rg65-gold-standard-dataset-65-pairs
IN premise — summaries/2026/08/24/wiki-Semantic_similarity-chunk-1.md
Created 2026-08-25T02:58:53+00:00
RG65 is a gold-standard evaluation dataset containing 65 human-judged word pairs used to benchmark computational semantic similarity measures.
Summary
RG65 is a fixed reference set of 65 word pairs whose similarity ratings were decided by humans, and it serves as the yardstick against which any computational measure of word meaning is scored. In practical terms, this gives the system a stable, agreed-upon target: if a model's similarity scores diverge from the human judgments in this set, something is wrong with the model, and the set is treated as ground truth rather than something to be derived.