/**
* @param dt
* @param eBoundaries
* @param listFeatIndsOfCurInp
* @param listFeatCountOfCurInp
* @throws IOException
*/
private void createNgramFeatures(
DependencyTree dt,
int weight,
Entity ent,
ArrayList<Integer> listFeatIndsOfCurInp,
ArrayList<Integer> listFeatCountOfCurInp)
throws IOException {
String[] feature = new String[0];
// collecting the unigram within a window of {-x, +x}
int x = ent.getStartWordIndex();
for (int i = x - 1; i >= x - 2 && i >= 0; i--) {
feature = new String[] {dt.allNodesByWordIndex[i].lemma + "$" + (i - x)};
GenericFeatVect.addNewFeatureInList(
feature, 1, listFeatIndsOfCurInp, listFeatCountOfCurInp, weight);
}
x = ent.getEndWordIndex();
for (int i = x + 1; i <= x + 2 && i < dt.allNodesByWordIndex.length; i++) {
feature = new String[] {dt.allNodesByWordIndex[i].lemma + "$" + (x - i)};
GenericFeatVect.addNewFeatureInList(
feature, 1, listFeatIndsOfCurInp, listFeatCountOfCurInp, weight);
}
}
/**
* @param curSen
* @param listFeatIndsOfCurInp
* @param listFeatCountOfCurInp
* @param e1
* @param e2
*/
private void extractNonTargetEntityFeatures(
Sentence curSen,
ArrayList<Integer> listFeatIndsOfCurInp,
ArrayList<Integer> listFeatCountOfCurInp,
Entity e1,
Entity e2) {
for (int e = 0; e < curSen.listOfEntities.size(); e++) {
Entity entOther = curSen.listOfEntities.get(e);
if (entOther.getNEcategory().equalsIgnoreCase("DISO")) {
GenericFeatVect.addNewFeatureInList(
new String[] {"DISOinsideSentence"}, 1, listFeatIndsOfCurInp, listFeatCountOfCurInp, 1);
// System.out.println(entOther.id);
DepTreeNode headOfEnt =
curSen.depTree.getHeadWordFromWordBoundaries(
entOther.getAllWordIndexes(), true, curSen);
int hwiOfEntOther = headOfEnt.wordIndex;
int wiOfImmediateGovernor = getNonConjGovernorIndex(curSen, hwiOfEntOther);
if (wiOfImmediateGovernor < 0) continue;
int wiOfNearestVerbGovernor = getNearestVerbGovernor(wiOfImmediateGovernor, curSen);
ArrayList<Integer> listOfWIofEntities = new ArrayList<Integer>();
listOfWIofEntities.addAll(e1.getAllWordIndexes());
listOfWIofEntities.addAll(e2.getAllWordIndexes());
// detect whether drugs are dependent of the verb governor
if (wiOfNearestVerbGovernor > -1
&& curSen.depTree.allNodesByWordIndex[wiOfNearestVerbGovernor].governAllWIsInList(
listOfWIofEntities)) {
/*
GenericFeatVect.addNewFeatureInList( new String[]{"bothEntDependOnVerbGovernorOfTheDISO"}, 1,
listFeatIndsOfCurInp, listFeatCountOfCurInp, 1);
//*/
GenericFeatVect.addNewFeatureInList(
new String[] {
"nearestVerbGovernorOfTheDISO="
+ curSen.arrLemmasByParser[wiOfNearestVerbGovernor].toLowerCase()
},
1,
listFeatIndsOfCurInp,
listFeatCountOfCurInp,
1);
// */
}
if (wiOfImmediateGovernor > -1
&& curSen.depTree.allNodesByWordIndex[wiOfImmediateGovernor].governAllWIsInList(
listOfWIofEntities)) {
/*
GenericFeatVect.addNewFeatureInList( new String[]{"bothEntDependOnImmediateGovernorOfTheDISO"}, 1,
listFeatIndsOfCurInp, listFeatCountOfCurInp, 1);
//*/
GenericFeatVect.addNewFeatureInList(
new String[] {
"immediateGovernorOfTheDISO="
