Back to Subject

Unit 6: Recommender Systems - Practice Quiz

INT423 — Machine Learning-Ii 50 Questions
0 Correct 0 Wrong 50 Left
0/50

1 What is the primary objective of a Recommender System?

A. To predict user preferences for items they have not yet interacted with
B. To reduce the dimensionality of a dataset
C. To cluster similar users without predicting ratings
D. To classify images

2 Which of the following best describes Content-Based Filtering?

A. It recommends items similar to those a user liked in the past based on item features.
B. It relies solely on demographic data.
C. It recommends items based on the preferences of similar users.
D. It uses matrix factorization to find latent features.

3 What is the 'Cold Start' problem in recommender systems?

A. The system overheats due to large data processing.
B. The difficulty in recommending items to a new user or recommending a new item due to lack of history.
C. The issue where popular items are recommended too often.
D. The time it takes to initialize the recommendation engine.

4 Which of the following represents 'Explicit Feedback'?

A. A user clicking on a product link.
B. A user watching a video for 10 minutes.
C. A user giving a movie a 5-star rating.
D. A user adding an item to a cart but not buying it.

5 Which of the following represents 'Implicit Feedback'?

A. Rating a song 4 out of 5.
B. Writing a text review.
C. Clicking on an advertisement.
D. Filling out a preference survey.

6 In Collaborative Filtering, what is the core underlying assumption?

A. Users' preferences change randomly over time.
B. Users who agreed in the past will tend to agree in the future.
C. The features of the items are more important than user interactions.
D. Items with similar descriptions are always rated similarly.

7 What is the purpose of Mean Normalization in collaborative filtering?

A. To increase the range of ratings.
B. To handle users who have not rated any items by treating missing ratings as the average.
C. To remove the effect of item features.
D. To convert binary labels into continuous variables.

8 In a binary label system (favs, likes, clicks), how is the target variable usually represented?

A. As a vector of text keywords.
B. As a discrete set {1, 2, 3, 4, 5}.
C. As 1 for interaction (positive) and 0 for no interaction.
D. As a continuous value between 0 and 1.

9 Which similarity measure is commonly used in Content-Based Filtering to compare document vectors?

A. Cosine Similarity
B. Euclidean Distance
C. Hamming Distance
D. Manhattan Distance

10 What is a major disadvantage of Content-Based Filtering?

A. It requires a large number of users to find similarities.
B. It tends to overspecialize and lacks serendipity (surprising recommendations).
C. It suffers from the cold start problem for new users.
D. It cannot handle binary data.

11 User-based Collaborative Filtering involves:

A. Finding items similar to the item the user is viewing.
B. Finding users who live in the same demographic area.
C. Filtering content based on keywords provided by the user.
D. Finding users similar to the target user and recommending what they liked.

12 Item-based Collaborative Filtering involves:

A. Calculating the similarity between items based on user co-ratings.
B. Clustering users into groups.
C. Using a decision tree to classify items.
D. Analyzing item descriptions to find keywords.

13 What is 'Matrix Factorization' in the context of recommender systems?

A. A technique to decompose the user-item interaction matrix into lower-dimensional latent factor matrices.
B. A method to sort the matrix by top-rated items.
C. A method to multiply two matrices to get the final ratings.
D. A way to normalize the mean of the matrix.

14 When using binary labels like 'clicks', which issue is most prominent compared to explicit ratings?

A. Data Scarcity
B. Ambiguity of negative feedback (missing data vs. dislike)
C. Lack of user identification
D. High computational cost

15 Which of the following is a key advantage of Hybrid Recommender Systems?

A. They only require implicit feedback.
B. They eliminate the need for data collection.
C. They can overcome limitations of individual approaches like the cold start problem.
D. They are computationally cheaper than simple algorithms.

16 In a Recommender System, 'Serendipity' refers to:

A. The consistency of recommendations over time.
B. The speed of the recommendation engine.
C. The ability to recommend items that are relevant but surprising to the user.
D. The accuracy of the prediction.

