Table 7 Statistical values on the split of non-toxic dialogues by interlocutor for each category predicted by Detoxify.
Non-Toxic Split | Expert | User | ||||||
|---|---|---|---|---|---|---|---|---|
mean | std | min | max | mean | std | min | max | |
toxicity | 2.4E-03 | 2.4E-02 | 3.0E-04 | 9.9E-01 | 1.5E-03 | 1.5E-02 | 3.1E-04 | 2.4E-03 |
Severe toxicity | 4.0E-06 | 4.7E-05 | 1.0E-06 | 6.5E-03 | 2.0E-06 | 1.0E-05 | 1.0E-06 | 4.0E-06 |
obscene | 2.8E-04 | 1.2E-02 | 1.8E-05 | 8.2E-01 | 5.6E-05 | 5.9E-04 | 1.8E-05 | 2.8E-04 |
threat | 8.5E-05 | 2.7E-03 | 1.2E-05 | 4.5E-01 | 4.5E-05 | 8.5E-04 | 1.4E-05 | 8.5E-05 |
insult | 8.1E-04 | 1.6E-02 | 7.4E-05 | 9.8E-01 | 5.0E-04 | 9.2E-03 | 7.2E-05 | 8.1E-04 |
Identity attack | 3.8E-04 | 3.8E-03 | 5.3E-05 | 5.2E-01 | 1.9E-04 | 2.8E-03 | 5.1E-05 | 3.8E-04 |
sexual explicit | 3.8E-04 | 1.2E-02 | 9.0E-06 | 8.4E-01 | 1.9E-04 | 9.1E-03 | 7.0E-06 | 3.8E-04 |
