Query-Free OpenWebText - Part 2: SPARQL Injection Corpus

This dataset is derived from OpenWebText and filtered so that it does not contain any SQL, SPARQL or Cypher in the plain text.

This is Part 2 of the Query-Free OpenWebText dataset, which contains SPARQL-injected versions.
SPARQL injected versions are derived from OpenWebText-clean

The OpenWebText SPARQL Injection Corpus consists of two variations (Injected-10% and Injected-30%) derived from the Clean OpenWebText dataset and augmented with SPARQL queries. This process was designed to simulate controlled pre-training bias (RQ2).

Dataset Format: Arrow