Recommendation Systems

Recommendation Systems

As presented at MIT - Data Analytics Club

Liron Zighelnic

Massachusetts Institution of Technology

Start discovering fashion you will LOVE!

Agenda – or: do you want to stay?

• What is recommendation systems?• Why recommendations are important?• Main methods and algorithms • Real life applications & who use it? (the question should be: who

doesn’t?)


About me

• The Co-founder and CEO of CurtainApp • An MBA candidate at MIT with a passion for business and technology• Did my undergrad in Engineering and Masters in Information Retrieval

(Search engine algorithms) • Have ten years of experience in tech, including software development,

technology management and product management, in a variety of industries including military, intelligence, mobile, venture capital and consulting.• LinkedIn: linkedin.com/in/lironzighelnic • Email: [email protected]


About Curtain

• Curtain is an intelligent mobile app that learns your taste and gives you personal fashion recommendations, making shopping fun and efficient• From the technology side we do Recommendation Algorithms, NLP,

Information Retrieval and a lot of fun

• Register to the beta at: www.curtainapp.com • Join us on Facebook: facebook.com/CurtainApp • Follow us on Twitter: twitter.com/thecurtainapp


Why recommendations are important?“We are leaving the age of information and entering the age of recommendation”

Chris Anderson “The Long Tail”


Why recommendations are important?• The world move from “one size fits all” solutions to personal tailor

made solutions• Users LOVE recommendations – 44% of consumers “strongly agree”

or “agree” that they want product recommendations based on past purchases1

1. Based on a survey done by Walker Sands and published in the 2014 Future of Retail study

http://www.walkersands.com/futureofretail


Why recommendations are important?The paradox of choices

Source: Sandglaz.com


Main methods and algorithms: User-based collaborative filtering

5 1 5 1

3 4 5 5

4 2 4 5

5 1

4 1 1

E.g., K-nearest neighbors


Main methods and algorithms – cont.: Item-based similarity

How to calculate vector similarity?

E.g., Cosine similarity


Real life applications

• There are SO many applications, almost in every field you can think of•We are going to speak about a few


Real life applications - dating

•OkCupid uses recommendation algorithms to be the “ultimate matchmaker” or as they put it “we use math to get you dates” 1


Real life applications - film recommendations• Netflix uses algorithms for film recommendations• E.g., someone who rented the romantic comedy

“10 Things I Hate About You” might be presented with “50 First Dates” 2• The company set a competition called the Netflix Prize in which engineers and researchers competed on building the best algorithm to predict user ratings for films, based on previous ratings, for a $1M award


Real life applications - music

• Pandora uses the properties of a song or artist to create a “radio-station” that plays music with similar properties• It uses user feedback to

refine the results, based on the attributes the user “likes” and “dislikes”

3


Real life applications - Google

•Many applications including: • Google Search• Google now• Google news• Recommendation

generate 38% more click- through1

41. Source Xavier Amatriain’s blog


Real life applications - fashion

•A very new domain•Many of the known

algorithms that were developed originally for book and movies are not relevant or should be further developed/adjusted

5


Real life applications – fashion – Cont.•Main applications:• Suggesting items the

user will like• Suggesting items that

will match the items the user already purchase helping the user to create an outfit 5


What do you think?

Is it important to have vertical solutions or can we come up with one algorithm that will be the best for all different domains?


Questions?


Recommended reading…

• Recommender Systems Handbook by Francesco Ricci, Lior Rokach, Bracha Shapira,Paul B. Kantor

• TED talk: Barry Schwartz - The paradox of choice http://www.ted.com/talks/barry_schwartz_on_the_paradox_of_choice?language=en

http://www.ted.com/talks/barry_schwartz_on_the_paradox_of_choice?language=en




Recommendation Systems - Why How and Real Life Applications

Data & Analytics

recommendation algorithms

product recommendations

technology management

main methods

item based similarity

age of information

product management