User Modeling and Recommender Systems Introduction to recommender

User Modeling and Recommender Systems: Introduction to recommender systems Adolfo Ruiz Calleja 06/09/2014

Index • What is a recommender system? • Classification of recommender systems • Introduction to the main paradigms of recommender systems • Example: Amazon 2

Index • What is a recommender system? – Approacher to avoid information overload – Definition of Recommender Systems – Some examples – Added value of the Recommender Systems • Classification of recommender systems • Introduction to the main paradigms of recommender systems • Example: Amazon 3

Approaches to avoid information overload • Information retrieval (IR) – Static content + dynamic query – The content is modelled – Example: a library search system • Information filtering (IF) – Static query + dynamic content – The query is modelled – Example: anti-spam filter 4

Definition of Recommender Systems (RS) are information filtering systems that seek to predict the preference that a user would give to an item USER Set of user attributes Set of user attributes ITEM Algorithm Set of user attributes Set of user attributes rating 5

Some Examples 6

Some Examples 7

Some Examples 8

Some Examples 9

Added value of the Recommender Systems • Provision of personalized recommendations – But it requires that the maintain a user profile • Allows to persuade each customer with personalized information • Serendipitous discovery • Enables to deal with the long tail – Which is very important in the Web 10

Added value of the Recommender Systems 11

Index • What is a recommender system? • Classification of recommender systems – Different classifications – Domain of the recommendation – Purpose of the recommendation – Context of the recommendation – Data collected – Recommendation algorithm • Introduction to the main paradigms of recommender systems • Example: Amazon 12

Different classifications • • • Domain of the recommender system Purpose of the recommendation Context of the recommendation Data collected Recommendation algorithms Others • Privacy • Interfaces • Software architecture 13

Domain of the recommendation: What is being recommended? • Many different examples – Text documents (web pages, news…) – Media (music, movies…) – Products (or product bundles) – Vendors – People – Sequences • Huge impact on the recommendation algorithm – Should it recommend twice the same item? – How important is time? 14

Purpose of the recommendation • The recommendation itself – E. g. sale a product • Education of the users – E. g. track user behavior to provide recommendations • Build a community around a particular product – E. g. booking 15

Context of the recommendation: What is the user doing? • Can the user be interrupted? – E. g. listening to music vs. shopping • Is the user alone or within a group? – E. g. recommend items to users vs. to groups 16

Data collected • • • How are the recommended items described? How are they collected? Whose opinion does the algorithm collect? How is this opinions collected? How are the profiles created? – Explicit / Implicit • What kind of personal information is collected? – It opens several ethical issues 17

Recommendation algorithm • Which information is taken into account to make the recommendation? • How honest is the recommendation? – Business rules may affect – External manipulation • Transparency of the algorithm 18

Index • What is a recommender system? • Classification of recommender systems • Introduction to the main paradigms of recommender systems – Idea – Not personalized – Content-based recommendation – Knowledge-based recommendation – Collaborative recommendation • Example: Amazon 19

Idea USER Set of user attributes Set of user attributes ITEM Algorithm Set of user attributes Set of user attributes rating 20

Not personalized • • Based on External Community Data Very little information from the user (if any) Simple algorithms They forget about the long tails • Example: Tripadvisor or Billboard 21

Content-based recommendation • User model is built analyzing user preferences and item attributes • Very little information from the user (if any) • Do not need to count with a large group of users • It is hard for them to deal with subjective characteristics of items • Hard to found massively used examples – Personalized news feeds 22

Knowledge-based recommendation • Subclass of content-based recommender systems • Need explicit information “from the outside” – Included by the user (constraint-based) – Knowledge from experts in the domain (cased-based) • Can deal with time spans • Can deal with visitors that only appear once • House, car or technology recommendation – Realtor 23

Collaborative recommendation • • Item model is a set of ratings User model is a set of ratings Many different techniques to match the ratings What to do with new things/people/systems? • Predominant paradigm 24

Index • What is a recommender system? • Classification of recommender systems • Introduction to the main paradigms of recommender systems • Example: Amazon 25