Wiki-Disease-Benchmark

This benchmark consist in 255 randomly selected disease descriptions, as of February 2024. Each disease description was labeled by two data annotators who reviewed each other's annotations to ensure accuracy and consistency across the dataset. 

This procedure involves collecting, parsing and extracting data from Wikipedia using a software routine that interfaces with an API \footnote{https://pypi.org/project/Wikipedia-API/} to systematically retrieve and collate information related to a predefined disease. Specifically, it searches for pages with a certain disease and, within those pages, extracts the "Sings and Symptoms" section.

This process has two steps:

Retrieve all the labels rdfs:label of triples in DBpedia \footnote{https://dbpedia.org/} that are a disease rdf:type dbo:Disease.

With these labels, go to each page of Wikipedia and scrape the section "Signs and Symptoms".

After extracting the text from Wikipedia, the phenotypical entities were annotated.