What are product recommendations, and how do they work?

CC 2.0 by Horia Varlan | http://flic.kr/p/7vjmof

2

Agenda

•  What are Product Recommenders •  Introducing Recommenders •  A Simple Example •  Recommender Evaluation

•  How do they work? •  Machine learning tool – Apache

Mahout

Namics Conference 2012

September 1, 2012

3

About Sentric

•  Spin-off of MeMo News AG, the leading provider for Social Media Monitoring & Analytics in Switzerland

•  Big Data expert, focused on Hadoop, HBase and Solr

•  Objective: Transforming data into insights

Intro

September 1, 2012

CC 2.0 by Dennis Wong | http://flic.kr/p/6C3RuV  

5

The Patterns

•  Each day we form opinions about things we like, don’t like, and don’t even care about.

•  People tend to like things … •  that similar people like •  that are similar to other things they like

•  These patterns can be used to predict such likes and dislikes.

Introducing Recommenders

September 1, 2012

6

Strategies for Discovering New Things

user-based – Look to what people with similar tastes seem to like Example:

Introducing Recommenders

September 1, 2012

7

Strategies for Discovering New Things

item-based – Figure out what items are like the ones you already like (again by looking to others’ apparent preferences)

Example:

Introducing Recommenders

September 1, 2012

8

Strategies for Discovering New Things

content-based – Suggest items based on particular attribute (again by looking to others’ apparent preferences)

Example:

September 1, 2012

Introducing Recommenders

September 1, 2012

9

Recommendation is all about predicting patterns of taste, and using them to discover new and desirable things you didn’t already know about.

Collaborative Filtering – Producing recommendations based on, and only based on, knowledge of users’ relationships to items.

Recommenders

User-based

Item-based

Content-based

The Definition of Recommendation

Introducing Recommenders

September 1, 2012

CC 2.0 by Will Scullin | http://flic.kr/p/6K9jb8  

11

The Workflow

•  Let’s start with a simple example

A Simple user-based Example

September 1, 2012

Create  Input  Data  

Create  a  Recommender  

Analyse  the  Output  

12

Input Data

•  Recommendations will base on input-data

•  Data takes the form of preferences –associations from users to items

Example: These values might be ratings on a scale of 1 to 5, where 1 indicates items the user can’t stand, and 5 indicates favorites.

A Simple user-based Example

September 1, 2012

1,101,5.0 1,102,3.0 1,103,2.5 2,101,2.0 2,102,2.5 2,103,5.0 2,104,2.0 3,101,2.5 3,104,4.0 3,105,4.5 3,107,5.0 4,101,5.0 4,103,3.0"4,104,4.5"4,106,4.0"5,101,4.0"5,102,3.0"5,103,2.0"5,104,4.0"5,105,3.5"5,106,4.0 "    

User 1 has a preference 3.0 for item 102

13

Trend Visualization

•  Trend visualization for positive users preferences (in petrol)

•  All other preferences are recognized as negative – the user doesn’t seem to like the item that much (red, dotted)

A Simple user-based Example

September 1, 2012

1,101,5.0 1,102,3.0 1,103,2.5 2,101,2.0 2,102,2.5 2,103,5.0 2,104,2.0 3,101,2.5 3,104,4.0 3,105,4.5 3,107,5.0 4,101,5.0 4,103,3.0"4,104,4.5"4,106,4.0"5,101,4.0"5,102,3.0"5,103,2.0"5,104,4.0"5,105,3.5"5,106,4.0 "    

1 5 3

101 102 103 104 105 106 107

2 4

14

Trend Visualization

Users 1 and 5 seem to have similar tastes. Both like 101, like 102 a little less, and like 103 less still

A Simple user-based Example

September 1, 2012

1 5

101 102 103 104 105 106 107

4 2

Users 1 and 4 seem to have similar tastes. Both seem to like 101 and 103 identically

Users 1 and 2 have tastes that seem to run counter to each other

15

Analyzing the Output

So what product might be recommended to user 1?

A Simple user-based Example

September 1, 2012

1 5 3

101 102 103 104 105 106 107

2 4

Obviously not 101, 102 or 103. User 1 already knows about these.

16

Analyzing the Output

The output could be: [item:104, value:4.257081]" The recommender engine did so because it estimated user 1’s preference for 104 to be about 4.3, and that was the highest among all the items eligible for recommendation. Questions: •  Is this the best recommendation for user 1? •  What exactly is a good recommendation?

