private static double normVector(List<Double> v1) {
double result = 0.0;
for (Double aV1 : v1) {
result = result + (aV1 * aV1);
}
return Math.sqrt(result);
}
public static double getInverseDocumentFrequency(GraphDatabaseService db, Long featureId) {
Double idf;
Double d = ((Integer) getDocumentSize(db)).doubleValue();
Double dt = ((Integer) getDocumentSizeForFeature(db, featureId)).doubleValue();
idf = Math.log(d / dt);
return idf;
}
public static Map<String, Object> getCosineSimilarityVector(GraphDatabaseService db) {
Map<String, List<LinkedHashMap<String, Object>>> documents = getFeaturesForAllClasses(db);
Map<String, List<LinkedHashMap<String, Object>>> results = new HashMap<>();
List<Integer> featureIndexList = getFeatureIndexList(db);
List<String> documentList = documents.keySet().stream().collect(Collectors.toList());
Collections.sort(documentList, (a, b) -> a.compareToIgnoreCase(b));
for (String key : documentList) {
List<LinkedHashMap<String, Object>> resultList = new ArrayList<>();
LinkedHashMap<String, Double> classMap = new LinkedHashMap<>();
List<Double> v1 =
featureIndexList
.stream()
.map(i -> documents.get(key).contains(i) ? featureIndexList.indexOf(i) : 0.0)
.collect(Collectors.toList());
documents
.keySet()
.stream()
.forEach(
otherKey -> {
List<Double> v2 =
featureIndexList
.stream()
.map(
i ->
documents.get(otherKey).contains(i)
? featureIndexList.indexOf(i)
: 0.0)
.collect(Collectors.toList());
classMap.put(otherKey, cosineSimilarity(v1, v2));
});
final List<LinkedHashMap<String, Object>> finalResultList = resultList;
classMap
.keySet()
.forEach(
ks -> {
LinkedHashMap<String, Object> localMap = new LinkedHashMap<>();
localMap.put("class", ks);
localMap.put("similarity", classMap.get(ks));
finalResultList.add(localMap);
});
Collections.sort(
finalResultList,
(a, b) -> ((String) a.get("class")).compareToIgnoreCase((String) b.get("class")));
results.put(key, finalResultList);
}
List<LinkedHashMap<String, Object>> similarityVector = new ArrayList<>();
for (String key : results.keySet()) {
List<Double> cosineVector;
cosineVector =
results
.get(key)
.stream()
.map(a -> Convert.toDouble(Math.round(100000 * (Double) a.get("similarity"))))
.collect(Collectors.toList());
LinkedHashMap<String, Object> row = new LinkedHashMap<>();
row.put("class", key);
row.put("vector", cosineVector);
similarityVector.add(row);
}
Collections.sort(
similarityVector,
(a, b) -> ((String) a.get("class")).compareToIgnoreCase((String) b.get("class")));
Map<String, Object> vectorMap = new LinkedHashMap<>();
List<ArrayList<Double>> vectors = new ArrayList<>();
List<String> classNames = new ArrayList<>();
for (LinkedHashMap<String, Object> val : similarityVector) {
vectors.add((ArrayList<Double>) val.get("vector"));
classNames.add((String) val.get("class"));
}
vectorMap.put("classes", classNames);
vectorMap.put("vectors", vectors);
return vectorMap;
}
