public static string DumpTopicsPopularity(IList <string> words, LDAAnalysis lda_analysis)
{
StringBuilder sb = new StringBuilder();
// Count how many docs count each topic in their top-5
int TOP_N = Math.Min(5, lda_analysis.NUM_TOPICS); // Must be less than or equal to 5
int[,] topics_popularity = new int[lda_analysis.NUM_TOPICS, TOP_N];
if (true)
{
TopicProbability[][] density_of_top5_topics_in_docs_sorted = lda_analysis.DensityOfTop5TopicsInDocsSorted; // [doc][n<5]
for (int doc = 0; doc < lda_analysis.NUM_DOCS; ++doc)
{
for (int n = 0; n < TOP_N; ++n)
{
int topic = density_of_top5_topics_in_docs_sorted[doc][n].topic;
++topics_popularity[topic, n];
}
}
}
// Show the descriptive keywords for each topic
if (true)
{
for (int topic = 0; topic < lda_analysis.NUM_TOPICS; ++topic)
{
string description = lda_analysis.GetDescriptionForTopic(words, topic, true, ";", false);
sb.AppendFormat("{0}", description);
sb.AppendLine();
for (int n = 0; n < TOP_N; ++n)
{
sb.AppendFormat("{0}de:{1}\t", n + 1, topics_popularity[topic, n]);
}
sb.AppendLine();
}
}
sb.AppendLine();
if (true)
{
for (int topic = 0; topic < lda_analysis.NUM_TOPICS; ++topic)
{
sb.AppendFormat("{0}\t", topic);
for (int n = 0; n < TOP_N; ++n)
{
sb.AppendFormat("{0}\t", topics_popularity[topic, n]);
}
sb.AppendLine();
}
}
return(sb.ToString());
}
private static double CalculateSimilarity_JS(LDAAnalysis lda_analysis, float[] distribution, int doc_i)
{
// Calculate the similarity of the "topic distribution" for this pair of words using the Jensen-Shannon divergence (always finite)
double JS_12 = 0.0;
double JS_21 = 0.0;
for (int topic = 0; topic < lda_analysis.NUM_TOPICS; ++topic)
{
double P = distribution[topic];
double Q = lda_analysis.DensityOfTopicsInDocuments[doc_i, topic];
double M = (P + Q) / 2.0;
if (0 != M)
{
JS_12 += P * Math.Log(P / M, 2);
JS_21 += Q * Math.Log(Q / M, 2);
}
}
return(1.0 - (JS_12 + JS_21) / 2.0);
}
public static int[] GetDocumentsSimilarToDistribution(LDAAnalysis lda_analysis, float[] distribution)
{
int[] docs = new int[lda_analysis.NUM_DOCS];
double[] similarities = new double[lda_analysis.NUM_DOCS];
// Initial ordering
for (int doc_i = 0; doc_i < lda_analysis.NUM_DOCS; ++doc_i)
{
docs[doc_i] = doc_i;
}
// Similarities
for (int doc_i = 0; doc_i < lda_analysis.NUM_DOCS; ++doc_i)
{
similarities[doc_i] = CalculateSimilarity_JS(lda_analysis, distribution, doc_i);
}
// Sort
Array.Sort(similarities, docs);
Array.Reverse(similarities);
Array.Reverse(docs);
return(docs);
}
