dragon.ir.clustering

Class LinkKMean

java.lang.Object

dragon.ir.clustering.AbstractClustering

dragon.ir.clustering.LinkKMean

All Implemented Interfaces:

Clustering

public class LinkKMean
extends AbstractClustering

Link-based K-Means Clustering Algorithm

The Link-based K-Means also incorporates pairwise object relationships into the clustering in addition to tradidtional local features such as terms

Copyright: Copyright (c) 2005

Company: IST, Drexel University

Version:

1.0

Author:

Davis Zhou

Field Summary

Fields inherited from class dragon.ir.clustering.AbstractClustering

clusterNum, clusterSet, featureFilter, indexReader, randomSeed, showProgress

Constructor Summary

Constructors

Constructor and Description
LinkKMean(Clustering initClustering, SparseMatrix links) The constructor with symmetric pairwise object relationship matrix, i.e.
LinkKMean(Clustering initClustering, SparseMatrix outLinks, SparseMatrix inLinks) The constructor with asymmetric pairwise object relationship matrix, i.e.

Method Summary

Modifier and Type	Method and Description
boolean	cluster(IRDoc[] arrDoc) Cluster given testing documents to the given number of clusters
protected DoubleDenseMatrix	estimateClassTransferProb(IRDoc[] arrDoc, int[] arrDocLabel)
protected double	getLogLikelihood(IRDoc doc, int clusterID, DoubleDenseMatrix transMatrix, double[] arrOutLinks, double[] arrInLinks)
int	getMaxIteration()
boolean	getUseWeight()
protected boolean	initialize(IRDoc[] arrDoc)
void	setMaxIteration(int iteration)
void	setUseWeight(boolean useWeight)

Methods inherited from class dragon.ir.clustering.AbstractClustering

cluster, getCluster, getClusterNum, getClusterSet, getFeatureFilter, getIndexReader, getRandomSeed, setFeatureFilter, setRandomSeed, setShowProgress

Methods inherited from class java.lang.Object

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

Constructor Detail

LinkKMean

public LinkKMean(Clustering initClustering,
                 SparseMatrix links)

The constructor with symmetric pairwise object relationship matrix, i.e. the linkage btween objects is undirected.

Parameters:

initClustering - the clustering method for initialization

links - pairwise object relationship matrix. The matrix should be symmetric.

LinkKMean

public LinkKMean(Clustering initClustering,
                 SparseMatrix outLinks,
                 SparseMatrix inLinks)

The constructor with asymmetric pairwise object relationship matrix, i.e. the linkage btween objects is directed.

Parameters:

initClustering - the clustering method for initialization

outLinks - pairwise object relationship matrix. Each row in this matrix stands for the linkages from the corressponding object to other objects.

inLinks - pairwise object relationship matrix. It is the transposed matrix of the out link matrix. Each row in this matrix stands for the linkages from other objects to the object the row corresponds to.

Method Detail

setUseWeight

public void setUseWeight(boolean useWeight)

getUseWeight

public boolean getUseWeight()

initialize

protected boolean initialize(IRDoc[] arrDoc)

cluster

public boolean cluster(IRDoc[] arrDoc)

Description copied from interface: Clustering

Cluster given testing documents to the given number of clusters

Parameters:

arrDoc - the document set for clustering

Returns:

true if clustering successfully

getMaxIteration

public int getMaxIteration()

setMaxIteration

public void setMaxIteration(int iteration)

getLogLikelihood

protected double getLogLikelihood(IRDoc doc,
                                  int clusterID,
                                  DoubleDenseMatrix transMatrix,
                                  double[] arrOutLinks,
                                  double[] arrInLinks)

estimateClassTransferProb

protected DoubleDenseMatrix estimateClassTransferProb(IRDoc[] arrDoc,
                                                      int[] arrDocLabel)