Why NER extracts structured information in NLP - Challenge Your Understanding

Challenge - 5 Problems

🧠 Conceptual

intermediate

Why does Named Entity Recognition (NER) extract structured information?

NER is used to find specific pieces of information in text, like names or dates. Why is this considered extracting structured information?

A Because NER converts unorganized text into labeled categories like person, location, or date, making data easier to analyze.

B Because NER translates text into another language to structure it.

C Because NER removes all punctuation to create a clean text format.

D Because NER summarizes the entire text into a short paragraph.

💡 Hint

Think about how NER tags parts of text with labels that computers can understand easily.

NER tags words or phrases with categories like 'person' or 'date'. This labeling turns messy text into organized data, which is structured information.

❓ Predict Output

intermediate

Output of NER entity extraction code

What is the output of this Python code using spaCy to extract entities?

NLP

import spacy
nlp = spacy.load('en_core_web_sm')
doc = nlp('Apple was founded by Steve Jobs in California.')
entities = [(ent.text, ent.label_) for ent in doc.ents]
print(entities)

A [('Apple', 'PERSON'), ('Steve Jobs', 'ORG'), ('California', 'LOC')]

B [('Apple', 'LOC'), ('Steve Jobs', 'GPE'), ('California', 'PERSON')]

C [('Apple', 'GPE'), ('Steve Jobs', 'PERSON'), ('California', 'ORG')]

D [('Apple', 'ORG'), ('Steve Jobs', 'PERSON'), ('California', 'GPE')]

💡 Hint

Remember that 'Apple' is a company (organization), 'Steve Jobs' is a person, and 'California' is a geopolitical entity.

spaCy labels 'Apple' as an organization (ORG), 'Steve Jobs' as a person (PERSON), and 'California' as a geopolitical entity (GPE).

❓ Model Choice

advanced

Choosing the best model for NER on noisy social media text

You want to extract structured information from tweets that contain slang, misspellings, and emojis. Which model is best suited for this NER task?

A A rule-based NER system using fixed dictionaries

B A simple logistic regression model trained on formal news articles

C A pre-trained BERT model fine-tuned on social media NER datasets

D A clustering algorithm that groups similar words without labels

💡 Hint

Consider which model can understand context and adapt to informal language.

BERT models fine-tuned on social media data can handle slang and misspellings better than fixed rules or simple models.

❓ Metrics

advanced

Evaluating NER model performance with F1 score

An NER model predicted 80 entities correctly, missed 20 entities, and predicted 10 entities incorrectly. What is the F1 score?

A 0.84

B 0.80

C 0.75

D 0.88

💡 Hint

Calculate precision and recall first, then use F1 = 2 * (precision * recall) / (precision + recall).

Precision = 80 / (80 + 10) = 0.8889, Recall = 80 / (80 + 20) = 0.8, F1 = 2 * 0.8889 * 0.8 / (0.8889 + 0.8) ≈ 0.842.

🔧 Debug

expert

Why does this NER model fail to extract entities from new domain text?

You trained an NER model on news articles but it performs poorly on medical reports. What is the most likely reason?

A The model architecture is incorrect and cannot process text longer than 100 words.

B The model was trained on a different domain and cannot generalize well to medical terms.

C The training data had too many entities, causing overfitting.

D The model uses a wrong loss function that ignores entity labels.

💡 Hint

Think about how domain differences affect model understanding.

Models trained on one domain often fail on very different domains because vocabulary and entity types differ, causing poor generalization.