11

Question 1

Question 1
What business value do recommender systems offer?
a) Sales assistance or dealing with information overload.
b) Improving the user experience.
c) Helping people find what they are looking for faster.
d) Finding interesting items that a user would not think of him/herself.
e) All of the above.

Answer 1

e) All of the above.

Question 2

Question 2
Which statement is NOT CORRECT?
a) The idea of up-selling is to sell more of a given product, usually at the time of purchase.
b) Cross-selling aims at selling an additional product or service.
c) Recommender systems aim at decreasing the so called hit, click through or lookers to bookers rates.
d) Down-selling means selling less of a product or service in order to maintain a sustainable, long-lasting customer relationship.

Answer 2

c) Recommender systems aim at decreasing the so called hit, click through or lookers to bookers rates.

Question 3

Question 3
Companies like Amazon, Netflix, YouTube, etc. usually
a) disclose how their recommendation system works.
b) do not disclose how their recommendation system works.

Answer 3

b) do not disclose how their recommendation system works.

Question 4

Question 4
The output of a recommendation system can be:
a) a relevance score, denoting the relevance of a product or service to a particular user.
b) a top N ranking where the top N most interesting products or services to a user are listed.
c) any of the above.

Answer 4

c) any of the above.

Question 5

Question 5
Unpersonalized recommendations can be based on
a) popularity.
b) novelty.
c) promotions.
d) all of the above.

Answer 5

d) all of the above.

Question 6

Question 6
The cold start problem implies that
a) it is hard to measure the performance of a recommendation system.
b) it is hard to provide good recommendations for new users or items.
c) the entire recommendation process is non-stationary.
d) every person’s preferences evolve in time, space and context.

Answer 6

b) it is hard to provide good recommendations for new users or items.

Question 7

Question 7
Recommendation systems can be evaluated in terms of
a) accuracy.
b) diversity.
c) novelty.
d) fairness.
e) all of the above.

Answer 7

e) all of the above.

Question 8

Question 8
Explicit user interest implies that the user is
a) aware that he or she is conveying an opinion.
b) aware that he or she is conveying an opinion.

Answer 8

a) aware that he or she is conveying an opinion.

Question 9

Question 9
Messaging with the customer service desk to ask for more specific product information is an example of
a) explicit user interest.
b) implicit user interest.

Answer 9

b) implicit user interest.

Question 10

Question 10
When compared to implicit user interest, explicit user interest is
a) more noisy and less robust.
b) more robust and less noisy.

Answer 10

b) more robust and less noisy.

Question 11

Question 11
Which is a typical problem when dealing with real-life rating matrices?
a) scalability.
b) sparsity.
c) rating bias.
d) long tail distribution.
e) all of the above.

Answer 11

e) all of the above.

Question 12

Question 12
Recommender systems have an intrinsic bias to give recommendations for the
a) popular items.
b) unpopular items.

Answer 12

a) popular items.

Question 13

Question 13
Which statement is NOT CORRECT?
a) The core idea of collaborative filtering is to give recommendations based on what other similar customers from the community have shown interest in. Hence, no user or item descriptive features are used.
b) Content filtering uses the user profile, contextual data and item features. This is then combined with other user’s ratings.
c) Knowledge based filtering makes use of user profile data, contextual data, item features and knowledge models.
d) Hybrid filtering uses all sources of data to make the recommendation: user profile data, contextual data, community data, item features and knowledge models.

Answer 13

b) Content filtering uses the user profile, contextual data and item features. This is then combined with other user’s ratings.

Question 14

Question 14
The goal of recommender system can be:
a) prediction.
b) ranking.
c) both.

Answer 14

c) both.

Question 15

Question 15
When evaluating recommender systems, the test set
a) cannot be used during model development.
b) can be used during model development.

Answer 15

a) cannot be used during model development.

Question 16

Question 16
When evaluating the prediction power of a recommendation system, we are aiming for
a) a high RMSE, a low MAD and a high Pearson correlation.
b) a low RMSE, a low MAD and a high Pearson correlation.
c) a low RMSE, a high MAD and a high Pearson correlation.
d) a low RMSE, a low MAD and a low Pearson correlation.

Answer 16

b) a low RMSE, a low MAD and a high Pearson correlation.

Question 17

Question 17
Which statement is CORRECT?
a) The more correct recommendations in the top k, the larger the average precision AP@k.
b) The average precision AP@k rewards for front-loading recommendations that are correct.
c) The average precision AP@k will never penalize for adding more recommendations.
d) All statements are correct.

Answer 17

d) All statements are correct.

Question 18

Question 18
Which statement is NOT CORRECT?
a) Spearman’s rank order correlation always ranges between -1 which represents perfect disagreement and +1 which represents perfect agreement.
b) Kendall’s tau is 1 for perfect agreement and -1 for perfect disagreement.
c) The Goodman-Kruskal 𝛾 will only consider the tied pairs.
d) The Goodman-Kruskal 𝛾 is +1 if there are no discordant pairs so perfect agreement, -1 if there are no concordant pairs so perfect disagreement.