+ curSen.arrLemmasByParser[wiOfImmediateGovernor].toLowerCase()
},
1,
listFeatIndsOfCurInp,
listFeatCountOfCurInp,
1);
// */
}
// *
if (wiOfImmediateGovernor == wiOfNearestVerbGovernor)
GenericFeatVect.addNewFeatureInList(
new String[] {"immediateGovernorIsVerbGovernorOfTheDISO"},
1,
listFeatIndsOfCurInp,
listFeatCountOfCurInp,
1);
// */
}
}
}
/**
* @param curSen
* @param listFeatIndsOfCurInp
* @param listFeatCountOfCurInp
* @param e1
* @param e2
* @param negWI
* @param listOfReducedGraph
*/
private void extractNegationScopeFeatures(
Sentence curSen,
ArrayList<Integer> listFeatIndsOfCurInp,
ArrayList<Integer> listFeatCountOfCurInp,
Entity e1,
Entity e2,
int negWI,
ArrayList<Integer> listOfReducedGraph) {
int wiOfImmediateGovernor = getNonConjGovernorIndex(curSen, negWI);
if (!listOfReducedGraph.contains(wiOfImmediateGovernor) || wiOfImmediateGovernor < 0) return;
int wiOfNearestVerbGovernor = getNearestVerbGovernor(wiOfImmediateGovernor, curSen);
// int dist = 0;
// *
if (!listOfReducedGraph.contains(wiOfNearestVerbGovernor)
// ||
// dist > 3
) wiOfNearestVerbGovernor = -1;
/*
if ( wiOfNearestVerbGovernor > -1 &&
curSen.depTree.allNodesByWordIndex[wiOfNearestVerbGovernor].getParentsWordIndexes().isEmpty()
)
GenericFeatVect.addNewFeatureInList( new String[]{"verbGovernorIsRoot"}, 1,
listFeatIndsOfCurInp, listFeatCountOfCurInp, 1);
*/
ArrayList<Integer> listOfWIofEntities = new ArrayList<Integer>();
listOfWIofEntities.addAll(e1.getAllWordIndexes());
listOfWIofEntities.addAll(e2.getAllWordIndexes());
/*
// detect whether drugs are dependent of the verb governor
if ( wiOfNearestVerbGovernor > -1
&& curSen.depTree.allNodesByWordIndex[wiOfNearestVerbGovernor].governAllWIsInList(listOfWIofEntities) )
GenericFeatVect.addNewFeatureInList( new String[]{"bothEntDependOnVerbGovernor"}, 1,
listFeatIndsOfCurInp, listFeatCountOfCurInp, 1);
*/
if (wiOfImmediateGovernor > -1
&& curSen.depTree.allNodesByWordIndex[wiOfImmediateGovernor].governAllWIsInList(
listOfWIofEntities))
GenericFeatVect.addNewFeatureInList(
new String[] {"bothEntDependOnImmediateGovernor"},
1,
listFeatIndsOfCurInp,
listFeatCountOfCurInp,
1);
if (wiOfImmediateGovernor == wiOfNearestVerbGovernor)
GenericFeatVect.addNewFeatureInList(
new String[] {"immediateGovernorIsVerbGovernor"},
1,
listFeatIndsOfCurInp,
listFeatCountOfCurInp,
1);
/*
GenericFeatVect.addNewFeatureInList( new String[]{"immediateGovernor="
+ curSen.arrLemmasByParser[wiOfImmediateGovernor].toLowerCase()}, 1,
listFeatIndsOfCurInp, listFeatCountOfCurInp, 1);
//*/
if (wiOfNearestVerbGovernor > -1)
GenericFeatVect.addNewFeatureInList(
new String[] {
"nearestVerbGovernor=" + curSen.arrLemmasByParser[wiOfNearestVerbGovernor].toLowerCase()
},
1,
listFeatIndsOfCurInp,
listFeatCountOfCurInp,
1);
// */
}
/**
* @param isSimplifyEntity
* @param tokenWithPos
* @param senID
* @param sentence
* @param listEnt
* @param listRel
* @param listDependencies
* @param medtType
* @param isResolveOverlappingEntities
* @param relToBeConsidered
* @param arrClauseBoundOfSen
* @return
* @throws IOException
*/
public boolean setInpVectFromDepGraphOfPairsAndTrigger(
Relation objRel,
Sentence objCurSen,
int medtType,
boolean discardDepRelUsingProbabilityInReducedGraph,
boolean useWalkFeatures,
boolean useRegExPatterns,
boolean useDepPatterns,
boolean useTriggers,
boolean triggersFromWholeRGinsteadOfLCP,
boolean useNegativeCues,
Entity e1,
Entity e2)
throws IOException {
ArrayList<Integer> listFeatIndsOfCurInp = new ArrayList<Integer>(),
listFeatCountOfCurInp = new ArrayList<Integer>();
DepTreeNode
headOfEnt1 =
objCurSen.depTree.getHeadWordFromWordBoundaries(
e1.getAllWordIndexes(), true, objCurSen),
headOfEnt2 =
objCurSen.depTree.getHeadWordFromWordBoundaries(
e2.getAllWordIndexes(), true, objCurSen);
DepTreeNode dn = null;
// System.out.println(e1.id + " " + e2.id);
// All nodes in the shortest path connecting the target pairs must be retained
// All nodes satisfying the 3 rules of MEDT kernel must be retained
dn =
objCurSen
.depTree
.clone()
.findMinimalSubTreeWithEntities(false, headOfEnt1, medtType, headOfEnt2);
/*
* If there is no minimal subtree/path between target entities then we do not consider to generate
* instance for training/testing.
*/
if (dn == null) {
FileUtility.writeInFile(
GenericFeatVect.vectOutFile,
e1.id + " " + e1.name + "\t" + e2.id + " " + e2.name + "\n\n",
true);
FileUtility.writeInFile(GenericFeatVect.vectOutFile, objCurSen.text + "\n\n", true);
return false;
}
GenericFeatVect.listOfAllInstances.add(
e1.id + " " + e1.name + "\t" + e2.id + " " + e2.name + "\n\n" + objCurSen.text + "\n\n");
// Construct feature set using e-walks and v-walks
if (dn != null && useWalkFeatures) {
createNgramFeatures(objCurSen.depTree, 1, e1, listFeatIndsOfCurInp, listFeatCountOfCurInp);
createNgramFeatures(objCurSen.depTree, 1, e2, listFeatIndsOfCurInp, listFeatCountOfCurInp);
// System.out.println(objCurSen.senID + " " + dn.wordIndex);
objCurSen.depTree.replaceEntitiesWithDummies(
dn, e1.boundaries, e2.boundaries, new ArrayList<Integer>());
createFeaturesFromInputGraph(
objCurSen.depTree,
dn,
"",
listFeatIndsOfCurInp,
listFeatCountOfCurInp,
new ArrayList<Integer>(),
e1.boundaries,
e2.boundaries);
}
if (useRegExPatterns) {
matchPPIpatternOnSentence(
objCurSen.text, e1.name, e2.name, listFeatIndsOfCurInp, listFeatCountOfCurInp);
matchPPIpatternOnSentence(
objCurSen.text, e2.name, e1.name, listFeatIndsOfCurInp, listFeatCountOfCurInp);
}
ArrayList<String> listOfDepRelsInReducedGraph = new ArrayList<String>();
ArrayList<ArrayList<Integer>> listNodesAndLCP =
new PatternsDepRelFromGraph()
.extractDepRelsInReducedGraph(
objRel,
objCurSen,
listOfDepRelsInReducedGraph,
discardDepRelUsingProbabilityInReducedGraph,
false);
if (useDepPatterns)
extractDepPatternFeatures(
listOfDepRelsInReducedGraph, listFeatIndsOfCurInp, listFeatCountOfCurInp);
// NOTE: listNodesAndLCP has two elements - (0) all the nodes in Reduced graph, and (1) the
// least common parents (LCPs) in Reduced graph.
