MyMediaLite Recommender System Library

MyMediaLite
a lightweight, multi-purpose library of recommender system algorithms

Zeno Gantner

University of Hildesheim

February 5, 2011

Zeno Gantner, University of Hildesheim: MyMediaLite Recommender System Library — http://ismll.de/mymedialite 1 / 16

Introduction

What are Recommender Systems?


Introduction

MyMediaLite: Recommender System Algorithm Library
functionality
rating prediction
item recommendation from implicit feedback
algorithm testbed

target groups
why use it?
recommender system researchers
simple
educators and students
free
application developers
scalable
misc info well-documented
written in C#, runs on Mono choice
GNU General Public License (GPL)
regular releases (1 or 2 per month)

Using MyMediaLite

Data Flow

hyperparameters

Recommender

interaction
data predictions
Model
user/item
attributes

disk

Using MyMediaLite

Methods Implemented in MyMediaLite
rating prediction
averages: global, user, item
linear baseline method by Koren and Bell
frequency-weighted Slope One
k-nearest neighbor (kNN):
user or item similarities, diﬀ. similarity measures
collaborative or attribute-/content-based
(biased) matrix factorization
item prediction from implicit feedback
random
most popular item
linear content-based model optimized for BPR (BPR-Linear)
support-vector machine using item attributes
k-nearest neighbor (kNN)
weighted regularized matrix factorization (WR-MF)
matrix factorization optimized for BPR (BPR-MF)

Using MyMediaLite

Command-Line Tools

one for each task: rating prediction, item recommendation
simple text format: CSV
pick method and parameters using command-line arguments
evaluate, store/load models

http://ismll.de/mymedialite/documentation/command_line.html


Using MyMediaLite

Embedding MyMediaLite: C#
using System ;
using M yM e d ia Lite . Data ;
using M yM e d ia Lite . E v a l ;
using M yM e d ia Lite . IO ;
using M yM e d ia Lite . ItemRecommendation ;

p u b l i c c l a s s Example
{
p u b l i c s t a t i c v o i d Main ( s t r i n g [ ] a r g s )
{
// l o a d t h e d a t a
v a r u s e r m a p p i n g = new E n t i t y M a p p i n g ( ) ;
v a r i t e m m a p p i n g = new E n t i t y M a p p i n g ( ) ;
v a r t r a i n i n g d a t a = ItemRecommenderData . Read ( a r g s [ 0 ] , u s e r m a p p i n g , i t e m m a p p i n g ) ;
var r e l e v a n t i t e m s = item mapping . I n t e r n a l I D s ;
v a r t e s t d a t a = ItemRecommenderData . Read ( a r g s [ 1 ] , u s e r m a p p i n g , i t e m m a p p i n g ) ;

// s e t up t h e recommender
v a r recommender = new M o s t P o p u l a r ( ) ;
recommender . S e t C o l l a b o r a t i v e D a t a ( t r a i n i n g d a t a ) ;
recommender . T r a i n ( ) ;

// m e a s u r e t h e a c c u r a c y on t h e t e s t d a t a s e t
v a r r e s u l t s = I t e m P r e d i c t i o n E v a l . E v a l u a t e ( recommender , t e s t d a t a , t r a i n i n g d a t a ,
relevant items );
C o n s o l e . W r i t e L i n e ( " prec@5 ={0} " , r e s u l t s [ " prec5 " ] ) ;

// make a p r e d i c t i o n f o r a c e r t a i n u s e r and i t e m
C o n s o l e . W r i t e L i n e ( recommender . P r e d i c t ( u s e r m a p p i n g . T o I n t e r n a l I D ( 1 ) ,
item mapping . ToInternalID ( 1 ) ) ) ;
}
}

Using MyMediaLite

Embedding MyMediaLite: Python

#! / u s r / b i n / e n v i p y

import clr
clr . AddReference ( " MyMediaLite . dll " )
from MyMediaLite import ∗

# load the data
user_mapping = Data . EntityMapping ( )
item_mapping = Data . EntityMapping ( )
train_data = IO . I t e m R ec o m m e n d e r Da t a . Read ( " u1 . base " , user_mapping , item_mapping )
relev ant_ite ms = item_mapping . InternalIDs
test_data = IO . I t e m R ec o m m e n d e r Da t a . Read ( " u1 . test " , user_mapping , item_mapping )

# s e t up t h e recommender
recommender = I te mR e co m me nd a ti o n . MostPopular ( )
recommender . S e t C o l l a b o r a t i v e D a t a ( train_data ) ;
recommender . Train ( )

# m e a s u r e t h e a c c u r a c y on t h e t e s t d a t a s e t
print Eval . I t e m Pr ed i ct i on Ev a l . Evaluate ( recommender , test_data , train_data , relevant_items )

# make a p r e d i c t i o n f o r a c e r t a i n u s e r and i t e m
print recommender . Predict ( user_mapping . ToInternalID ( 1 ) , item_mapping . ToInternalID ( 1 ) )


Using MyMediaLite

Embedding MyMediaLite: Ruby
#! / u s r / b i n / e n v i r

require ’ MyMediaLite ’

min_rating = 1
max_rating = 5

# load the data
user_mapping = MyMediaLite : : Data : : EntityMapping . new ( )
item_mapping = MyMediaLite : : Data : : EntityMapping . new ( )

train_data = MyMediaLite : : IO : : R a t i n g P r e d i c t i o n D a t a . Read ( " u1 . base " , min_rating , max_rating ,
user_mapping , item_mapping )
test_data = MyMediaLite : : IO : : R a t i n g P r e d i c t i o n D a t a . Read ( " u1 . test " , min_rating , max_rating ,
user_mapping , item_mapping )

# s e t up t h e recommender
recommender = MyMediaLite : : RatingPrediction : : UserItemBaseline . new ( )
recommender . MinRating = min_rating
recommender . MaxRating = max_rating
recommender . Ratings = train_data
recommender . Train ( )

# m e a s u r e t h e a c c u r a c y on t h e t e s t d a t a s e t
eval_results = MyMediaLite : : Eval : : RatingEval : : Evaluate ( recommender , test_data )
eval_results . each do | entry |
puts " #{ entry } "
end

# make a p r e d i c t i o n f o r a c e r t a i n u s e r and i t e m
puts recommender . Predict ( user_mapping . ToInternalID ( 1 ) , item_mapping . ToInternalID ( 1 ) )


Extending MyMediaLite

Roll Your Own Recommendation Method

It’s easy.

for basic functionality
deﬁne model data structures
write Train() method
write Predict() method

That’s all!



Roll Your Own: Deﬁne Model Data Structures



Roll Your Own: Write Train() Method



Roll Your Own: Write Predict() Method



Roll Your Own Recommendation Method

It’s easy.

You do not need to worry about including the new method to the
command-line tools, reﬂection takes care of that.

advanced functionality
CanPredict() method
load/store models
on-line updates


Conclusion

MyMediaLite
future work
more methods (contributions welcome . . . )
additional scenarios: context-aware recommendation, tags, . . .
distributed/parallel computing

Methods now shipped with MyMediaLite were
used in the MyMedia field trials (>50,000 users).

acknowledgements
authors: Zeno Gantner, Steffen Rendle, Christoph Freudenthaler
funding by EC FP7 project “Dynamic Personalization of Multimedia”
(MyMedia) under grant agreement no. 215006.
feedback, patches, suggestions: Thorsten Angermann, Fu Changhong,
Andreas Hoffmann, Artus Krohn-Grimberghe, Christina Lichtenth¨ler,
a
Damir Logar, Thai-Nghe Nguyen

Conclusion

MyMediaLite

homepage: http://ismll.de/mymedialite

fork it: http://gitorious.org/mymedialite

follow us: http://twitter.com/mymedialite

send feedback/patches: mymedialite@ismll.de

MyMediaLite: simple — free — scalable — well-documented


MyMediaLite Recommender System Library

Recommended

Recommended

More Related Content

What's hot

What's hot (11)

Viewers also liked

Viewers also liked (20)

Similar to MyMediaLite Recommender System Library

Similar to MyMediaLite Recommender System Library (20)

Recently uploaded

Recently uploaded (20)

MyMediaLite Recommender System Library