vishalbelsare / extractacy Goto Github PK

View Code? Open in Web Editor NEW

This project forked from jenojp/extractacy

Spacy pipeline object for extracting values that correspond to a named entity (e.g., birth dates, account numbers, laboratory results)

License: MIT License

Python 100.00%

extractacy's Introduction

extractacy - pattern extraction and named entity linking for spaCy

spaCy pipeline object for extracting values that correspond to a named entity (e.g., birth dates, account numbers, or laboratory results)

Installation and usage

Install the library.

pip install extractacy

Import library and spaCy.

import spacy
from spacy.pipeline import EntityRuler
from extractacy.extract import ValueExtractor

Load spacy language model. Set up an EntityRuler for the example.

nlp = spacy.load("en_core_web_sm")
# Set up entity ruler
ruler = nlp.add_pipe("entity_ruler")
patterns = [
    {"label": "TEMP_READING", "pattern": [{"LOWER": "temperature"}]},
    {"label": "TEMP_READING", "pattern": [{"LOWER": "temp"}]},
    {
        "label": "DISCHARGE_DATE",
        "pattern": [{"LOWER": "discharge"}, {"LOWER": "date"}],
    },
    
]
ruler.add_patterns(patterns)

Define which entities you would like to link patterns to. Each entity needs 3 things:

patterns to search for (list). This relies on spaCy token matching syntax.
n_tokens to search around a named entity (int or sent)
direction (right, left, both)

# Define ent_patterns for value extraction
ent_patterns = {
    "DISCHARGE_DATE": {"patterns": [[{"SHAPE": "dd/dd/dddd"}],[{"SHAPE": "dd/d/dddd"}]],"n": 2, "direction": "right"},
    "TEMP_READING": {"patterns": [[
                        {"LIKE_NUM": True},
                        {"LOWER": {"IN": ["f", "c", "farenheit", "celcius", "centigrade", "degrees"]}
                        },
                    ]
                ],
                "n": "sent",
                "direction": "both"
        },
}

Add ValueExtractor to spaCy processing pipeline

nlp.add_pipe("valext", config={"ent_patterns":ent_patterns}, last=True)

doc = nlp("Discharge Date: 11/15/2008. Patient had temp reading of 102.6 degrees.")
for e in doc.ents:
    if e._.value_extract:
        print(e.text, e.label_, e._.value_extract)
        
## Discharge Date DISCHARGE_DATE 11/15/2008
## temp reading TEMP_READING 102.6 degrees