Table 2 Statistics of the tailored metrics calculated over the released data.

From: A dataset of synthetic art dialogues with ChatGPT

Attribute vs Metric

WER ↓

BLEU-1 ↑

Jaccard ↑

Accuracy ↑

Levenshtein Sim. ↑

Semantic similarity ↑

Painting_profile–Artist_name

4.5e-3 ± 0.04

0.99 ± 0.05

0.99 ± 0.06

0.99 ± 0.05

0.99 ± 0.01

0.99 ± 0.02

Painting_profile–Typical_triggered_emotion_in_viewers

0.16 ± 0.36

0.81 ± 0.33

0.79 ± 0.38

0.86 ± 0.34

0.82 ± 0.31

0.92 ± 0.15

Painting_profile–Artistic_movement_or_school

0.6e-3 ± 0.07

0.61 ± 0.48

0.61 ± 0.48

0.62 ± 0.48

0.62 ± 0.48

0.7 ± 0.38

User_profile—User_emotion

0.91 ± 0.28

0.08 ± 0.27

0.08 ± 0.27

0.09 ± 0.29

0.15 ± 0.27

0.41 ± 0.19

User_profile–User_preferred_artistic_movement

0.46 ± 0.48

0.44 ± 0.48

0.45 ± 0.48

0.49 ± 0.49

0.53 ± 0.41

0.57 ± 0.41

  1. Results show that the synthetic dialogues have high quality for profile recovery.