Final Exam: Recommender Systems Flashcards
What is a recommender system?
An information filtering system that provides suggestions for items that are most pertinent to a particular user. An important part of the information and e-commerce ecosystem.
What is the value of a recommender system?
An important part of the information and e-commerce ecosystem.
What are the types of recommender systems?
- Content-based filtering
- Collaborative filtering
- Hybrid approach
What is the breakdown of collaborative filtering?
- Model-based filtering (Bayesian Networks and Neural Networks)
- Memory-base filtering (Item-based approach and User-based approach)
What is an item profile?
A profile that contains the description or features of a collection of similar items.
What is a user profile?
A profile that contains all the item profiles whose items are rated positively by the user.
What are the drawbacks of content-based recommender systems?
- Require in-depth knowledge of the item features
- Nontrivial aggregation
- Limited capacity to expand on the users’ existing choices or interests
What are the advantages of content-based recommender systems?
- Quickly adapting to user’s dynamic preferences
- Does not rely on other users’ data
- Ensure security and privacy of user data
- Overcome the cold-start problem existing in the collaborative algorithms
- Common in personalized news, publications, web page recommendations, etc.
What is collaborative filtering-based recommender systems?
Use the similarity between users or the similarity between items. Items liked by similar users will be recommended to the user.
What are the drawbacks of collaborative filtering-based recommender systems?
- Nontrivial and time-consuming to find the neighborhood of a target user
- Rely on a LARGE data set of users’ activities
- Cold-start problem
- Privacy concerns
What are the advantages of collaborative filtering-based recommender systems?
- Does not require item features
- Ability to expand on the user’s existing interests by discovering new items
- Flexible algorithms
What is memory-based recommendation?
A conversion from a utility matrix to a recommendation.
1. The target user’s profile must exist in the utility matrix
2. Item-based, user-based
What is a model-based recommendation?
A recommendation utilized to predict the user’s rating for an unrated item.
1. The target user’s profile does not necessarily exist in the training process
What is key across both user-based filtering and item-based collaborative filtering?
Similarity Computation
What is an example of a similarity computation?
Pearson Correlation
What is a cold start?
When insufficient data is utilized to draw meaningful inference.
Remedy: Incentivize users to provide feedback; collect additional data about user and items
What is a shilling attack?
When fake data is implemented into a database to increase or decrease some items’ popularity.
Remedy: Monitor abnormalities of user behavior
What is a latency problem (lack of exploration)?
When new items are frequently inserted into the database, but the system fails to explore.
Remedy: Design algorithms to encourage the exploration of new items.
What is the sparsity problem?
When there are too many NaNs in the input data.
Remedy: Design hybrid approaches, potentially model-based methods.
What is scalability?
When some algorithms are hard to scale to large amounts of data.
Remedy: Dimension reduction; clusters of data.
What is the formula for Pearson’s Correlation?
Pxy = corr(x,y) = E[(X-Mean of x)(Y-Mean of y)]/SDy x SDx
