org.knowceans.corpus.analysis
Class AtmTopicsConverter

java.lang.Object
  org.knowceans.corpus.analysis.TopicsConverter
      org.knowceans.corpus.analysis.AtmTopicsConverter

public class AtmTopicsConverter
extends TopicsConverter

ActorMediaTopicAnalyser analyses the results of an LS-AMQ run. It performs the same operations as TopicAnalyser but replaces documents by actors.

Author:

heinrich

Field Summary

private java.util.Vector<java.lang.String>	allActors
private java.util.HashSet<java.lang.Integer>	queryActors

Constructor Summary

AtmTopicsConverter()

Method Summary

protected void	analyseMatching(java.lang.String modelname, java.lang.String suffix) Analyse binary actor--topic relations using KL divergence.
private static double	kl(double[] p, double[] q) KL-divergence between distributions p and q.
static void	main(java.lang.String[] args)
private double[][]	matchActors(double[][][] psi, int p, int q, boolean queryActorsOnly) Build matrix of actor--actor matches according to the Kullback-Leibler-divergence, M_ij = KL(demand_i||supply_j).
private java.util.HashSet<java.lang.Integer>	readActors(java.lang.String modelname) read the actor indices of all actors and those that posted queries.
void	run(java.lang.String directory, java.lang.String model, java.lang.String suffix) performs topic analysis, including query topics
static void	saveMatching(java.lang.String filename, java.util.Vector<org.knowceans.map.TreeMultiMap<java.lang.Double,java.lang.Integer>> actors, java.lang.String comment, java.lang.String additional) save the matching information between actors.
private org.knowceans.map.TreeMultiMap<java.lang.Double,java.lang.Integer>	sort(double[] ds, int i, double d) sort the array in descending order.

Methods inherited from class org.knowceans.corpus.analysis.TopicsConverter

analyse, extractTopics, normaliseRows, posterior, posterior, printMatrix, run, saveTopics, test, truncateMap

Methods inherited from class java.lang.Object

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

Field Detail

allActors

private java.util.Vector<java.lang.String> allActors

queryActors

private java.util.HashSet<java.lang.Integer> queryActors

Constructor Detail

AtmTopicsConverter

public AtmTopicsConverter()

Method Detail

main

public static void main(java.lang.String[] args)

run

public void run(java.lang.String directory,
                java.lang.String model,
                java.lang.String suffix)

performs topic analysis, including query topics

analyseMatching

protected void analyseMatching(java.lang.String modelname,
                               java.lang.String suffix)

Analyse binary actor--topic relations using KL divergence.

Parameters:

modelname - model file name without extension

suffix - of binary files added to model name

comment - for original matrix

saveMatching

public static void saveMatching(java.lang.String filename,
                                java.util.Vector<org.knowceans.map.TreeMultiMap<java.lang.Double,java.lang.Integer>> actors,
                                java.lang.String comment,
                                java.lang.String additional)

save the matching information between actors.

Parameters:

filename -

actors -

comment -

additional -

sort

private org.knowceans.map.TreeMultiMap<java.lang.Double,java.lang.Integer> sort(double[] ds,
                                                                                int i,
                                                                                double d)

sort the array in descending order. XXX: very similar to TopicAnalyser.extractTopics()

Parameters:

ds - array

i - maximum elements in return type or -1

d - minimum element value

Returns:

matchActors

private double[][] matchActors(double[][][] psi,
                               int p,
                               int q,
                               boolean queryActorsOnly)

Build matrix of actor--actor matches according to the Kullback-Leibler-divergence, M_ij = KL(demand_i||supply_j).

Parameters:

psi - matrix[][][] of actor--topic associations, by relations in first index

p - relation of left KL argument

q - relation of right KL argument

queryActorsOnly - set KL divergence to Double.INF for all non-querying actors

Returns:

kl

private static double kl(double[] p,
                         double[] q)

KL-divergence between distributions p and q. (The measure is asymmetric: KL(p||q) = int p(x) * log(p(x)/q(x)) dx).

Parameters:

p - discrete pdf

q - discrete pdf (q.length = p.length)

Returns:

KL divergence

readActors

private java.util.HashSet<java.lang.Integer> readActors(java.lang.String modelname)

read the actor indices of all actors and those that posted queries.

Parameters:

modelname -

Returns: