RegEx with spaCy

Natural Language Processing with spaCy

Azadeh Mobasher

Principal Data Scientist

What is RegEx?

Rule-based information extraction (IR) is useful for many NLP tasks
Regular expression (RegEx) is used with complex string matching patterns
RegEx finds and retrieves patterns or replace matching patterns

RegEx: link and phone information extraction

RegEx strengths and weaknesses

Pros:

Enables writing robust rules to retrieve information
Can allow us to find many types of variance in strings
Runs fast
Supported by programming languages

Cons:

Syntax is challenging for beginners
Requires knowledge of all the ways a pattern may be mentioned in texts

RegEx in Python

Python comes prepackaged with a RegEx library, re.
The first step in using re package is to define a pattern.
The resulting pattern is used to find matching content.

import re

pattern = r"((\d){3}-(\d){3}-(\d){4})"
text = "Our phone number is 832-123-5555 and their phone number is 425-123-4567."

RegEx in Python

We use .finditer() method from re package

iter_matches = re.finditer(pattern, text)

for match in iter_matches:
    start_char = match.start()
    end_char = match.end()

    print ("Start character: ", start_char, "| End character: ", end_char, 
           "| Matching text: ", text[start_char:end_char])

>>> Start character:  20 | End character:  32 | Matching text:  832-123-5555
Start character:  59 | End character:  71 | Matching text:  425-123-4567

RegEx in spaCy

RegEx in three pipeline components: Matcher, PhraseMatcher and EntityRuler.

text = "Our phone number is 832-123-5555 and their phone number is 425-123-4567."

nlp = spacy.blank("en")
patterns = [{"label": "PHONE_NUMBER", "pattern": [{"SHAPE": "ddd"},
            {"ORTH": "-"}, {"SHAPE": "ddd"},
            {"ORTH": "-"}, {"SHAPE": "dddd"}]}]

ruler = nlp.add_pipe("entity_ruler")
ruler.add_patterns(patterns)
doc = nlp(text)
print ([(ent.text, ent.label_) for ent in doc.ents])

>>> [('832-123-5555', 'PHONE_NUMBER'), ('425-123-4567', 'PHONE_NUMBER')]

Let's practice!

Natural Language Processing with spaCy