Abstract: As we use the internet for more and more things, for a website, knowing what a person needs, wants and would want is not just a nice-to-have feature but almost a must-have.
Today recommendation systems suggest products, films/shows, music, friends/dates, etc. The recommender works by finding similar items to a given item or user. For that, each item first needs to be quantified and represented in way that allows comparison against each-other.
This workshop will begin with the origins of recommendation systems, discuss how they are built and where they are today. We will review the tools and techniques used to build a recommender. Attendees will understand what type of data is necessary, and they will get a sense of what would make an effective recommender.
Having a good recommendation system might be the lowest-hanging-fruit for most websites and companies. It's relatively easy to achieve and it provides immediate value addition to the consumer. We will discuss how to measure such value gain and the challenges in quantifying that.
This workshop will review the two major approaches to creating recommendations: ""content-based filtering"" and ""collaborative filtering"". We also will discuss the opposing forces of ""exploitation"" (recommending the most similar thing) and ""exploration"" (recommending something exciting).
We will wrap by discussing the future of these systems (value and potential), and make a case for a recommender that's personalized to each user's interests and also inclination.
Bio: Coming Soon!