// Do not allow POBJ-POBJ binary relations
public List<RelationInstance> extractBinaryRelations(List<RelationInstance> instanceList) {
List<RelationInstance> newInstanceList = new ArrayList<RelationInstance>();
for (RelationInstance instance : instanceList) {
for (Argument arg1 : instance.getArguments()) {
for (Argument arg2 : instance.getArguments()) {
if (arg1 == arg2) continue;
if (arg1.getArgumentType().equals("SUBJ")) {
if (arg1 == arg2) continue;
if (arg2.getArgumentType().equals("DOBJ")
|| arg2.getArgumentType().startsWith("POBJ")
|| arg2.getArgumentType().equals("SUBJ")) {
newInstanceList.add(createBinaryInstance(instance, arg1, arg2));
}
}
if (arg1.getArgumentType().equals("DOBJ")) {
if (arg1 == arg2) continue;
if (arg2.getArgumentType().startsWith("POBJ")
|| arg2.getArgumentType().equals("DOBJ")) {
newInstanceList.add(createBinaryInstance(instance, arg1, arg2));
}
}
}
}
}
return newInstanceList;
}
private RelationInstance createBinaryInstance(
RelationInstance instance, Argument arg1, Argument arg2) {
RelationInstance newInstance = new RelationInstance();
String preposition = "";
if (arg2.getArgumentType().startsWith("POBJ")) {
preposition = arg2.getArgumentType().substring(5).toLowerCase();
}
newInstance.setAnnotatedSentence(instance.getAnnotatedSentence());
newInstance.setNormalizedRelation(instance.getNormalizedRelation() + " " + preposition);
newInstance.setOriginalRelation(instance.getOriginalRelation());
newInstance.addArgument(arg1);
newInstance.addArgument(arg2);
return newInstance;
}
private void evaluate(
String normalizedSentence, String originalSentence, String entity1, String entity2)
throws FileNotFoundException, UnsupportedEncodingException {
List<RelationInstance> instances = null;
try {
instances = relationExtraction.extractRelations(normalizedSentence, true);
if (instances.size() > 0) {
CoreMap sentence = instances.get(0).getAnnotatedSentence();
SemanticGraph dependencies = sentence.get(CollapsedCCProcessedDependenciesAnnotation.class);
System.out.println("========== Sentence =========");
System.out.println(sentence.toString());
if (dependencies != null) System.out.println(dependencies.toFormattedString());
System.out.println("======= N-ary Instaces ======");
for (RelationInstance instance : instances) {
System.out.println(instance);
}
}
instances = extractBinaryRelations(instances);
} catch (Exception e) {
e.printStackTrace();
System.out.println("Resuming...");
}
boolean relationMatched = false;
String concatenatedRelations = "";
if (instances != null && instances.size() > 0) {
System.out.println("======= Binary Instaces ======");
for (RelationInstance instance : instances) {
System.out.println("Instance: " + instance.getOriginalRelation());
boolean containMention1 = false;
boolean containMention2 = false;
for (Argument arg : instance.getArguments()) {
System.out.println("\tArg: [" + arg.getEntityId() + "] - Type: " + arg.getArgumentType());
// .endsWith() (previously .contains()) is a hack for bad annotated entities in the ground
// truth, such as Andre [[[Agassi]]].
if (arg.getEntityName().endsWith(PLACEHOLDER_ENTITY1)) {
containMention1 = true;
}
if (arg.getEntityName().endsWith(PLACEHOLDER_ENTITY2)) {
containMention2 = true;
}
}
if (containMention1 && containMention2) {
if (concatenatedRelations.isEmpty()) {
concatenatedRelations = instance.getOriginalRelation();
} else {
// concatenatedRelations += " ,, " + instance.getOriginalRelation();
concatenatedRelations += " " + instance.getOriginalRelation();
}
relationMatched = true;
}
}
}
if (!relationMatched) {
ps.println(entity1 + "\t---\t" + entity2 + "\t" + originalSentence);
} else {
ps.println(
entity1 + "\t" + concatenatedRelations.trim() + "\t" + entity2 + "\t" + originalSentence);
}
}