edu.stanford.nlp.coref.md.MentionDetectionClassifier Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of stanford-corenlp Show documentation
Show all versions of stanford-corenlp Show documentation
Stanford CoreNLP provides a set of natural language analysis tools which can take raw English language text input and give the base forms of words, their parts of speech, whether they are names of companies, people, etc., normalize dates, times, and numeric quantities, mark up the structure of sentences in terms of phrases and word dependencies, and indicate which noun phrases refer to the same entities. It provides the foundational building blocks for higher level text understanding applications.
package edu.stanford.nlp.coref.md;
import edu.stanford.nlp.util.logging.Redwood;
import java.io.IOException;
import java.io.Serializable;
import java.util.List;
import java.util.Map;
import java.util.Properties;
import java.util.Set;
import edu.stanford.nlp.coref.hybrid.rf.RandomForest;
import edu.stanford.nlp.coref.data.Dictionaries;
import edu.stanford.nlp.coref.data.Mention;
import edu.stanford.nlp.io.IOUtils;
import edu.stanford.nlp.ling.CoreLabel;
import edu.stanford.nlp.ling.RVFDatum;
import edu.stanford.nlp.stats.ClassicCounter;
import edu.stanford.nlp.stats.Counter;
import edu.stanford.nlp.stats.Counters;
import edu.stanford.nlp.util.Generics;
public class MentionDetectionClassifier implements Serializable {
/** A logger for this class */
private static Redwood.RedwoodChannels log = Redwood.channels(MentionDetectionClassifier.class);
private static final long serialVersionUID = -4100580709477023158L;
public RandomForest rf;
public MentionDetectionClassifier(RandomForest rf) {
this.rf = rf;
}
public static Counter extractFeatures(Mention p, Set shares, Set neStrings, Dictionaries dict, Properties props) {
Counter features = new ClassicCounter<>();
String span = p.lowercaseNormalizedSpanString();
String ner = p.headWord.ner();
int sIdx = p.startIndex;
int eIdx = p.endIndex;
List sent = p.sentenceWords;
CoreLabel preWord = (sIdx==0)? null : sent.get(sIdx-1);
CoreLabel nextWord = (eIdx == sent.size())? null : sent.get(eIdx);
CoreLabel firstWord = p.originalSpan.get(0);
CoreLabel lastWord = p.originalSpan.get(p.originalSpan.size()-1);
features.incrementCount("B-NETYPE-"+ner);
if(neStrings.contains(span)) {
features.incrementCount("B-NE-STRING-EXIST");
if( ( preWord==null || !preWord.ner().equals(ner) ) && ( nextWord==null || !nextWord.ner().equals(ner) ) ) {
features.incrementCount("B-NE-FULLSPAN");
}
}
if(preWord!=null) features.incrementCount("B-PRECEDINGWORD-"+preWord.word());
if(nextWord!=null) features.incrementCount("B-FOLLOWINGWORD-"+nextWord.word());
if(preWord!=null) features.incrementCount("B-PRECEDINGPOS-"+preWord.tag());
if(nextWord!=null) features.incrementCount("B-FOLLOWINGPOS-"+nextWord.tag());
features.incrementCount("B-FIRSTWORD-"+firstWord.word());
features.incrementCount("B-FIRSTPOS-"+firstWord.tag());
features.incrementCount("B-LASTWORD-"+lastWord.word());
features.incrementCount("B-LASTWORD-"+lastWord.tag());
for(Mention s : shares) {
if(s==p) continue;
if(s.insideIn(p)) {
features.incrementCount("B-BIGGER-THAN-ANOTHER");
break;
}
}
for(Mention s : shares) {
if(s==p) continue;
if(p.insideIn(s)) {
features.incrementCount("B-SMALLER-THAN-ANOTHER");
break;
}
}
return features;
}
public static MentionDetectionClassifier loadMentionDetectionClassifier(String filename) throws ClassNotFoundException, IOException {
log.info("loading MentionDetectionClassifier ...");
MentionDetectionClassifier mdc = IOUtils.readObjectFromURLOrClasspathOrFileSystem(filename);
log.info("done");
return mdc;
}
public double probabilityOf(Mention p, Set shares, Set neStrings, Dictionaries dict, Properties props) {
try {
boolean dummyLabel = false;
RVFDatum datum = new RVFDatum<>(extractFeatures(p, shares, neStrings, dict, props), dummyLabel);
return rf.probabilityOfTrue(datum);
} catch (Exception e) {
throw new RuntimeException(e);
}
}
public void classifyMentions(List> predictedMentions, Dictionaries dict, Properties props) {
Set neStrings = Generics.newHashSet();
for (List predictedMention : predictedMentions) {
for (Mention m : predictedMention) {
String ne = m.headWord.ner();
if (ne.equals("O")) continue;
for (CoreLabel cl : m.originalSpan) {
if (!cl.ner().equals(ne)) continue;
}
neStrings.add(m.lowercaseNormalizedSpanString());
}
}
for (List predicts : predictedMentions) {
Map> headPositions = Generics.newHashMap();
for (Mention p : predicts) {
if (!headPositions.containsKey(p.headIndex)) headPositions.put(p.headIndex, Generics.newHashSet());
headPositions.get(p.headIndex).add(p);
}
Set remove = Generics.newHashSet();
for (int hPos : headPositions.keySet()) {
Set shares = headPositions.get(hPos);
if (shares.size() > 1) {
Counter probs = new ClassicCounter<>();
for (Mention p : shares) {
double trueProb = probabilityOf(p, shares, neStrings, dict, props);
probs.incrementCount(p, trueProb);
}
// add to remove
Mention keep = Counters.argmax(probs, (m1, m2) -> m1.spanToString().compareTo(m2.spanToString()));
probs.remove(keep);
remove.addAll(probs.keySet());
}
}
for (Mention r : remove) {
predicts.remove(r);
}
}
}
}