Latent Dirichlet Allocation (LDA) in NLP - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What does the 'topic' represent in LDA?

In Latent Dirichlet Allocation, what does a 'topic' most accurately represent?

A A single word that best describes a document

B A fixed label assigned to each document

C A cluster of documents grouped by similarity

D A distribution over words showing which words are likely to appear together

💡 Hint

Think about how LDA models topics as probabilities over vocabulary.

In LDA, each topic is a probability distribution over words, indicating which words are likely to co-occur in that topic.

❓ Predict Output

intermediate

Output of LDA topic distribution for a document

Given the following Python code using sklearn's LDA, what is the shape of doc_topic_dist?

NLP

from sklearn.decomposition import LatentDirichletAllocation
from sklearn.feature_extraction.text import CountVectorizer

docs = ["apple banana apple", "banana orange banana", "apple orange orange"]
vectorizer = CountVectorizer()
dtm = vectorizer.fit_transform(docs)
lda = LatentDirichletAllocation(n_components=2, random_state=0)
lda.fit(dtm)
doc_topic_dist = lda.transform(dtm)
print(doc_topic_dist.shape)

A (3, 2)

B (2, 3)

C (3, 3)

D (2, 2)

💡 Hint

Check how many documents and topics are in the model.

The transform method returns the topic distribution for each document. There are 3 documents and 2 topics, so the shape is (3, 2).

❓ Model Choice

advanced

Choosing the number of topics in LDA

You want to model topics in a large collection of news articles using LDA. Which approach is best to decide the number of topics?

A Use domain knowledge and try multiple values, then select based on coherence or perplexity scores

B Always set the number of topics to 10 by default

C Set the number of topics equal to the number of documents

D Use the number of unique words as the number of topics

💡 Hint

Think about how to balance model complexity and interpretability.

Choosing the number of topics is often done by testing multiple values and evaluating metrics like coherence or perplexity, combined with domain knowledge.

❓ Metrics

advanced

Interpreting LDA perplexity score

After training an LDA model, you get a perplexity score of 1200 on your test set. What does a lower perplexity score indicate?

A The model is overfitting the training data

B The model predicts the test data better, indicating better generalization

C The model has fewer topics

D The model is ignoring rare words

💡 Hint

Perplexity measures how well the model predicts unseen data.

A lower perplexity means the model better predicts the test data, showing better generalization.

🔧 Debug

expert

Identifying error in LDA code snippet

What error will this code raise?

NLP

from sklearn.decomposition import LatentDirichletAllocation
from sklearn.feature_extraction.text import CountVectorizer

docs = ["cat dog", "dog mouse", "cat mouse"]
vectorizer = CountVectorizer()
dtm = vectorizer.fit_transform(docs)
lda = LatentDirichletAllocation(n_components=3, random_state=0)
lda.fit_transform(dtm)
print(lda.components_.shape)

A TypeError: fit_transform() missing 1 required positional argument

B AttributeError: 'LatentDirichletAllocation' object has no attribute 'components_'

C (3, 3)

D ValueError: n_components must be less than or equal to number of features

💡 Hint

Check the shape of the document-term matrix and the number of topics.

The document-term matrix has 3 features (unique words). The model has 3 topics, so components_ shape is (3, 3). No error occurs.