Dataset Card for the Wikipedia-Physics Corpus

Wikipedia-Physics Corpus contains 102,409 paragraphs extracted from 6642 key physics-related Wikipedia articles as well as word-sense labels for 1,186 occurrences of "Planck" across 885 paragraphs.

This table details the main columns of the dataset:

Column	Description
Text	Full text of the paragraph
Characters	Number of unicode characters in the paragraph
Subwords	Number of BERT subwords in the paragraph
PlanckOccurrences	Number of occurrences of "Planck" in the paragraph
PlanckLabels	A list of word-sense labels for all occurrences of "Planck" in the paragraph

The primary purpose of the Wikipedia-Physics Corpus lies in its use for analyzing the meaning of concepts in physics. For further insights into the corpus and the underlying research project (Network Epistemology in Practice) please refer to this paper [link coming soon].

Construction

These articles were selected using the PetScan tool, which generated a list of all pages categorized under "physics" or its immediate subcategories. Markup was removed and minimal cleaning applied to produce plain text paragraphs, and while references were removed, all formulas were retained.

Additionally, 1,186 occurrences of the term "Planck" were labeled across 885 paragraphs. These occurrences were identified using a case-insensitive regular expression to capture variations like "PLANCK" and "Planck(2015)"" while excluding irrelevant forms like "planckian". The labels denote distinct meanings of the target term, such as PERSON, CONSTANT, UNITS, LAW, MPS, MISSION, and FOKKER.

Details

Developer: Arno Simons
Funded by: The European Union under Grant agreement ID: 101044932
Language: English