A Simple user-based Example

September 1, 2012

CC 2.0 by larsaaboe | http://flic.kr/p/7nJpV8  

18

Evaluating a Recommender

Goal: Evaluate how closely the estimated preferences match the actual preferences.

How?

A Simple user-based Example

September 1, 2012

Prepare Split Run Analyse Reasonable data set  

30% for test 70 % for training

Compare estimates with test data à Calculate a score

Produce estimate preferences with training data

Experiment with other recommenders

19

Evaluating a Recommender

Example evaluation output for a particular recommender engine

Note: A score of 0.0 would mean perfect estimation

A Simple user-based Example

September 1, 2012

Item 1 Item 2 Item 3

Actual 3.0 5.0 4.0

Estimate 3.5 2.0 5.0

Difference 0.5 3.0 1.0

Average distance = (0.5+3.0+1.0)/3=1.5

Root-mean-square =√((0.52+3.02+1.02)/3)=1.8484

CC 2.0 by amtrak_russ | http://flic.kr/p/6fAPej  

21

Apache Mahout

•  Mahout … •  Open-source machine learning library from

Apache (Java) •  Can be used for large data collections – it’s

scalable, build upon Apache Hadoop •  Implements algorithms such as

Classification, Recommenders, Clustering •  Incubates a number of techniques and

algorithms

•  ML it’s a hype! But …

In a Nutshell

September 1, 2012

22

Create a Recommender

A Simple Recommender

A Simple user-based Example

September 1, 2012

class RecommenderExample {" … main(String[] args) throws … {" DataModel model = new FileDataModel(new File(“examle.csv")); " UserSimilarity similarity = " new PearsonCorrelationSimilarity(model);" UserNeighborhood neighborhood = " new NearestNUserNeighborhood(2, similarity, model);" Recommender recommender = " new GenericUserBasedRecommender(model, neighborhood, similarity);" List<RecommendedItem> recommendations = recommender.recommend(1, 1);"" for (RecommendedItem recommendation : recommendations) {" System.out.println(recommendation);" }"}}"  

23

Component Interaction

A user-based Recommender

September 1, 2012

<<interface>>  Recommender  

<<interface>>  UserSimilarity  

<<interface>>  UserNeighborhood  

<<interface>>  DataModel  ApplicaAon  

24

UserNeighborhood

NearestNUserNeighborhood

A neighborhood around user 1 is chosen to consist of the three most similar users: 5, 4, and 2

ThresholdUserNeighborhood Defining a neighborhood of most-similar users with a similarity threshold

Algorithms

September 1, 2012

2  

1  5  

4  3  

2  

1  5  

4  3  

25

User Similarity

Implementations of this interface define a notion of similarity between two users. Implementations should return values in the range -1.0 to 1.0, with 1.0 representing perfect similarity.

Algorithms

September 1, 2012

<<interface>> UserSimilarity"

LogLikelihoodSimilarity"

TanimotoCoefficientSimilarity" ..."

EuclideanDistanceSimilarity"

PearsonCorrelationSimilarity"

UncenteredCosineSimilarity"

26

User Similarity

Similarity between data objects can be represented in a variety of ways: •  Distance between data objects is sum of the

distances of each attribute of the data objects (i.e. Euclidean Distance)

•  Measuring how the attributes of both data objects change with respect to the variation of the mean value for the attributes (Pearson Correlation coefficient)

•  Using the word frequencies for each document, the normalized dot product of the frequencies can be used as a measure of similarity (cosine similarity)

•  An a few more ..

Algorithms

September 1, 2012

27

Euclidean Distance

Similarity between two data objects:

Mathematically & Plot

September 1, 2012

User 1

User 2

User 3 User 4

User 5

0

1

2

3

4

5

0 1 2 3 4 5

102

101

28

Pearson Correlation

Similarity between two data objects:

Mathematically & Plot

September 1, 2012

101

102

103

104

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

0 1 2 3 4 5

Use

r 5

User 1

29

Thank you!

Questions? Jean-Pierre König, [email protected]

Namics Conference 2012

September 1, 2012

30

Literatur & Links

•  References The content of this presentation is based on: •  Chapter 1, 2 and 4 of the following book:

Owen, Anil, Dunning, Friedman. Mahout in Action. Shelter Island, NY: Manning Publications Co., 2012.

•  Chapter “Discussion of Similarity Metrics” of the following publication: Shanley Philip. Data Mining Portfolio.

•  Links http://bitly.com/bundles/jpkoenig/1

A Simple user-based Example

September 1, 2012