COURSE OVERVIEW 

User analytics for recommender systems

Georgia Koutrika, Research Director, “Athena” Research and Innovation Center, Athens, Greece

We increasingly get our news, entertainment, and information via computers, tablets, and phones. The proliferation of digital content in a plurality of forms (including e-news, movies, and online courses), along with the popularity of portable devices has created immense opportunities as well as challenges for systems in order to provide users with information and services that best serve the users’ needs. User analytics provide insights into content consumption and user behavior, which can be applied to improve content and services rendered to users. For example, analyzing people’s searches helps search engines, such as Google, understand user search intent and improve search results. Analyzing people’s clicks (Google News), purchases (Amazon) or movies viewed (Netflix) can help generate personalized recommendations for items that the users would like to read, buy, and so forth.

In this course, we study user analytics in the context of recommender systems. Recommender systems provide advice on movies, products, travel, leisure activities, and many other topics, and have become very popular in systems, such as Google News, Amazon, Quora, and Yelp. The aim of the course is to present methods for deriving knowledge from user actions (e.g., purchases, ratings, comments, and so forth) and generating successful recommendations of elements for the users. Real-life recommender systems are examined along the way and measures of “success” in recommender systems are presented.

Outline of the lecture:

PART 1:

  1. What is a recommender system?
  2. Analyzing user past selections: Content-based recommendations
  3. Analyzing user and item relationships:  Collaborative Filtering
    1. Neighborhood-based methods
    2. Model-based
    3. Clustering
    4. Association Rule

PART 2:

  1. Modern Recommendation Methods
    1. Multi-Armed Bandits
    2. Matrix Factorization and Extensions

.                     1. The Netflix Prize

.                     2. From SVD to MF

.                     3. Ad-hoc extensions

3. Graphical models

4. Deep learning

PART 3:

  1. Hybrid systems
    1. Combining predictions
    2. Blending of different types or sources
    3. Diverse Recommendations
  2. Analyzing user attention
    1. Page optimization
  3. Evaluation
  4. Lessons Learnt