RARE (Randomized AMRs with Rewired Edges)
Introduced by Kachwala et al. in Rematch: Robust and Efficient Matching of Local Knowledge Graphs to Improve Structural and Semantic Similarity
RARE consists of English AMR pairs with similarity scores that reflect the structural differences between them.
Given that AMRs are graphical representations of text, an AMR similarity metric should be sensitive to structural variations between AMRs, even if its labels remain unchanged. Since there is no established evaluation of AMR metrics on structural similarity, we have developed a new benchmark dataset called Randomized AMRs with Rewired Edges (RARE). RARE consists of English AMR pairs with similarity scores that reflect the structural differences between them.
It can be used to evaluate AMR metrics on their sensitivity at detecting structural changes between pairs of AMRs.
Papers
| Paper | Code | Results | Date | Stars |
|---|
