Information Extraction (IE)

Information Extraction (IE) is the task of automatically pulling structured information (entities, relationships, facts) out of unstructured text. It builds directly on POS tagging, syntactic parsing, and NER from Unit 2.

Unstructured: "Riya joined Google as a Software Engineer on 5 August 2026."

Structured:
{
  "person": "Riya",
  "organization": "Google",
  "role": "Software Engineer",
  "start_date": "2026-08-05"
}

Core Sub-tasks of Information Extraction

Relation Extraction — Rule-Based (Pattern Matching)

import re

text = "Riya works at Google. Aman is employed by Microsoft."
pattern = r'(\w+) (?:works at|is employed by) (\w+)'
matches = re.findall(pattern, text)
print(matches)
# [('Riya', 'Google'), ('Aman', 'Microsoft')]

Relation Extraction Using Dependency Parsing (Recall Unit 2)

Dependency parses expose grammatical relations (subject, object) that can be turned into (entity, relation, entity) triples.

import spacy

nlp = spacy.load("en_core_web_sm")
doc = nlp("Riya founded a startup in 2024.")

for token in doc:
    if token.dep_ == "nsubj":
        subject = token.text
    if token.dep_ == "dobj":
        obj = token.text
    if token.pos_ == "VERB":
        verb = token.text

print(f"({subject}, {verb}, {obj})")
# (Riya, founded, startup)

Combining NER + Pattern Matching for Structured Extraction

import spacy
import re

nlp = spacy.load("en_core_web_sm")
text = "Riya joined Google as a Software Engineer on 5 August 2026."
doc = nlp(text)

person = [ent.text for ent in doc.ents if ent.label_ == "PERSON"]
org = [ent.text for ent in doc.ents if ent.label_ == "ORG"]
date = [ent.text for ent in doc.ents if ent.label_ == "DATE"]

role_match = re.search(r'as an? ([\w\s]+?) on', text)
role = role_match.group(1) if role_match else None

record = {
    "person": person[0] if person else None,
    "organization": org[0] if org else None,
    "role": role,
    "date": date[0] if date else None
}
print(record)
# {'person': 'Riya', 'organization': 'Google', 'role': 'Software Engineer', 'date': '5 August 2026'}

Real-World Applications of Information Extraction

Information Extraction is often the final structured output of an entire NLP pipeline — feeding chatbots, search engines, and analytics dashboards with clean, queryable facts pulled from free text.