Answer 18

c) The Goodman-Kruskal 𝛾 will only consider the tied pairs.

Question 19

Question 19
Which statement is CORRECT?
a) Serendipity measures how surprising are the successful recommendations. It is also sometimes referred to as the deviation from the natural prediction.
b) Scalability measures how well the system scales to a high number of users and items. It can be decomposed into offline model construction time, latency or the time from requesting a recommendation until the system responds, and throughput, or the number of recommendations per time unit.
c) The uplift effect captures whether the recommendation was actually necessary. In other words, it makes no sense to recommend an item the user was going to buy away.
d) All statements are correct.

Answer 19

d) All statements are correct.

Question 20

Question 20
Collaborative filtering methods
a) use the rating matrix to make a prediction.
b) don’t use the rating matrix to make a prediction.

Answer 20

a) use the rating matrix to make a prediction.

Question 21

Question 21
The key underlying assumption of collaborative filtering is that users who had similar tastes in the past,
a) will have different tastes in the future.
b) will have similar tastes in the future.

Answer 21

b) will have similar tastes in the future.

Question 22

Question 22
The Pearson correlation over 2 pairs of items is
a) 0.
b) -1.
c) +1.
d) either +1 or -1.

Answer 22

d) either +1 or -1.

Question 23

Question 23
A high cosine measures implies
a) high similarity.
b) low similarity.

Answer 23

a) high similarity.

Question 24

Question 24
A key characteristic of the Pearson and cosine similarity is that
a) differences in average rating behavior of the users are not considered.
b) differences in average rating behavior of the users are considered.

Answer 24

a) differences in average rating behavior of the users are not considered.

Question 25

Question 25
The adjusted cosine measure
a) does not take into account rating bias.
b) takes into account rating bias.

Answer 25

b) takes into account rating bias.

Question 26

Question 26
The Jaccard index can be used for
a) only binary data.
b) only continuous data.
c) both binary and continuous data.

Answer 26

a) only binary data.

Question 27

Question 27
The key idea of item-item collaborative filtering is to apply the k-nearest neighbor idea and look for
a) similar items.
b) similar users.
c) similar items and users.

Answer 27

a) similar items.

Question 28

Question 28
According to research, item similarities are considered to be
a) less stable than user similarities.
b) more stable than user similarities.

Answer 28

b) more stable than user similarities.

Question 29

Question 29
Which statement is NOT CORRECT?
a) K-nearest neighbor based filtering methods assume a minimum number of ratings for new users and items in order to work well.
b) K-nearest neighbor based filtering methods are also known to suffer from the popularity bias which implies that items liked by more users will have a higher chance of being recommended.
c) K-nearest neighbor based filtering methods do not scale well for most real-world scenarios.
d) The Geuens et al. study illustrated that user-based collaborative filtering methods always achieve better accuracy than item-based collaborative filtering methods.

Answer 29

d) The Geuens et al. study illustrated that user-based collaborative filtering methods always achieve better accuracy than item-based collaborative filtering methods.

Question 30

Question 30
In content filtering, the recommendations are based on
a) the content of items.
b) other user’s opinions.

Answer 30

a) the content of items.

Question 31

Question 31
Content filtering recommendation systems start by building
a) item profiles for each user.
b) item profiles for each item.

Answer 31

b) item profiles for each item.

Question 32

Question 32
In content filtering, the item profile can be constructed as the set of features with the
a) smallest TF-IDF scores together with their scores.
b) highest TF-IDF scores together with their scores.

Answer 32

b) highest TF-IDF scores together with their scores.

Question 33

Question 33
Which statement is CORRECT?
a) In content filtering, a user profile can be constructed as the average of the rated item profiles.
b) In content filtering, given a user profile and an item profile, we can then estimate the utility or possible interest using the cosine similarity measure between both.
c) In content filtering, we can also use a classification algorithm such as a decision tree or logistic regression to make recommendations.
d) All statements are correct.

Answer 33

d) All statements are correct.

Question 34

Question 34
Which statement is CORRECT?
a) In content based filtering, there is a cold start problem for items but not for users.
b) In content based filtering, there is a cold start problem for users but not for items.
c) In content based filtering, there is a cold start problem for both items and users.
d) In content based filtering, there is no cold start problem.

Answer 34

b) In content based filtering, there is a cold start problem for users but not for items.

Question 35

Question 35
In content based filtering, there is
a) a sparsity problem.
b) no sparsity problem.

Answer 35

b) no sparsity problem.

Question 36

Question 36
Which statement is CORRECT.
a) A disadvantage of content filtering is that the tagging or identification of the key features can get quite intensive, especially when it concerns images, movies or music.
b) In content filtering, there is a risk of overspecialization by never recommending items outside a user’s content profile.
c) Both statements are correct.

Answer 36

c) Both statements are correct.