flowxai
/

moderation

+# moderation evaluation
+**Threshold 0.84 (calibrated on the validation split, objective macro_f1).** The config default of 0.5 scored 0.923 against 0.931 for the calibrated value, on validation. Every table below is on test, at the calibrated threshold.
+## Per language
+| Language | Support | P | R | F1 | Notes |
+|---|---|---|---|---|---|
+| `bg` Bulgarian | 60 | 1.000 | 0.967 | 0.983 |  |
+| `cs` Czech | 60 | 1.000 | 1.000 | 1.000 |  |
+| `da` Danish | 60 | 1.000 | 0.983 | 0.992 |  |
+| `de` German | 60 | 1.000 | 0.983 | 0.992 |  |
+| `el` Greek | 60 | 1.000 | 1.000 | 1.000 |  |
+| `en` English | 59 | 0.983 | 1.000 | 0.992 |  |
+| `es` Spanish | 59 | 1.000 | 0.983 | 0.991 |  |
+| `et` Estonian | 60 | 1.000 | 1.000 | 1.000 |  |
+| `fi` Finnish | 60 | 1.000 | 1.000 | 1.000 |  |
+| `fr` French | 59 | 1.000 | 1.000 | 1.000 |  |
+| `ga` Irish | 60 | 0.952 | 0.983 | 0.967 |  |
+| `hr` Croatian | 58 | 0.983 | 1.000 | 0.991 |  |
+| `hu` Hungarian | 60 | 1.000 | 1.000 | 1.000 |  |
+| `it` Italian | 59 | 1.000 | 1.000 | 1.000 |  |
+| `lt` Lithuanian | 60 | 0.984 | 1.000 | 0.992 |  |
+| `lv` Latvian | 60 | 1.000 | 0.967 | 0.983 |  |
+| `mt` Maltese | 59 | 0.982 | 0.949 | 0.966 | not in base model pretraining |
+| `nl` Dutch | 60 | 1.000 | 0.983 | 0.992 |  |
+| `pl` Polish | 60 | 1.000 | 1.000 | 1.000 |  |
+| `pt` Portuguese | 60 | 0.968 | 1.000 | 0.984 |  |
+| `ro` Romanian | 60 | 1.000 | 1.000 | 1.000 |  |
+| `sk` Slovak | 60 | 1.000 | 0.983 | 0.992 |  |
+| `sl` Slovenian | 59 | 1.000 | 1.000 | 1.000 |  |
+| `sv` Swedish | 60 | 1.000 | 1.000 | 1.000 |  |
+| `tr` Turkish | 60 | 0.984 | 1.000 | 0.992 |  |
+| `az` Azerbaijani | 60 | 1.000 | 0.967 | 0.983 |  |
+The base-model note is a fact about pretraining, not a cause of the score beside it. `nsfw` Maltese carried the same note at 0.000 and reached 1.000 on corpus size alone, with nothing about the base model changed. Check how many examples a weak score rests on before reaching for this.
+## Per register
+| Register | Support | P | R | F1 | FPR |
+|---|---|---|---|---|---|
+| `cyber_intrusion` | 130 | 1.000 | 1.000 | 1.000 | 0.000 |
+| `cyber_intrusion_near_miss` | 0 | 0.000 | 0.000 | 0.000 | 0.000 |
+| `defamation` | 130 | 1.000 | 1.000 | 1.000 | 0.000 |
+| `defamation_near_miss` | 0 | 0.000 | 0.000 | 0.000 | 0.019 |
+| `election_integrity` | 130 | 1.000 | 0.992 | 0.996 | 0.000 |
+| `election_integrity_near_miss` | 0 | 0.000 | 0.000 | 0.000 | 0.000 |
+| `extremism` | 129 | 1.000 | 0.984 | 0.992 | 0.000 |
+| `extremism_near_miss` | 0 | 0.000 | 0.000 | 0.000 | 0.038 |
+| `fraud_deception` | 130 | 1.000 | 1.000 | 1.000 | 0.000 |
+| `fraud_deception_near_miss` | 0 | 0.000 | 0.000 | 0.000 | 0.058 |
+| `hate_incitement` | 126 | 1.000 | 1.000 | 1.000 | 0.000 |
+| `hate_incitement_near_miss` | 0 | 0.000 | 0.000 | 0.000 | 0.000 |
+| `illicit_drugs` | 130 | 1.000 | 1.000 | 1.000 | 0.000 |
+| `illicit_drugs_near_miss` | 0 | 0.000 | 0.000 | 0.000 | 0.000 |
+| `mundane_informational` | 0 | 0.000 | 0.000 | 0.000 | 0.000 |
+| `mundane_operational` | 0 | 0.000 | 0.000 | 0.000 | 0.000 |
+| `mundane_transactional` | 0 | 0.000 | 0.000 | 0.000 | 0.000 |
+| `property_crime` | 129 | 1.000 | 0.977 | 0.988 | 0.000 |
+| `property_crime_near_miss` | 0 | 0.000 | 0.000 | 0.000 | 0.019 |
+| `self_harm` | 129 | 1.000 | 0.992 | 0.996 | 0.000 |
+| `self_harm_near_miss` | 0 | 0.000 | 0.000 | 0.000 | 0.019 |
+| `sexual_exploitation` | 129 | 1.000 | 0.969 | 0.984 | 0.000 |
+| `sexual_exploitation_near_miss` | 0 | 0.000 | 0.000 | 0.000 | 0.019 |
+| `violent_facilitation` | 130 | 1.000 | 0.977 | 0.988 | 0.000 |
+| `violent_facilitation_near_miss` | 0 | 0.000 | 0.000 | 0.000 | 0.019 |
+| `weapons_cbrn` | 130 | 1.000 | 0.992 | 0.996 | 0.000 |
+| `weapons_cbrn_near_miss` | 0 | 0.000 | 0.000 | 0.000 | 0.000 |
+## Known weaknesses
+The three weakest languages by F1: `mt` at 0.966, `ga` at 0.967, `az` at 0.983.
+These are published rather than dropped. A coverage table with the bad rows removed is not a coverage table.