Document-term matrix in NLP - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output of Document-Term Matrix Creation

What is the output of the following code that creates a document-term matrix from two simple documents?

NLP

from sklearn.feature_extraction.text import CountVectorizer
corpus = ['apple orange apple', 'orange banana orange']
vectorizer = CountVectorizer()
X = vectorizer.fit_transform(corpus)
print(X.toarray())

A

[[1 2 0]
 [0 1 2]]

B

[[1 1 1]
 [1 1 1]]

C

[[2 0 1]
 [0 1 2]]

D

[[2 1 0]
 [0 2 1]]

💡 Hint

CountVectorizer counts how many times each word appears in each document.

The vectorizer finds three unique words: 'apple', 'banana', and 'orange'. The first document has 'apple' twice and 'orange' once, so its vector is [2, 0, 1]. The second document has 'orange' twice and 'banana' once, so its vector is [0, 1, 2]. However, the order of words is alphabetical: 'apple', 'banana', 'orange'. So the matrix is [[2, 0, 1], [0, 1, 2]]. The printed output matches option C.

🧠 Conceptual

intermediate

Understanding Document-Term Matrix Dimensions

If you create a document-term matrix from 5 documents containing a total of 100 unique words, what will be the shape (rows, columns) of the matrix?

A 100 rows and 5 columns

B 100 rows and 100 columns

C 5 rows and 5 columns

D 5 rows and 100 columns

💡 Hint

Rows represent documents, columns represent unique words.

A document-term matrix has one row per document and one column per unique word found in all documents. So with 5 documents and 100 unique words, the matrix shape is (5, 100).

❓ Metrics

advanced

Choosing the Right Metric for Document-Term Matrix Similarity

Which metric is most appropriate to measure similarity between two document vectors from a document-term matrix when the goal is to find documents with similar topics regardless of length?

A Cosine similarity

B Euclidean distance

C Manhattan distance

D Jaccard index

💡 Hint

Consider a metric that ignores vector length and focuses on direction.

Cosine similarity measures the angle between two vectors, ignoring their length. This is ideal for document vectors where length varies but topic similarity depends on word distribution.

🔧 Debug

advanced

Identifying the Error in Document-Term Matrix Code

What error will the following code raise? from sklearn.feature_extraction.text import CountVectorizer corpus = ['cat dog', 'dog mouse'] vectorizer = CountVectorizer() X = vectorizer.fit_transform(corpus) print(X[0, 1])

A No error, prints the count of the second word in the first document

B TypeError: 'csr_matrix' object is not subscriptable

C AttributeError: 'CountVectorizer' object has no attribute 'fit_transform'

D IndexError: index out of range

💡 Hint

csr_matrix supports indexing with X[i, j].

No error is raised. X is a csr_matrix, which supports direct indexing like X[0, 1]. Here, the vocabulary is ['cat', 'dog', 'mouse'] (alphabetical order), so X[0, 1] is the count of 'dog' in the first document, which is 1.

❓ Model Choice

expert

Best Model to Use with Document-Term Matrix for Text Classification

Given a document-term matrix representing text data, which machine learning model is generally most suitable for classifying documents into categories when the data is high-dimensional and sparse?

A K-Nearest Neighbors (KNN)

B Support Vector Machine (SVM) with linear kernel

C Decision Tree

D Naive Bayes

💡 Hint

Consider models that handle high-dimensional sparse data well and avoid overfitting.

SVM with a linear kernel works well on high-dimensional sparse data like document-term matrices. It finds a hyperplane that separates classes effectively and handles sparsity better than KNN or Decision Trees. Naive Bayes is also common but SVM often performs better in complex text classification.