MPPSC 2020 Prelims Paper-I · Question 60 of 100

Correct — D, Collaborative Filtering. It is the technique used by recommender (recommendation) systems to predict what a user will like by collecting the tastes and preferences of many users. Its core assumption: if two users agreed on many items in the past, they are likely to agree again — so items liked by 'similar' users are recommended. It powers the 'customers who bought this also bought…' and 'recommended for you' features on Amazon, Netflix and YouTube.

• (a)Social Networking — Social networking is about building and maintaining online connections between people; it is not a method of automatically predicting a user's interests.
• (b)Social Targetting — Social targeting aims ads or content at users using their social-profile data; it is not the described automatic-prediction-from-preferences technique.
• (c)Collaborative Publishing — Collaborative publishing is many authors jointly producing content (e.g., a wiki) — it has nothing to do with predicting a user's interest.

Recommender systems predict user preferences. There are two main approaches: collaborative filtering, which uses patterns across many users' ratings/behaviour, and content-based filtering, which uses item attributes matched to a user's past likes. The definition given — automatic prediction by collecting preferences from users — is collaborative filtering.

All four options share 'social'/'collaborative' buzzwords — that is the trap. Match the definition ('predict a user's interest by collecting preferences') to collaborative FILTERING, not publishing, networking or targeting.

• Collaborative filtering predicts a user's interests from the collected preferences of many users.
• Key idea: users who agreed before will agree again (user-user or item-item similarity).
• It powers recommendations on Amazon, Netflix and YouTube.
• Contrast: content-based filtering uses item features, not other users' behaviour.

• Confusing collaborative filtering with collaborative publishing (wikis) or social networking
• Assuming it needs a social-network graph rather than users' preference data

IT sections give a definition and ask for the term ('predicting user interest from collected preferences' = collaborative filtering). Keep the four 'social/collaborative' terms distinct.

No directly related past PYQ was found.

• practice — not a real PYQ

  A recommender system that suggests items based on the attributes of items a user previously liked (rather than other users' behaviour) uses:

  • (a)Content-based filtering
  • (b)Collaborative filtering
  • (c)Social targeting
  • (d)Collaborative publishing

  Answer(a) Content-based filtering — it matches item features to the user's own history.

• practice — not a real PYQ

  'Customers who bought this also bought…' on e-commerce sites is a classic example of:

  • (a)Collaborative filtering
  • (b)Social networking
  • (c)Collaborative publishing
  • (d)Spam filtering

  Answer(a) Collaborative filtering — recommendations drawn from many users' buying patterns.