public String printCategoryDistribution(SortedMap<Integer, Double> categoryDistribution) {
StringBuilder scoreValues = new StringBuilder();
LOG.debug("output category distribution:");
scoreValues.append("scores<-c(");
for (Map.Entry<Integer, Double> score : categoryDistribution.entrySet()) {
String name = treeCache.getNameById(score.getKey(), String.valueOf(score.getKey()));
LOG.debug(name + "\t\t" + score.getValue() + "\t" + treeCache.getDepth(score.getKey()));
if (scoreValues.length() > 10) {
scoreValues.append(", ");
}
scoreValues.append(score.getValue());
}
scoreValues.append(")");
return scoreValues.toString();
}
/**
* Given concept vector which represents a text, calculate the category probabilities.
*
* @param cv
* @param conceptLimit only conceptLimit concepts are used to calculate the category path
* distribution
* @return category id -> scores
*/
public SortedMap<Integer, Double> getCategoryDistribution(ConceptVector cv, int conceptLimit)
throws WikitException {
ConceptIterator conceptIterator = cv.orderedIterator();
// Category ID --> Category
Map<Integer, Category> bags = new HashMap<>();
int count = 0;
while (conceptIterator.next() && count++ < conceptLimit) {
int conceptId = conceptIterator.getId();
ConceptItem conceptItem = new ConceptItem(conceptId, conceptIterator.getValue());
Set<Integer> catIds = treeCache.getCategoryIdsByConceptId(conceptId);
conceptItem.catIds = catIds;
for (int catId : catIds) {
if (bags.containsKey(catId)) {
bags.get(catId).addItem(conceptItem);
} else {
Category category = new Category(catId);
category.addItem(conceptItem);
bags.put(catId, category);
}
}
}
double totalScore = 0;
SortedMap<Integer, Double> sortedScores = null;
// category id -> score
Map<Integer, Double> scores = new HashMap<>();
// LOG.info("Method 1:");
// for (Map.Entry<Integer, Category> entry : bags.entrySet()) {
// double categoryScore = 0;
// Category category = entry.getValue();
// for (ConceptItem item : category.concepts()) {
// categoryScore += item.value;
// }
// double normalizedScore = categoryScore/entry.getValue().size();
// scores.put(entry.getKey(), normalizedScore);
// totalScore += normalizedScore;
// }
//
// sortedScores = new TreeMap<Integer,Double>(new ValueComparator(scores));
// sortedScores.putAll(scores);
// Method 2, take link into account"
double LAMBDA = 0.6;
totalScore = 0;
scores = new HashMap<>();
for (Map.Entry<Integer, Category> entry : bags.entrySet()) {
double categoryScore = 0;
Category category = entry.getValue();
for (ConceptItem conceptItem : category.concepts()) {
double v1 = conceptItem.value;
double v2 = 0.0;
Set<Integer> inlinkIds = conceptCache.getInlinkIds(conceptItem.id);
for (int inlinkId : inlinkIds) {
if (category.hasConcept(inlinkId)) {
v2 += category.getConcept(inlinkId).value;
// System.out.println(inlink + "==>" + item.id + "\t" + item.title);
}
}
if (inlinkIds.size() > 0) {
v2 = v2 / inlinkIds.size(); // normalize
}
// if item connected with
double v = LAMBDA * v1 + (1 - LAMBDA) * v2;
categoryScore += v;
}
double normalizedScore = categoryScore / category.size();
scores.put(category.id, normalizedScore);
totalScore += normalizedScore;
}
sortedScores = new TreeMap<Integer, Double>(new ValueComparator<>(scores));
boolean normalize = true;
if (normalize) {
// normalized the value
for (Map.Entry<Integer, Double> entry : scores.entrySet()) {
sortedScores.put(entry.getKey(), entry.getValue() / totalScore);
}
} else {
sortedScores.putAll(scores);
}
return sortedScores;
}