Precision: 16.84Recall: 9.72F1: 11.43
Papers with Code · CC BY-SA 4.0
Dialogue
Dialogue is notoriously hard to evaluate. Past approaches have used human evaluation.
논문
BLEU: 1.99Dis-1: 0.0056Dis-2: 0.0431Dis-3: 0.125
Perplexity: 32.48
Slot Accuracy: 0.973Joint SA: 0.600Inform: 75.1Inform_mct: 90.8
mauve: 0.968
Joint: 85
1 in 10 R@2: 5
Joint: 73.6
Request: 97.5Area: -Food: -Price: -
relevance (human): 2.72interest (human): 2.53
Avg F1: 21.99
Request: 96.5Area: 90Food: 84Price: 94
Avg F1: 16.18
Avg F1: 11.9
BLEU-1: 14.7ROUGE-L: 14.79METEOR: 6.39CIDr: 19.09
BLEU-1: 49.581BLEU-2: 39.24Distinct-1: 1.44Distinct-2: 8.89
Joint: 75.5
Avg F1: 19.09
Avg F1: 19.77