └── trends/  
│   └── site_links/    
├── processed/  
│   ├── trends.csv 
│   ├── websites.csv 
│   ├── bing.csv         
│   ├── google.csv
│   ├── labeled_links.csv
│   ├── links.csv 
│   └── images/ [NOT INCLUDED]
├── datasets/  
└── extracted/   

• raw/: Contains the initial keyword data.
• processed/: Contains intermediate data files generated during the data collection and processing pipeline.
• datasets/: Contains the final balanced datasets with raw features, split by search engine and ready for machine learning.
• extracted/: Contains the extracted features, split by search engine.

Note: The raw/screenshots/, raw/search_results/, raw/site_links/, and processed/images/ folders are not included in this dataset.

Data Collection and Processing

Our data collection and processing pipeline is open-source and available in the following repository: https://github.com/goker-dev/hyperlinks. This repository provides detailed instructions on how to reproduce the data collection and processing steps.

Citation

If you use this dataset in your research, please cite the following paper: