MyMediaGraph

A fork of the MyMediaLite framework for recommendation algorithm evaluation with mainly two additions:

• A library of graph vertex ranking recommendation algorithms (my-media-graph-extension)
• Various diversity metrics (my-media-lite-4j/src/main/java/org/mymedialite/eval/measures/diversity)

How to use

• Build the project with maven. Don't run the tests, takes too long.

  mvn clean package -DskipTests

• Run a recommender on the included MovieLens dataset, here the Most Popular-algo (in the project root dir):

  java -jar my-media-main/target/my-media-main-1.0-SNAPSHOT-jar-with-dependencies.jar MostPopular data/ml_medium_7-3_split train.csv test.csv attributes.csv

• Run a recommender (BmANP3) with a particular parameter set:

  java -jar my-media-main/target/my-media-main-1.0-SNAPSHOT-jar-with-dependencies.jar "BmANP3{numWalks=100000,beta=0.9}" data/ml_medium_7-3_split train.csv test.csv attributes.csv

• Get the recommendation quality statistics from data/ml_medium_7-3_split/METRICS.csv.

A list of the included recommenders together with the corresponding identifiers is available at MyMediaGraph/my-media-main/src/main/java/ch/uzh/ifi/ddis/mymedialite/main/ItemRecommenderEvaluate.java. To test additional recommenders, expand the list with classes implementing the ItemRecommender interface.

Publication

Our RecSys'15 paper presenting accuracy and diversity performance of vertex ranking recommendation algorithms: Blockbusters and Wallflowers: Accurate, Diverse, and Scalable Recommendations with Random Walks

Mapping of recommendation algorithm names:

Name in Paper	Command line Identifier
Perfect	Perfect
Random	Random
MostPop	MostPopular
I-kNN	ItemKNN
WI-kNN	WeightedItemKNN
BPRMF	BPRMF2
#3-Paths	IB-3Path
	Laplacian
	P3Matrix
	P3RandomWalkCached
	P3AlphaMatrix
	BmP3Alpha
	P5Matrix
	P3PopAbsNorm
	BmANP3
	HybridS
	BmHybridS

Related projects

MyMediaLite project

Java port of MyMediaLite