Collaborative filtering getting_started

Collaborative Filtering Algorithms :
Getting Started
Vivek A. Ganesan
vivganes@gmail.com
Big Data Gods Meetup, Santa Clara, CA May 13,
2013

Before we start
Copyright 2013, Vivek A. Ganesan, All rights reserved 1
o A BIG thank you to our sponsors –
Big Data Cloud
o Meeting Space
o Support
o Check out their big data training

Introduction
o Program Outline
o This is an opt-in program, it is FREE! (as in beer)
o We do social coding (which means you share your
code as open source, Apache v2 license)
o Program duration = 1 month, weekly sprints
o Weekly meetup (topical + social coding + Q/A)
o A weekend hackathon (Sat. afternoon) alternate
weeks (deep technical immersion)
o Demo at the end of the program

Agenda
o Introduction to CF Algorithms
o When to use CF?
o Metrics
o Exercise
o Questions?

Introduction to CF Algorithms
o A family of algorithms used to predict
o The preference of an user for an item, given
o a matrix of user preferences for items, where
o preferences must be expressed numerically (for e.g.
user ratings of item on a 1 to 5 integer scale)
o Collaborative because it only looks at user
preferences and does not take in to account user or
item attributes
o Filtering, is math speak for selecting a subset

CF : Common sense version
o Out of a large group of users who have rated
items :
o Pick a “small” subset of users who are “similar” to
you
o Now, for an item that you have not yet rated but your
“similar” users have rated :
o Figure out an “average” rating for the item from your
“similar” group of users
o Weigh it with your rating history and predict a rating

CF : Visual
User/Movie Sleepless in Seattle Titanic Terminator 2
Alice 5 5 3
Bob 1 3 5
Chandra 3 5 4
Dawood 2 3 5
Eduardo (you or
active user)
2 4
?

A sample approach
o Compute Eduardo’s “similarity” to all other
users
o Pick the three users “most similar” to Eduardo
o Weigh their ratings for Terminator 2 by their
degree of similarity to Eduardo
o Make sure that the predicted rating is within
the given scale (0 to 5)
o … and predict Eduardo’s rating for Terminator 2

Step 1 : Measuring Similarity
o Start with a distance metric
o There are several : let’s pick Euclidean for e.g.
o For n space, square root of sum of squared
differences
o Convert it to a similarity score (0 to 1)
o 1/(1 + Euclidean Distance) (adding 1 to avoid
division by zero)
o 0 for no match, 1 for perfect match

CF : Distances & Similarities
Alice Bob Chandra Dawood
3.16 & 0.24 1.414 & 0.414 1.414 & 0.414 1 & 0.5
• Pick the top three users most similar to Eduardo :
• Dawood, Bob and Chandra
• Weigh their ratings for Terminator 2 by their
degree of similarity to Eduardo :
• (0.414 x 5) + (0.414 x 4) + (0.5 x 5) = 6.226
• Ooops – too big a rating (0 to 5 scale)!
• Divide by sum of similarities (0.414 + 0.414 + 0.5)
• Answer : 6.226/1.328 = 4.688 (our prediction)

Improvements
o Some users rate movies consistently higher and
others rate them consistently lower
o Adjust for this by adding distance from mean
and then finally adding mean of the active
user
o Consult the Group Lens paper for details
o Use other measures that solves for “grade
inflation” e.g. Pearson’s

A recommendation engine
o Imagine a much larger data set of users and
movie ratings
o Do the same math for all users against all other
users
o Then predict ratings for those movies for which
users have not yet rated
o For a given user, pick the top N predicted rating
movies and recommend those

Collaborative filtering getting_started

More Related Content

Similar to Collaborative filtering getting_started (20)

Recently uploaded (20)

Collaborative filtering getting_started