Examining bias perpetuation in academic search engines: an algorithm audit of Google and Semantic Scholar

Main dataset (main.csv)

The main file contains an entry (N=28530) per search result in all collected pages. It comprises the following columns:

id: Unique identifier of the file (corresponds to the last part of the filename)

filename: Name of the file associated with the row (the file is in serp_html.zip)

engine: The search engine used (Google Scholar or Semantic Scholar).

browser: The web browser used for the search (Firefox or Chrome)

region: The geographical region where the search was made.

year: The year when the search was made

month: The month when the search was made

day: The day when the search was made

query: The full search query that was used

query_type: The type of the search query (health or technology)

topic: The topic associated with the search query ('covid vaccines', 'cryptocurrencies', 'internet', 'social media',  'vaccines', 'coffee')

trt: Treatment variable associated with the search (benefits or risks).

url: The URL of the (article) search result

title: The title of the (article) search result.

authorship: The author(s) of the (article) search result.

abstract_id: Unique identifier for the abstract of the (article) search result which connects with annotated-abstracts_v0.6.xlsx

abstract_hash: Hash value of the abstract for data integrity

link_n: The total number of results in the search page

rank: The rank of the search result on the search engine results page.

annotation: Any annotations associated with the (article's abstract) search result. One of: '3. Confirms both benefits and risks',  '4. Confirms neither benefits nor risks', '1. Confirms benefits',   '2. Confirms risks', '5. Abstract not related to {topic}')

valence:  -1 for abstracts containing risks, 0 for neutral abstracts, 1 for abstracts only containing benefits

Annotated abstracts (annotated-abstracts_v0.6.xlsx)

Manually annotated abstracts resulting from the searches.

Raw search engine result pages (serp_html.zip)

The zip contains an HTML per search engine result page collected (N=2853). See column filename from the main dataset.