Introduction to Recommender Systems

Recommender systems have changed the way people find products, information, and even other people. They study patterns of behavior to know what someone will prefer from among a collection of things he has never experienced. The technology behind recommender systems has evolved over the past 20 years into a rich collection of tools that enable the practitioner or researcher to develop effective recommenders. We will study the most important of those tools, including how they work, how to use them, how to evaluate them, and their strengths and weaknesses in practice.

The algorithms we will study include content-based filtering, user-user collaborative filtering, item-item collaborative filtering, dimensionality reduction, and interactive critique-based recommenders. The approach will be hands-on, with six week projects, each of which will involve implementation and evaluation of some type of recommender.

In addition to topical lectures, this course includes interviews and guest lectures with experts from both academia and industry.

Beginning in February 2015, you will be able to earn a Verified Certificate by verifying your identity via a webcam and a government-issued ID. This option will provide formal recognition of your achievements in the course and includes the University of Minnesota logo. Before then, you can complete a “test run” of the exam. You can then re-take the exam after the Verified Certificate becomes available. For information regarding Verified Certificates, see https://courserahelp.zendesk.com/hc/en-us/articles/201212399-Verified-Certificates

Introduction to Recommender Systems
This module introduces recommender systems and the course. It includes a detailed taxonomy of the types of recommender systems, and also includes a detailed tour of Amazon.com’s recommenders. There is an introductory assessment in the final lesson that leads you through exploring recommender systems on their own.

Non-Personalized Recommenders
This module covers non-personalized recommender systems, including recommendation based on summary statistics and on product-association rules. These recommenders, which are widely used in practice, include overall popularity (how many people like this? what’s the average rating?) and product-to-product recommenders such as “people who bought this item also bought” recommenders. There is an assessment at the end of the module that has you compute non-personalized recommendations.

Content-Based Recommenders
This module covers content-based recommender systems. These systems build a profile of content preferences based on the content attributes associated with items the users has liked or disliked. We’ll discuss common mechanisms for building and maintaining content preference profiles and have an assessment that has you complete hand computations of content profiles and recommendations.

User-User Collaborative Filtering
This module covers user-user collaborative filtering recommender systems. This classic method matches a user against other users with similar preferences and then combines the preferences of those “nearest neighbor” users to form predictions and recommendations. We cover a number of tunings and variations on the algorithm, and have an assessment where you implement your own user-user CF recommender in a spreadsheet.

Evaluation
This module focuses on metrics and evaluation. It introduces a variety of metric types, individual metrics, experimental techniques, and evaluation goals. In many ways, it is at the heart of the course -- what’s the point in having lots of different algorithms if you can’t tell which is better in a situation? The assessment at the end of this module takes you through a set of situations to test your understanding of effective evaluation.

Item Based
This module introduces item-item collaborative filtering, an early innovation that improved run-time performance by computing relationships among items from user rating data. We also look at the interesting case of unary implicit data (like it or don’t know) and have an assessment that has you compute item-item recommendations in a spreadsheet.

Dimensionality Reduction
This module introduces matrix factorization recommendation algorithms, the class of algorithms that seems to be among the most promising today for good recommendation quality and scalability. We introduce you to the concepts behind these algorithms, some specific implementations, and a look at current directions. Your last assessment in the course involves computing predictions and recommendations from factored-matrix representations of ratings matrices.

Advanced Topics
This is our concluding module; it includes coverage of topics such as security threats and the cold-start problem as well as a number of other practical issues. This module also consists of a three-part final exam, covering modules 6-8.