17 If a user has rated Item A (5 stars) and Item B (5 stars), and a second user rated Item A (5 stars), Item-Based CF would likely predict:

A. The second user will dislike Item B.
B. The second user will give Item B a high rating.
C. No prediction is possible.
D. The second user is a bot.

18 Which technique is best suited for a system where users rarely rate items but generate many search queries?

A. Matrix Factorization on explicit ratings
B. Demographic Filtering
C. User-based Collaborative Filtering
D. Content-Based Filtering

19 What is the 'Long Tail' phenomenon in recommender systems?

A. Recommendations are presented in a long list.
B. The algorithm takes a long time to converge.
C. The system requires long user IDs.
D. A small number of popular items generate most interactions, while many niche items have few interactions.

20 How does Mean Normalization help with bias?

A. It removes users who always give 1-star ratings.
B. It adjusts for users who are consistently harsh or generous in their ratings.
C. It converts all ratings to positive integers.
D. It ignores the item bias.

21 Which algorithm is commonly used to train Matrix Factorization models?

A. Decision Trees
B. Apriori Algorithm
C. K-Means Clustering
D. Alternating Least Squares (ALS)

22 A Weighted Hybrid Recommender System works by:

A. Running algorithms in a sequence where one refines the other.
B. Selecting one algorithm randomly.
C. Using content filtering only when collaborative filtering fails.
D. Combining the scores of different recommendation techniques with specific weights.

23 What is a 'Switching' Hybrid System?

A. It switches between positive and negative ratings.
B. It switches the user interface based on preferences.
C. It swaps the item ID with the user ID.
D. It chooses a recommendation technique based on the current situation (e.g., data availability).

24 In the context of binary labels, what is 'Confidence' often associated with?

A. The strength of the interaction (e.g., frequency of clicks or duration of view).
B. The percentage of items rated.
C. The confidence interval of the error.
D. The probability that the user is a human.

25 What is the primary input for a Content-Based Filtering algorithm?

A. Item Profiles (Features) and User Profiles.
B. Demographic data of all users.
C. Social Network Graphs.
D. A User-Item Rating Matrix.

26 Collaborative Filtering generally outperforms Content-Based Filtering in which scenario?

A. When there are no user ratings available.
B. When items have rich, structured metadata.
C. When recommending to a brand new user.
D. When identifying cross-genre or complex patterns that are hard to feature-engineer.

27 Which metric is commonly used to evaluate a Recommender System utilizing explicit ratings?

A. Accuracy
B. Jaccard Index
C. F1-Score
D. Root Mean Squared Error (RMSE)

28 What is the 'Grey Sheep' problem?

A. Users whose opinions do not consistently agree or disagree with any group of people.
B. The problem of duplicate accounts.
C. Items that are black and white images.
D. Users who only rate popular items.

29 In a user-item matrix used for CF, what does 'Sparsity' refer to?

A. The matrix has low rank.
B. The matrix is small in size.
C. Most entries in the matrix are empty (unknown ratings).
D. The ratings are all low numbers.

30 Which feature engineering technique is essential for Content-Based filtering of text documents?

A. Fourier Transform
B. Pixel normalization
C. Min-Max Scaling of IDs
D. TF-IDF (Term Frequency-Inverse Document Frequency)

31 Latent factors in Matrix Factorization usually represent:

A. Hidden characteristics inferred from data patterns.
B. Explicit categories like 'Action' or 'Comedy'.
C. The timestamp of the rating.
D. The user's age and location.

32 If a dataset consists only of 'Purchase' vs 'Non-Purchase', which type of filtering is applied?

A. Regression Analysis
B. Explicit Rating CF
C. Implicit Feedback CF
D. Sentiment Analysis

33 Why is Mean Normalization important when using Cosine Similarity for centered data (Pearson Correlation)?

A. It removes the user ID from the calculation.
B. It ensures all vectors have unit length.
C. It speeds up the computation.
D. It transforms the cosine similarity into Pearson Correlation Coefficient.

