OSimUnr

The OSimUnr dataset consists of nearly one million noun word-pairs (wps) for the English and Turkish languages. Final relatedness scores of such word-pairs are obtained through graph-based WordNet similarity/relatedness approximation methods, which are known to be roughly 50% correlated with human annotated relatedness scores. The dataset does not contain any human-annotated data. Word-pairs are automatically selected by a complex pipeline, designed for a specific scenario: orthographically-similar-but-semantically-unrelated (OSimUnr).