lecture 12 Flashcards
What is an example of a common recommender system application?
Movie recommendation (e.g., Netflix).
What major event in 2006 sparked interest in recommender systems?
Netflix’s $1M prize for improving their rating prediction model.
What is explicit feedback in recommender systems?
Direct user ratings for items.
What is collaborative filtering?
A method where users provide ratings to help filter recommendations.
What is the main drawback of explicit feedback?
Data sparsity—users only rate a small number of items.
Which company was one of the first to use personalized recommendations?
Amazon.
How does Amazon’s recommendation system work?
It predicts what users will like based on their past interactions.
What are other use cases for recommender systems?
News article recommendations, product recommendations, and friend suggestions.
What is the abstract task of recommendation?
Predicting relationships between two sets of objects.
What types of relations can be used in recommendation models?
Binary (e.g., like/dislike) or numeric (e.g., ratings).
What is an example of a binary recommendation relation?
A user following another user on social media.
What is an example of a numeric recommendation relation?
A user rating a movie on a scale of 1 to 5.
How do recommendation algorithms affect social media?
They determine what content users see based on past behavior.
What are potential negative effects of recommendation systems?
Filter bubbles, bias reinforcement, and increased polarization.
Why are recommendation algorithms a key component of social media platforms?
They optimize engagement by showing personalized content.
What is the structure of a numeric recommendation problem?
A matrix with users as rows, items as columns, and ratings as values.
What is an embedding model in recommendation systems?
A model that represents users and items as vectors in a shared space.
Why are embedding models useful for recommendations?
They allow learning of hidden patterns in user-item interactions.
What is the key idea behind matrix factorization?
Decomposing a rating matrix into two smaller matrices for users and items.
What does the user embedding vector represent?
A user’s preferences in a lower-dimensional space.
What does the item embedding vector represent?
An item’s characteristics in the same lower-dimensional space.
What mathematical operation is commonly used to predict ratings?
The dot product of user and item embeddings.
What is the goal of optimizing embeddings?
Minimizing the difference between predicted and actual ratings.
What is alternating least squares (ALS)?
An optimization method where user and item embeddings are updated alternately.
What is stochastic gradient descent (SGD) used for in embeddings?
To iteratively adjust embeddings based on prediction errors.
What is a challenge in recommender systems?
Handling missing values in the rating matrix.
What is a common way to deal with missing ratings?
Optimizing the model only on known ratings.
What is a major advantage of recommender systems?
They improve user experience by providing personalized content.
What is a takeaway from embedding models?
They transform users and items into a shared numerical space for better predictions.
