7: Recommender Systems

Question 1

What is a recommender system ?

Answer 1

A system which attempts to recommend information items likely to be of interest to the users.

Eg movies, TV, music, books, news, jokes, by Netflix or Amazon.

Question 2

What are the two ways to collect data from users ?

Answer 2

Explicit:

• Ask a user to rate an item
• Ask a user to create a (wish) list

Implicit:

• Keep a record of the items a user searches/views/purchases
• Analyse the user’s social network to discover likes and dislikes

Question 3

What other data can be used ?

Answer 3

• Domain-based knowledge (category)
• Patterns and relations
• Higher level context

Question 4

Types recommendation algorithms ?

Answer 4

• Popularity-based
• Demographic based
• Content based

Question 5

What is item co-occurrence ?

Answer 5

‘People who like this also like …’

• For each pair of items, count how often they are bought together (slow, but offline)
• For each item that is being purchased, rank the other items by their co-occurrence counts.
• Recommend the items from the top pairs that were bought together in the past.

Question 6

Content-based recommendation ?

Answer 6

Offers similar products with similar features, eg another computer with higher RAM and it costs about the same.

• Hard to guess which features are important for particular user
• Hard to compare different product features in different product domains

Question 7

Collaborative Filtering ?

Answer 7

• Used for spam control

Eg is an email address is blacklisted by one user, future messages can be blocked for all users.

Question 8

Give 2 types of CF ?

Answer 8

• User-based CF: find the most similar users and calculate their average rating for the new item
• Item-based CF: find the most similar items and calculate their current user’s average ratings.

Question 9

What are the advantages and disadvantages with CF ?

Answer 9

• Pros: simplicity, bc no content analysis is necessary
• Cons:
• • Sparsity (not enough data)
• • Cold-start (no past similar purchases)
• • Users may get recommendations for low quality products bought by friends

Question 10

What factors should be taken into account for designing recommendation systems ?

Answer 10

• Explanations
• Minimum nonsense recommendations
• Privacy
• Spam recommender
• No vandalism/offensive/explicit content

Question 11

What data to collect and how ?

Answer 11

• Find ways to collect as much input as possible without being disruptive
• A lot of data that can train a system: votes, clicks, page-view time, purchases, tagging, adding a title.

Question 12

What’s more important between:

• Data collection algorithm
• Data collected

Answer 12

Data collected