net.librec.recommender.content

Class HFTRecommender

java.lang.Object

net.librec.recommender.TensorRecommender

net.librec.recommender.content.HFTRecommender

All Implemented Interfaces:

Recommender

public class HFTRecommender
extends TensorRecommender

HFT Recommender McAuley J, Leskovec J. Hidden factors and hidden topics: understanding rating dimensions with review text[C] Proceedings of the 7th ACM conference on Recommender systems. ACM, 2013: 165-172.

Author:

ChenXu

Field Summary

Fields

Modifier and Type	Field and Description
protected float	initMean init mean
protected float	initStd init standard deviation
protected DenseVector	itemBiases user biases
protected DenseMatrix	itemFactors item latent factors
protected int	K
protected int	numberOfWords
protected double[][]	phiks
protected double	regBias bias regularization
protected float	regItem item regularization
protected float	regUser user regularization
com.google.common.collect.BiMap<java.lang.String,java.lang.Integer>	reviewMappingData
protected SparseStringMatrix	reviewMatrix
protected Randoms	rn
protected StringUtil	str
protected double[][]	thetaus
protected SparseStringMatrix	topicAssignment
protected DenseMatrix	topicToWord
protected SparseMatrix	trainMatrix
protected DenseVector	userBiases user biases
protected DenseMatrix	userFactors user latent factors

Fields inherited from class net.librec.recommender.TensorRecommender

allFeaturesMappingData, conf, context, decay, dimensions, earlyStop, globalMean, isBoldDriver, isRanking, itemDimension, itemMappingData, lastLoss, learnRate, LOG, loss, maxLearnRate, maxRate, minRate, numDimensions, numFactors, numItems, numIterations, numUsers, recommendedList, reg, testTensor, topN, trainTensor, userDimension, userMappingData, validTensor, verbose

Constructor Summary

Constructors

Constructor and Description
HFTRecommender()

Method Summary

Methods

Modifier and Type	Method and Description
protected void	calculatePhis()
protected void	calculateThetas()
java.util.Map<Measure.MeasureValue,java.lang.Double>	evaluateMap() evaluate Map
java.util.List<RecommendedItem>	getRecommendedList() get Recommended List
protected double	predict(int[] indices) predict a specific rating for user userIdx on item itemIdx with some other contexts indices, note that the prediction is not bounded.
protected double	predict(int u, int j)
protected java.lang.String	sampleTopicsToWords(java.lang.String[] wordsList, int u)
protected void	sampleZ()
protected void	setup() setup
protected void	trainModel() The training approach is SGD instead of L-BFGS, so it can be slow if the dataset is big.
protected double[]	updateArray(double[] oldValues, double[] newValues) Update function for thetas and phiks, check if softmax comes in to NaN and update the parameters.

Methods inherited from class net.librec.recommender.TensorRecommender

cleanup, evaluate, getDataModel, isConverged, loadModel, predict, recommend, recommend, recommendRank, recommendRating, saveModel, setContext, updateLRate

Methods inherited from class java.lang.Object

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

Field Detail

trainMatrix

protected SparseMatrix trainMatrix

reviewMatrix

protected SparseStringMatrix reviewMatrix

topicToWord

protected DenseMatrix topicToWord

topicAssignment

protected SparseStringMatrix topicAssignment

K

protected int K

numberOfWords

protected int numberOfWords

userBiases

protected DenseVector userBiases

user biases

itemBiases

protected DenseVector itemBiases

user biases

userFactors

protected DenseMatrix userFactors

user latent factors

itemFactors

protected DenseMatrix itemFactors

item latent factors

initMean

protected float initMean

init mean

initStd

protected float initStd

init standard deviation

regBias

protected double regBias

bias regularization

regUser

protected float regUser

user regularization

regItem

protected float regItem

item regularization

reviewMappingData

public com.google.common.collect.BiMap<java.lang.String,java.lang.Integer> reviewMappingData

str

protected StringUtil str

rn

protected Randoms rn

thetaus

protected double[][] thetaus

phiks

protected double[][] phiks

Constructor Detail

HFTRecommender

public HFTRecommender()

Method Detail

setup

protected void setup()
              throws LibrecException

Description copied from class: TensorRecommender

setup

Overrides:

setup in class TensorRecommender

Throws:

LibrecException - if error occurs during setting up

sampleZ

protected void sampleZ()
                throws java.lang.Exception

Throws:

java.lang.Exception

updateArray

protected double[] updateArray(double[] oldValues,
                   double[] newValues)
                        throws java.lang.Exception

Update function for thetas and phiks, check if softmax comes in to NaN and update the parameters.

Parameters:

oldValues - old values of the parameter

newValues - new values to update the parameter

Returns:

the old values if new values contain NaN

Throws:

java.lang.Exception - if error occurs

calculateThetas

protected void calculateThetas()

calculatePhis

protected void calculatePhis()

sampleTopicsToWords

protected java.lang.String sampleTopicsToWords(java.lang.String[] wordsList,
                                   int u)
                                        throws java.lang.Exception

Throws:

java.lang.Exception

trainModel

protected void trainModel()

The training approach is SGD instead of L-BFGS, so it can be slow if the dataset is big.

Specified by:

trainModel in class TensorRecommender

predict

protected double predict(int[] indices)

Description copied from class: TensorRecommender

predict a specific rating for user userIdx on item itemIdx with some other contexts indices, note that the prediction is not bounded. It is useful for building models with no need to bound predictions.

Specified by:

predict in class TensorRecommender

Parameters:

indices - user index, item index and context indices

Returns:

predictive rating for user userIdx on item itemIdx with some other contexts indices without bound

predict

protected double predict(int u,
             int j)

evaluateMap

public java.util.Map<Measure.MeasureValue,java.lang.Double> evaluateMap()
                                                                 throws LibrecException

Description copied from interface: Recommender

evaluate Map

Specified by:

evaluateMap in interface Recommender

Overrides:

evaluateMap in class TensorRecommender

Returns:

evaluate map

Throws:

LibrecException - if error occurs during constructing evaluate map

getRecommendedList

public java.util.List<RecommendedItem> getRecommendedList()

Description copied from interface: Recommender

get Recommended List

Specified by:

getRecommendedList in interface Recommender

Overrides:

getRecommendedList in class TensorRecommender

Returns:

recommended list