Submissions are evaluated using
Top-1 Accuracy
for both prediction targets.

Two metrics are computed:
Verb Accuracy
Noun Accuracy
The final competition score is:
Final Score = 0.5 * Verb Accuracy + 0.5 * Noun Accuracy
The leaderboard is divided into:

Public leaderboard (subset of test data)

Private leaderboard (hidden test subset)

Final rankings are determined by the private score.