SVM for text classification in NLP - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

How does SVM handle text data?

Support Vector Machines (SVM) are used for text classification. How does SVM process text data before training?

A SVM directly uses raw text strings as input features without any transformation.

B SVM requires text to be translated into another language before training.

C SVM converts text into numerical vectors using techniques like TF-IDF or word embeddings before training.

D SVM uses the length of the text only as the feature for classification.

💡 Hint

Think about how computers understand text data for machine learning.

SVM cannot work with raw text strings. Text must be converted into numbers, often using methods like TF-IDF or word embeddings, so the model can process it.

❓ Predict Output

intermediate

Output of SVM prediction on sample text

Given the following Python code using sklearn's SVM for text classification, what is the printed output?

NLP

from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.svm import SVC

texts = ['I love apples', 'I hate bananas']
labels = [1, 0]

vectorizer = TfidfVectorizer()
X = vectorizer.fit_transform(texts)

model = SVC(kernel='linear')
model.fit(X, labels)

new_text = ['I love bananas']
X_new = vectorizer.transform(new_text)
prediction = model.predict(X_new)
print(prediction[0])

A 0

B 1

C Error due to unseen words in new_text

D Array with multiple predictions

💡 Hint

Consider how SVM predicts based on learned features and similarity.

The model learned that 'love' is associated with label 1. Even though 'bananas' was in the training data with label 0, the presence of 'love' influences the prediction to 1.

❓ Hyperparameter

advanced

Choosing the SVM kernel for text classification

Which kernel is generally best suited for SVM when classifying text data represented by TF-IDF vectors?

A Sigmoid kernel, because it mimics neural networks.

B Polynomial kernel, because text data requires complex curved boundaries.

C RBF kernel, because it handles non-linear data better than linear kernel.

D Linear kernel, because text data is often linearly separable in high-dimensional space.

💡 Hint

Think about the nature of TF-IDF vectors and their dimensionality.

Text data transformed by TF-IDF is usually high-dimensional and sparse, making it often linearly separable. The linear kernel is efficient and effective in this case.

❓ Metrics

advanced

Evaluating SVM model performance on imbalanced text data

You trained an SVM classifier on imbalanced text data. Which metric is most reliable to evaluate the model's performance?

A F1-score, because it balances precision and recall.

B Precision, because it measures how many predicted positives are correct.

C Recall, because it measures how many actual positives are found.

D Accuracy, because it shows overall correct predictions.

💡 Hint

Consider what happens when classes are imbalanced.

Accuracy can be misleading on imbalanced data. F1-score combines precision and recall, giving a better sense of model performance on minority classes.

🔧 Debug

expert

Why does this SVM training code raise an error?

Examine the code below. Why does it raise an error during training?

NLP

from sklearn.feature_extraction.text import CountVectorizer
from sklearn.svm import SVC

texts = ['good movie', 'bad movie', 'great film']
labels = [1, 0, 1]

vectorizer = CountVectorizer()
X = vectorizer.fit_transform(texts)

model = SVC(kernel='linear')
model.fit(X, labels)

A The labels list length does not match the number of text samples.

B The kernel parameter 'linear' is invalid.

C SVC requires labels to be strings, not integers.

D CountVectorizer cannot be used with SVM.

💡 Hint

Check the size of inputs and labels carefully.

The labels list has only 2 elements but there are 3 text samples. This mismatch causes a ValueError during model training.