if (listNodesAndLCP == null) listNodesAndLCP = new ArrayList<ArrayList<Integer>>();
if (useTriggers && listNodesAndLCP.size() > 0) {
if (triggersFromWholeRGinsteadOfLCP)
for (int i = 0; i < listNodesAndLCP.get(0).size(); i++)
addTriggerWordFeatures(
objCurSen.depGraph.allNodesByWordIndex[listNodesAndLCP.get(0).get(i)].word,
listFeatIndsOfCurInp,
listFeatCountOfCurInp);
else
for (int i = 0; i < listNodesAndLCP.get(1).size(); i++)
addTriggerWordFeatures(
objCurSen.depGraph.allNodesByWordIndex[listNodesAndLCP.get(1).get(i)].word,
listFeatIndsOfCurInp,
listFeatCountOfCurInp);
}
// *
if (useNegativeCues && listNodesAndLCP.size() > 0) {
for (int i = 0; i < listNodesAndLCP.get(0).size(); i++) {
// System.out.println(objCurSen.senID + " " + dn.wordIndex);
// addNegativeWordFeatures( graph.allNodesByWordIndex[listNodesAndLCP.get(0).get(i)].word,
// listFeatIndsOfCurInp, listFeatCountOfCurInp);
if (Triggers.listOfNegativeWords.contains(
objCurSen.depGraph.allNodesByWordIndex[listNodesAndLCP.get(0).get(i)].lemma)) {
// add a feature indicating that there is a negative word in the reduced graph
String[] feature =
new String[] {
"HasNegWord@$"
+ objCurSen.depGraph.allNodesByWordIndex[listNodesAndLCP.get(0).get(i)].lemma,
};
GenericFeatVect.addNewFeatureInList(
feature, 1, listFeatIndsOfCurInp, listFeatCountOfCurInp, 1);
if (objCurSen.depGraph.allNodesByWordIndex[listNodesAndLCP.get(0).get(i) + 1].lemma
.equalsIgnoreCase("be")
|| objCurSen.depGraph.allNodesByWordIndex[listNodesAndLCP.get(0).get(i) + 1].pos
.equalsIgnoreCase("IN")) {
feature =
new String[] {
"WordNextToNegCue@$"
+ objCurSen
.depGraph
.allNodesByWordIndex[listNodesAndLCP.get(0).get(i) + 2]
.lemma,
};
} else {
feature =
new String[] {
"WordNextToNegCue@$"
+ objCurSen
.depGraph
.allNodesByWordIndex[listNodesAndLCP.get(0).get(i) + 1]
.lemma,
};
}
GenericFeatVect.addNewFeatureInList(
feature, 1, listFeatIndsOfCurInp, listFeatCountOfCurInp, 1);
// *
// extract negation scope features
if (objCurSen.depGraph.allNodesByWordIndex[listNodesAndLCP.get(0).get(i)].lemma.matches(
"(no|not)"))
extractNegationScopeFeatures(
objCurSen,
listFeatIndsOfCurInp,
listFeatCountOfCurInp,
e1,
e2,
listNodesAndLCP.get(0).get(i),
listNodesAndLCP.get(0));
// */
}
}
}
// */
// add Zhou et al. 2005 features
new ExtAceFeatVect()
.getZhouEtAl2005FeatVal(objCurSen, e1, e2, listFeatIndsOfCurInp, listFeatCountOfCurInp);
extractNonTargetEntityFeatures(objCurSen, listFeatIndsOfCurInp, listFeatCountOfCurInp, e1, e2);
GenericFeatVect.sortFeatValByIndx(listFeatIndsOfCurInp, listFeatCountOfCurInp);
GenericFeatVect.listOfAllInstancesWithFeat.add(listFeatIndsOfCurInp);
GenericFeatVect.listOfAllInstancesWithFeatCount.add(listFeatCountOfCurInp);
return true;
}
