edu.umass.cs.mallet.projects.seg_plus_coref.coreference

Class MultipleCorefClusterer

java.lang.Object

edu.umass.cs.mallet.projects.seg_plus_coref.coreference.CorefClusterAdv

edu.umass.cs.mallet.projects.seg_plus_coref.coreference.MultipleCorefClusterer

public class MultipleCorefClusterer
extends CorefClusterAdv

Clusters multiple objects simultaneously (e.g. Authors, venues, and papers). Trains a separate MaxEnt classifier to weight edges between objects of the same type. Partitions resulting graph such that constraints between objects are respected (e.g. clustering two authors requires their venues be in the same cluster)

Nested Class Summary

Nested classes/interfaces inherited from class edu.umass.cs.mallet.projects.seg_plus_coref.coreference.CorefClusterAdv

CorefClusterAdv.EdgeComparator, CorefClusterAdv.PseudoEdge, CorefClusterAdv.PseudoEdgeComparator, CorefClusterAdv.PseudoVertex

Field Summary

Fields

Modifier and Type	Field and Description
static String	AUTHOR
static String	PAPER Constants for the types of citation nodes
static String	VENUE

Constructor Summary

Constructors

Constructor and Description
MultipleCorefClusterer(Pipe[] _pipes)

Method Summary

Modifier and Type	Method and Description
Collection[]	clusterMentions(InstanceList[] ilists, List[] mentions, int optimalBest, boolean stochastic) Returns a list of collections representing the clustering of "ilists"
void	mergeVertices(salvo.jesus.graph.WeightedGraph g, salvo.jesus.graph.VertexImpl v1, salvo.jesus.graph.VertexImpl v2) This is where we enforce constraints between object types.
Collection	partitionGraph(salvo.jesus.graph.WeightedGraph origGraph) Partition the graph by greed agglomerative merging.
void	setIndices(InstanceList[] ilists) Sets the mapping from citation type to index
void	testClassifiers(InstanceList[] ilists)
void	train(InstanceList[] ilists) Train the underlying classifiers with "ilists" as trainingData

Methods inherited from class edu.umass.cs.mallet.projects.seg_plus_coref.coreference.CorefClusterAdv

absoluteCluster, addVerticesToGraph, buildGraphFromPseudoEdges, chooseEdge2, chooseEdge3, clusterMentions, clusterMentions, collectionAvg, completeGraphNBest, computeInitialTreeObjScore, constructEdgesFromPseudoEdges, constructEdgesUsingTrainedClusterer, constructEdgesUsingTrainedClusterer, constructEdgesUsingTrainedClusterer, constructOptimalEdgesUsingNBest, copyGraph, createGraph, createGraph, createGraph, createPseudoEdges, createPseudoVertices, evaluateAgainstKey, evaluatePartitioning, evaluatePartitioningExternal, evaluatePartitioningExternal, exportGraph, getClassifier, getClusteringFromPseudo, getCollectionOfOriginalObjects, getPseudoClustering, getUnNormalizedScores, hasNextIndexList, inSameCluster, loadME, nextIndexList, nextIndexListStochastic, numSingletons, printGraph, printParamDetails, printParams, setConfWeightedScores, setFullPartition, setKeyPartitioning, setNBestInference, setOptimality, setRBeamSize, setSearchParams, setThreshold, setTrueNumStop, testClassifier, testClassifier, train, trainClassifier, typicalClusterAdv, typicalClusterPartition, updateGraphNBest, updateScore, weightOfConfig

Methods inherited from class java.lang.Object

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

Field Detail

PAPER

public static final String PAPER

Constants for the types of citation nodes

See Also:

Constant Field Values

VENUE

public static final String VENUE

See Also:

Constant Field Values

AUTHOR

public static final String AUTHOR

See Also:

Constant Field Values

Constructor Detail

MultipleCorefClusterer

public MultipleCorefClusterer(Pipe[] _pipes)

Method Detail

train

public void train(InstanceList[] ilists)

Train the underlying classifiers with "ilists" as trainingData

setIndices

public void setIndices(InstanceList[] ilists)

Sets the mapping from citation type to index

testClassifiers

public void testClassifiers(InstanceList[] ilists)

clusterMentions

public Collection[] clusterMentions(InstanceList[] ilists,
                                    List[] mentions,
                                    int optimalBest,
                                    boolean stochastic)

Returns a list of collections representing the clustering of "ilists"

mergeVertices

public void mergeVertices(salvo.jesus.graph.WeightedGraph g,
                          salvo.jesus.graph.VertexImpl v1,
                          salvo.jesus.graph.VertexImpl v2)

This is where we enforce constraints between object types. Constraints are: If v1,v2 are Citation objects, find their corresponding VenueCitations and AuthorCitations and merge them. We do this by first simply merging the paper vertices, then merging their corresponding venue and author vertices.

Overrides:

mergeVertices in class CorefClusterAdv

partitionGraph

public Collection partitionGraph(salvo.jesus.graph.WeightedGraph origGraph)

Partition the graph by greed agglomerative merging. Copied from CorefClusterAdv.java. Added bookkeeping for paperVertex2VenueVertex hash.

Overrides:

partitionGraph in class CorefClusterAdv