The F1 score is the harmonic mean of precision and recall — a single figure that rates a result well only when it gets both right, and drops sharply whenever one is traded away for the other.
| Term | F1 Score |
|---|---|
| Category | Statistics and Evidence |
| Also known as | F1 |
| Where it appears | Semantic enrichment evaluation |
What it means in RankGear
RankGear reports the F1 score inside semantic enrichment evaluation, where the app scores how well an automated pass tagged, classified, or matched entities and topics against a known reference set. Precision asks how many of the items the pass flagged were actually correct; recall asks how many of the genuinely correct items it managed to find. The F1 score folds those two into one number so evaluation runs can be ranked against each other without reading two columns side by side.
How to interpret it
F1 runs from 0 to 1 (shown as 0 to 100% in some views), and because it is a harmonic mean rather than a plain average it stays low whenever precision and recall are far apart. A pass that finds almost everything but is frequently wrong, and one that is nearly always right but misses most items, both score poorly — only a run that is strong on both climbs toward 1. Read it next to the raw precision and recall figures and the sample size behind them: a high F1 on a handful of items is far less trustworthy than a slightly lower one measured across hundreds.
| Precision | Share of flagged items that were actually correct |
|---|---|
| Recall | Share of the truly correct items that were found |
| F1 | Harmonic mean of the two — low unless both are high |
Example
Say a semantic enrichment run labels 100 passages with entity tags. It flags 40 passages, 32 of them correct, giving precision of 0.80; but 50 passages truly should have been flagged, so recall is only 0.64. The F1 score works out to about 0.71 — pulled closer to the weaker recall figure than a simple average would be, a clear signal that the run is still missing more than a third of the entities it should catch.
Important considerations
- F1 weights precision and recall equally, so when a miss costs more than a false flag (or the reverse), the single score can hide the trade-off that matters most to you.
- It says nothing about which items were confused — two runs with an identical F1 can fail in completely different ways, so pair it with the underlying error breakdown.
- Small evaluation sets make F1 jump around; a gap between two runs measured on only a few dozen items may be noise rather than a real difference.
- F1 is an internal indicator of enrichment quality, not a Google ranking score. A better F1 means a cleaner enrichment pass — it does not, on its own, move a page up in search results.
Related terms
Part of the RankGear glossary · how RankGear measures · the 870 factors.