edu.stanford.nlp.cluster
Class PLSI

java.lang.Object
  |
  +--edu.stanford.nlp.cluster.AbstractClusteringMethod
        |
        +--edu.stanford.nlp.cluster.PLSI

All Implemented Interfaces:

ClusteringMethod

public class PLSI

extends AbstractClusteringMethod

Probabilistic Latent Semantic Indexing. PLSI (Hofmann, 1999) uses Expectation Maximization to optimize the log-likelihood of a generative model for documents.

Field Summary

Fields inherited from class edu.stanford.nlp.cluster.AbstractClusteringMethod

clusters, db, method, nc, nd, nt

Constructor Summary

PLSI()
Sets values for db, nt, nd;

Method Summary

SimpleClusters	cluster(DataCollection data, int num_clusters) Perform default number of iterations (60)
double	Epr_z_dw(int z, int d, int w) Expectation Step: calculates posterior probabilities for latent variables z based on current estimates of parameters.
void	initialize() Initialize each class with arbitrary priors P(z), P(w|z), P(d|z).
void	oneIteration() Maximization Step
Cluster	oneIteration(int z) Re-estimates parameters using posterior probabilities given in the E-step.
void	perform_n_iterations(int n) Loops through n iterations of EM

Methods inherited from class edu.stanford.nlp.cluster.AbstractClusteringMethod

cluster, evaluate, evaluate, initialize, toString, toXMLString

Methods inherited from class java.lang.Object

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

Constructor Detail

PLSI

public PLSI()

Sets values for db, nt, nd;

Method Detail

initialize

public void initialize()

Initialize each class with arbitrary priors P(z), P(w|z), P(d|z). these priors are set randomly, but ensure that sum_over_z(P(z))=1; sum_over_zw(P(w|z))=1; and sum_over_zd(P(d|z))=1

Epr_z_dw

public double Epr_z_dw(int z,
                       int d,
                       int w)

Expectation Step: calculates posterior probabilities for latent variables z based on current estimates of parameters.

Returns:

P(z|d,w), calculated from the priors in the aspect model

oneIteration

public Cluster oneIteration(int z)

Re-estimates parameters using posterior probabilities given in the E-step.

oneIteration

public void oneIteration()

Maximization Step

perform_n_iterations

public void perform_n_iterations(int n)

Loops through n iterations of EM

cluster

public SimpleClusters cluster(DataCollection data,
                              int num_clusters)

Perform default number of iterations (60)