Testing a Recommender System for Self-Actualization · 2017-07-04 · Testing a Recommender System...
26
Testing a Recommender System for Self-Actualization Daricia Wilkinson, Saadhika Sivakumar , Pratitee Sinha, Bart P. Knijnenburg Clemson University
Transcript of Testing a Recommender System for Self-Actualization · 2017-07-04 · Testing a Recommender System...
Testing a Recommender System for Self-Actualization
Daricia Wilkinson, Saadhika Sivakumar , Pratitee Sinha, Bart P. Knijnenburg
Clemson University
Authors
Daricia Wilkinson
Saadhika Sivakumar
Pratitee Sinha
Bart P. Knijnenburg
Example
VS
GoalsRecommender Systems for Self-Actualization aims to:
Support rather than replace decision making
Focus on exploration rather than consumption
Attempt to cover users’ various tastes
Research Plan
DEVELOP TEST EVALUATE
Research Plan
DEVELOP TEST EVALUATE
Algorithmic Features
Incorrect negative predictions
items = max(average predicted rating – user predicted rating)
“Things we think you’ll hate”
Low valued predictions are never recommended
allows users to correct or confirm low-valued predictions
Algorithmic Features
Unknown Preferences
items = max(predicted ratingmf – predicted ratingknn)2
“Things we have no clue about”
difficult for recommenders to predict items for which there is insufficient information about whether the user will like them or not.
we estimate the system’s confidence by computing the difference in user-predicted ratings for different algorithms
Algorithmic Features
Novel Items
users = max(% top rated items with (#ratings<threshold))items = min(#ratings)
“Things you’ll be among the first to try”
Addresses the cold start problem
users may at times actually be excited to try new things, even if it does not always fit their preferences.
Algorithmic Features
Controversial Items
items = max(var(neighbors’ predicted ratings))
“Things that are controversial”
“safe” recommendations do not challenge a user’s tastes beyond what is generally agreed upon as “good” among like-minded users
items that some like-minded users really like, but others hate
Research Plan
DEVELOP TEST EVALUATE
Testing
More things you might like ~300Things you might hate
Things we are not sure about
Things you’ll be among the first to try
Things that are controversial
In addition to the top 10 things you might like:
Testing
15 20 1
RATE RATE CHOOSE
Research Plan
DEVELOP TEST EVALUATE
Evaluation
http://bit.ly/umuai http://bit.ly/userexperiments
EvaluationAspects
Perceived Recommendation quality, diversity, novelty
System & Choice Satisfaction
Choice and tradeoff difficulty
Perceived taste coverage
Fear of missing things
Taste clarification potential
Taste development potential
Perceived choice conformity
Behavior
Objective coverage
Objective choice conformity
Existing scales
New scales
Summary
New direction for recommenders that supports our aspirational selves
Contribution to the theory of recommender systems evaluation
Acknowledge multidimensionality of users
Thank You
Daricia Wilkinson
Saadhika Sivakumar
Pratitee Sinha
Bart P. Knijnenburg
[email protected] [email protected]@g.clemson.edu [email protected]
