SpaCy: Training a custom Named Entity Recognition (NER) model

Photo by Arthur Mazi on Unsplash

Training a custom Named Entity Recognition (NER) model using spaCy involves several steps: preparing a labeled dataset, training the model, and then evaluating its performance.

Preparing a Labeled Dataset

For training, you’ll need a dataset where each entity you want the model to recognize is labeled. The dataset format for spaCy should look like this:

TRAIN_DATA = [
    # GPS_COORDINATES
    ("Coordinates: 123.45 N, 67.89 W", {"entities": [(13, 29, "GPS_COORDINATES")]}),
    ("Location: 12.34 S, 56.78 E", {"entities": [(11, 24, "GPS_COORDINATES")]}),
    ("Lat/Long: 98.76 S, 54.32 W", {"entities": [(10, 24, "GPS_COORDINATES")]}),
    ("Position: 65.43 N, 21.45 E", {"entities": [(10, 25, "GPS_COORDINATES")]}),
    ("Coords: 45.67 S, 78.90 W", {"entities": [(8, 22, "GPS_COORDINATES")]}),
    ("Latitude/Longitude: 34.56 N, 78.90 E", {"entities": [(20, 37, "GPS_COORDINATES")]}),
    ("GPS: 23.45 N, 56.78 W", {"entities": [(5, 19, "GPS_COORDINATES")]}),
    ("Place: 67.89 S, 12.34 E", {"entities": [(7, 21, "GPS_COORDINATES")]}),
    ("Geolocation: 45.67 N, 89.01 W", {"entities": [(13, 28, "GPS_COORDINATES")]}),
    ("Mapping: 78.90 N, 12.34 E", {"entities": [(9, 24, "GPS_COORDINATES")]}),

    # BANK_ACCOUNT
    ("Bank account: 1234-5678-1234-5678", {"entities": [(13, 33…