إضافة قواعد مخصّصة للكيانات

Pre-trained NER never saw your product names or internal jargon. Custom rules let you catch the terms it always misses. 🎯

spaCy's EntityRuler adds entities from patterns you define, no retraining required. It plugs straight into the pipeline.

Insert the ruler with add_pipe. Placing it before ner lets your rules take priority over the model's guesses.

Each rule is a dict with a label and a pattern. The label is the entity type you want assigned to a match.

Feed a list of pattern dicts to ruler.add_patterns. Now those exact phrases get tagged every time they appear.

Beyond plain strings, patterns can be token lists that match on attributes like lowercase text, giving flexible rules.

A token pattern with several entries catches phrases, so San Pedro Lab is tagged as one entity, not three words.

Because the ruler sits before ner, your custom labels win on conflicts, while the model still handles everything else.

Run a sentence through the updated pipeline and loop doc.ents to confirm your new terms now show the right label.

Export rules to disk with to_disk so your whole team reuses the same custom entities instead of redefining them.

Custom rules are a fast first step. Later you can train the model on labeled examples for tricky, fuzzy cases. 🚀

How do you make custom rules override the model?

The EntityRuler adds entities from string or token patterns, runs before ner, and saves to disk for reuse. ✅