Content-based filtering in ML Python - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

How does content-based filtering recommend items?

Which of the following best describes how content-based filtering recommends items to a user?

A It recommends items similar to those the user liked before based on item features.

B It recommends items liked by other users with similar tastes.

C It randomly recommends popular items regardless of user preferences.

D It recommends items based on the time of day and user location.

💡 Hint

Think about how the system uses the user's past preferences and item details.

Content-based filtering uses the features of items the user liked to find similar items to recommend.

❓ Predict Output

intermediate

Output of cosine similarity calculation

What is the output of this Python code that calculates cosine similarity between two item feature vectors?

ML Python

from sklearn.metrics.pairwise import cosine_similarity
import numpy as np

item1 = np.array([[1, 0, 1, 0]])
item2 = np.array([[0, 1, 1, 0]])
similarity = cosine_similarity(item1, item2)
print(round(similarity[0][0], 2))

A 0.33

B 0.50

C 0.67

D 0.75

💡 Hint

Recall cosine similarity formula and count overlapping features.

Cosine similarity is the dot product divided by the product of vector lengths. Here it equals 0.5.

❓ Model Choice

advanced

Best model for content-based filtering with sparse user-item data

Which model is most suitable for content-based filtering when user-item interaction data is sparse but item features are rich?

A Collaborative filtering using user similarity

B Matrix factorization using only user-item ratings

C K-Nearest Neighbors using item feature vectors

D Random forest classifier on user demographic data

💡 Hint

Focus on models that use item features directly.

K-Nearest Neighbors on item features works well when user-item data is sparse but item features are available.

❓ Hyperparameter

advanced

Choosing the number of neighbors in content-based KNN

In a content-based filtering system using K-Nearest Neighbors on item features, what is the effect of increasing the number of neighbors (k)?

A It increases recommendation diversity but may reduce relevance.

B It always improves recommendation accuracy.

C It decreases the number of recommended items.

D It causes the model to ignore item features.

💡 Hint

Think about how more neighbors affect similarity and variety.

Increasing k includes more items that are less similar, increasing diversity but possibly lowering relevance.

🔧 Debug

expert

Why does this content-based filtering code produce identical recommendations for all users?

Given this simplified content-based filtering code, why do all users get the same recommended items? ```python import numpy as np user_profiles = { 'user1': np.array([1, 0, 1]), 'user2': np.array([1, 0, 1]), 'user3': np.array([1, 0, 1]) } item_features = { 'itemA': np.array([1, 0, 0]), 'itemB': np.array([0, 1, 1]), 'itemC': np.array([1, 0, 1]) } recommendations = {} for user, profile in user_profiles.items(): scores = {} for item, features in item_features.items(): scores[item] = np.dot(profile, features) recommended = sorted(scores, key=scores.get, reverse=True)[:2] recommendations[user] = recommended print(recommendations) ```

A The code uses Euclidean distance instead of dot product.

B The item features are normalized incorrectly causing identical scores.

C The sorting function is not stable, causing random recommendations.

D All user profiles are identical, so dot products and recommendations are the same.

💡 Hint

Check the user profiles carefully.

Since all user profiles are the same vector, their dot products with items are identical, leading to identical recommendations.