34 Which of the following is a limitation of Collaborative Filtering?

A. It yields recommendations that are too obvious.
B. It is strictly rule-based.
C. It requires domain knowledge to engineer features.
D. It cannot recommend items if no one else has rated them (New Item problem).

35 In a Cascade Hybrid System:

A. The system cascades into a random selection.
B. One recommender refines the recommendations given by another.
C. All recommenders run in parallel.
D. The weights of recommenders change dynamically.

36 Which strategy helps in solving the Cold Start problem for a new user?

A. Applying Matrix Factorization immediately.
B. Waiting for the user to rate 100 items.
C. Asking the user to select preferred genres during onboarding.
D. Using Item-Based Collaborative Filtering.

37 What is the primary advantage of Model-Based CF over Memory-Based CF?

A. It gives exact results based on neighbors.
B. It does not require training.
C. It handles sparsity better and offers faster prediction times.
D. It is easier to implement.

38 What does the 'Banana' problem refer to in Recommender Systems?

A. The shape of the loss function.
B. Users buying bananas only once.
C. Recommending items like bananas which are bought frequently but don't indicate distinct taste.
D. A coding error in Python.

39 Which loss function is minimized in standard Matrix Factorization for explicit ratings?

A. Log-Likelihood
B. Squared Error (between actual and predicted rating) + Regularization
C. Hinge Loss
D. Cross-Entropy Loss

40 How does a 'Demographic-based' recommender work?

A. It uses the age, gender, and location of users to find similar groups.
B. It uses only purchase history.
C. It uses the text content of reviews.
D. It uses satellite imagery.

41 Which of the following is an example of a use case for Association Rule Learning in recommendations?

A. Predicting the rating of a movie.
B. Face recognition.
C. Determining if an email is spam.
D. Market Basket Analysis (e.g., 'Frequently Bought Together').

42 Binary cross-entropy is a suitable loss function when:

A. Predicting the price of a house.
B. Predicting a probability of interaction (click/no-click).
C. Predicting a star rating from 1 to 5.
D. Clustering users.

43 What is the scalability challenge in User-Based Collaborative Filtering (Memory-Based)?

A. It cannot handle text data.
B. Computing similarity between millions of users in real-time is computationally expensive.
C. It only works with binary data.
D. It requires too much hard drive space.

44 In the context of Mean Normalization, if a user has not rated any movies, what prediction does the algorithm default to?

A. The maximum possible rating.
B. Zero.
C. A random number.
D. The average rating of the specific movie by other users.

45 What is 'Collaborative Filtering' distinct from?

A. Using machine learning.
B. Analyzing the internal attributes or content of the item.
C. Predicting future behavior.
D. Using matrices.

46 Regularization is added to the cost function in Matrix Factorization to:

A. Increase the number of latent features.
B. Ensure all ratings are positive.
C. Prevent overfitting by penalizing large values in the feature matrices.
D. Make the code run faster.

47 Which of these is a binary label?

A. Review Sentiment: Positive (0.8 score)
B. Favorite: Yes
C. View Duration: 120 seconds
D. Rating: 4.5 stars

48 Comparison: Which method requires domain knowledge for feature extraction?

A. Content-Based Filtering
B. Collaborative Filtering
C. Matrix Factorization
D. User-User KNN

49 Precision@k is a metric used to evaluate:

A. The exact numerical rating accuracy.
B. The number of users who rated k items.
C. The proportion of recommended items in the top-k set that are relevant.
D. The time taken to generate k recommendations.

50 In a Hybrid system, 'Feature Augmentation' refers to:

A. Adding more RAM to the server.
B. Using the output of one recommender as a feature input for another.
C. Adding random noise to the data.
D. Increasing the font size of recommendations.
Previous MCQ
Unit 5