Package es.upm.etsisi.cf4j.recommender.matrixFactorization

Class BNMF

java.lang.Object

es.upm.etsisi.cf4j.recommender.Recommender

es.upm.etsisi.cf4j.recommender.matrixFactorization.BNMF

public class BNMF
extends Recommender

Implements Hernando, A., Bobadilla, J., & Ortega, F. (2016). A non negative matrix factorization for collaborative filtering recommender systems on a Bayesian probabilistic model. Knowledge-Based Systems, 97, 188-202.

Field Summary

Fields

Modifier and Type	Field	Description
protected double[][]	a	User factors
protected double	alpha	This hyper-parameter is related to the possibility of obtaining overlapping groups of users sharing the same tastes.
protected double[][]	b	Item factors
protected double	beta	This hyper-parameter represents the amount of evidence that the algorithm requires to deduce that a group of users likes an item.
protected static double	DEFAULT_R
protected double[][]	epsilonMinus	Epsilon- parameters
protected double[][]	epsilonPlus	Epsilon+ parameters
protected double[][]	gamma	Gamma parameters
protected int	numFactors	Number of factors
protected int	numIters	Number of iterations
protected double	r	Hyper-parameter of the binomial distribution.

Fields inherited from class es.upm.etsisi.cf4j.recommender.Recommender

datamodel

Constructor Summary

Constructors

Constructor	Description
BNMF(DataModel datamodel, int numFactors, int numIters, double alpha, double beta)	Model constructor
BNMF(DataModel datamodel, int numFactors, int numIters, double alpha, double beta, double r, long seed)	Model constructor
BNMF(DataModel datamodel, int numFactors, int numIters, double alpha, double beta, long seed)	Model constructor
BNMF(DataModel datamodel, Map<String,Object> params)	Model constructor from a Map containing the model's hyper-parameters values.

Method Summary

Modifier and Type	Method	Description
void	fit()	Estimates model parameters given the hyper-parameters
double	getAlpha()	Get the alpha value
double	getBeta()	Get the beta value
double[]	getEpsilonMinus(int itemIndex)	Get the epsilon- vector of an item
double[]	getEpsilonPlus(int itemIndex)	Get the epsilon+ vector of an item
double[]	getGamma(int userIndex)	Get the gamma vector of an user
double[]	getItemFactors(int itemIndex)	Get the latent factors vector of an item (bi)
int	getNumFactors()	Get the number of factors of the model
int	getNumIters()	Get the number of iterations
double	getR()	Get the r value
double[]	getUserFactors(int userIndex)	Get the latent factors vector of a user (au)
double	predict(int userIndex, int itemIndex)	Computes a rating prediction
String	toString()

Methods inherited from class es.upm.etsisi.cf4j.recommender.Recommender

getDataModel, predict

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Field Detail

DEFAULT_R

protected static final double DEFAULT_R

See Also:

Constant Field Values

a

protected double[][] a

User factors

b

protected double[][] b

Item factors

gamma

protected double[][] gamma

Gamma parameters

epsilonPlus

protected double[][] epsilonPlus

Epsilon+ parameters

epsilonMinus

protected double[][] epsilonMinus

Epsilon- parameters

alpha

protected final double alpha

This hyper-parameter is related to the possibility of obtaining overlapping groups of users sharing the same tastes.

beta

protected final double beta

This hyper-parameter represents the amount of evidence that the algorithm requires to deduce that a group of users likes an item.

r

protected final double r

Hyper-parameter of the binomial distribution.

numFactors

protected final int numFactors

Number of factors

numIters

protected final int numIters

Number of iterations

Constructor Detail

BNMF

public BNMF(DataModel datamodel,
            Map<String,Object> params)

Model constructor from a Map containing the model's hyper-parameters values. Map object must contains the following keys:

• numFactors: int value with the number of latent factors.
• numIters:: int value with the number of iterations.
• alpha: double value with the alpha hyper-parameter. This value is related to the possibility of obtaining overlapping groups of users sharing the same tastes.
• beta: double value with the beta hyper-parameter. This value represents the amount of evidences that the algorithm requires to deduce that a group of users likes an item.
• seed (optional): random seed for generating random numbers. If missing, random value is used.

Parameters:

datamodel - DataModel instance

params - Model's hyper-parameters values

BNMF

public BNMF(DataModel datamodel,
            int numFactors,
            int numIters,
            double alpha,
            double beta)

Model constructor

Parameters:

datamodel - DataModel instance

numFactors - Number of factors

numIters - Number of iterations

alpha - This parameter is related to the possibility of obtaining overlapping groups of users sharing the same tastes

beta - Amount of evidences that the algorithm requires to deduce that a group of users likes an item

BNMF

public BNMF(DataModel datamodel,
            int numFactors,
            int numIters,
            double alpha,
            double beta,
            long seed)

Model constructor

Parameters:

datamodel - DataModel instance

numFactors - Number of factors

numIters - Number of iterations

alpha - This parameter is related to the possibility of obtaining overlapping groups of users sharing the same tastes

beta - Amount of evidences that the algorithm requires to deduce that a group of users likes an item

seed - Seed for random numbers generation

BNMF

public BNMF(DataModel datamodel,
            int numFactors,
            int numIters,
            double alpha,
            double beta,
            double r,
            long seed)

Model constructor

Parameters:

datamodel - DataModel instance

numFactors - Number of factors

numIters - Number of iterations

alpha - This parameter is related to the possibility of obtaining overlapping groups of users sharing the same tastes

beta - Amount of evidences that the algorithm requires to deduce that a group of users likes an item

r - Parameter of the binomial distribution (fixed to 4)

seed - Seed for random numbers generation

Method Detail

getNumFactors

public int getNumFactors()

Get the number of factors of the model

Returns:

Number of factors

getNumIters

public int getNumIters()

Get the number of iterations

Returns:

Number of iterations

getAlpha

public double getAlpha()

Get the alpha value

Returns:

Alpha value

getBeta

public double getBeta()

Get the beta value

Returns:

Beta value

getR

public double getR()

Get the r value

Returns:

r value

getGamma

public double[] getGamma(int userIndex)

Get the gamma vector of an user

Parameters:

userIndex - user index

Returns:

User's gamma vector

getEpsilonPlus

public double[] getEpsilonPlus(int itemIndex)

Get the epsilon+ vector of an item

Parameters:

itemIndex - item index

Returns:

Item's epsilon+ vector

getEpsilonMinus

public double[] getEpsilonMinus(int itemIndex)

Get the epsilon- vector of an item

Parameters:

itemIndex - item index

Returns:

Item's epsilon- vector

getUserFactors

public double[] getUserFactors(int userIndex)

Get the latent factors vector of a user (au)

Parameters:

userIndex - User

Returns:

Latent factors vector

getItemFactors

public double[] getItemFactors(int itemIndex)

Get the latent factors vector of an item (bi)

Parameters:

itemIndex - User

Returns:

Latent factors vector

fit

public void fit()

Description copied from class: Recommender

Estimates model parameters given the hyper-parameters

Specified by:

fit in class Recommender

predict

public double predict(int userIndex,
                      int itemIndex)

Description copied from class: Recommender

Computes a rating prediction

Specified by:

predict in class Recommender

Parameters:

userIndex - Index of the user in the array of Users of the DataModel instance

itemIndex - Index of the item in the array of Items of the DataModel instance

Returns:

Prediction

toString

public String toString()

Overrides:

toString in class Object