Table 4 Inter-annotator agreement scores per domain and data source reported in terms of average per label F1 scores, macro-averaged F1 and accuracy (and standard deviation in brackets)
From: Classifying patient and professional voice in social media health posts
| Â | Cardio/Reddit | Cardio/Twitter | Skin/Reddit | Skin/Twitter | All |
|---|---|---|---|---|---|
F1: Other | 0.90 (0.01) | 0.93 (0.03) | 0.89 (0.02) | 0.95 (0.03) | 0.93 (0.03) |
F1: Patient voice | 0.96 (0.01) | 0.69 (0.09) | 0.97 (0.01) | 0.53 (0.19) | 0.93 (0.03) |
F1: Professional Voice | 0.85 (0.03) | 0.59 (0.07) | 0.18 (0.15) | – (–) | 0.59 (0.06) |
Macro averaged F1 | 0.90 (0.03) | 0.73 (0.06) | 0.68 (0.05) | 0.74 (0.11) | 0.81 (0.04) |
Accuracy (%) | 0.94 (0.01) | 0.87 (0.04) | 0.95 (0.01) | 0.91 (0.05) | 0.92 (0.03) |