02 - Intro to Recommender Systems

Question 1

What are examples of Recommender Systems?

Answer 1

• Popular (Netflix)
• Frequently bought together (Amazon)
• Customers who bought this item also bought (Amazon)
• Daily Mix (Spotify)

Question 2

What are examples of non-obvious Recommender Systems?

Answer 2

Google recommendations for locations and restaurants

Question 3

What are the the main types of recommendations?

Answer 3

• Rating prediction
• Ton-n ranking

Question 4

Are Recommender Systems also Machine Learning?

Answer 4

• Sometimes RecSys are based on standard ML algorithms (Tabular data, Regression, Classification)
• Many non-ML algorithms

Question 5

What is (much) more important in RecSys than typically in machine learning?

Answer 5

For RecSys online evaluations are (much) more important

Question 6

Why are Recommender Systems important?

Answer 6

• Recommender Systems are everywhere
• One of the most common fields to apply ML
• Strong impact on business, society, environment
• AutoML could really advance RecSys

Question 7

How could AutoML advance RecSys?

Answer 7

• Make Algorithm Selection and Tuning possible
• Ease the development of RecSys/ Make RecSys available for small businesses
• Reproducibility

Question 8

What types of data are there for recommender systems?

Answer 8

• Tabular
• Images (medicine or fashion)
• Text
• Time Series

Question 9

What domains of applications are there for recommender systems?

Answer 9

• Movies, Music
• Health
• Finance
• E-Commerce

Question 10

What is part of the developement process of RecSys?

Answer 10

• Recommendation Algorithms
• User Interfaces
• Architectures (APIs, high Performance, Realtime)
• Business Aspects (Money)
• Economy, society and environment
• Ethics (data privacy, discrimination)

Question 11

What recommendation approaches are there?

Answer 11

• Popularity based
• Collaborative Filtering
• Rating prediction
• Content-based Filtering
• Product based
• Link-based (Zitate oder Hyperlinks)
• Personalized

Question 12

On what criteria to decide for a recommender systems algorithm?

Answer 12

• Applicability
• Complexity
• Serendpity
• Privacy

Question 13

When can Collaborative Filterung work?

Answer 13

Collaborative Filtering only works for user with (many) ratings and items with at least one rating

Question 14

When can Popularity work?

Answer 14

Popularity works for every User but only for items with many ratings/views

Question 15

When can Content-based Filtering work?

Answer 15

Content-based Filtering works for every User and every Item

Question 16

What is the Algorithm Selection Problem?

Answer 16

• What algorithm to use for a specific problem?
• Why is there no algorithm for every problem?

Question 17

What are the benefits of popularity based recommendation approaches?

Answer 17

• Simple
• Uses few resources
• Works quite well
• Data privacy (not good for companies, they want information)

Question 18

What are product based recommendations?

Answer 18

• Frequently bought together
• User who bought … also bought