/**
* chooses the cluster with the highest number of semantic concepts that correspond to different
* search keywords
*
* @param graph the graph
* @param correspondingKeywords Mapping of semantic concepts (nodeIDs) to search keywords
*/
public void filterClusterByInterconnectionLevel(
WTPGraph graph, Map<String, String> correspondingKeywords) {
Set<Set<String>> clusters = getClusters(graph);
printClusters(clusters);
Set<Set<String>> resultClusters = new HashSet<Set<String>>();
// get biggest number of corresponding keywords of one cluster
int maxNumberOfCorrespondingKeywords = 0;
for (Set<String> cluster : clusters) {
int numOfDifferentCorrespondingKeywords =
getNumberOfCorrespondingKeywords(cluster, correspondingKeywords);
System.out.println(
"Number of different corresponding keywords: " + numOfDifferentCorrespondingKeywords);
if (numOfDifferentCorrespondingKeywords > maxNumberOfCorrespondingKeywords) {
maxNumberOfCorrespondingKeywords = numOfDifferentCorrespondingKeywords;
resultClusters = new HashSet<Set<String>>();
resultClusters.add(cluster);
} else if (numOfDifferentCorrespondingKeywords == maxNumberOfCorrespondingKeywords) {
resultClusters.add(cluster);
}
}
Set<String> survivingNodes = new HashSet<String>();
for (Set<String> cluster : resultClusters) {
survivingNodes.addAll(cluster);
}
Set<Node> dyingNodes = new HashSet<Node>();
for (Node n : graph.getGraph().getNodeSet()) {
if (!survivingNodes.contains(n.getId())) dyingNodes.add(n);
}
graph.getGraph().getNodeSet().removeAll(dyingNodes);
}
/**
* This methods deletes all nodes of the graph that are not part of a path
*
* @param paths all paths that are interconnecting nodes, which correspond to different keywords
* @param graph the graph
*/
private void filterInterconnectedNodes(List<Path> paths, WTPGraph graph) {
// NodeIDs of the nodes that are part of at least one of the path
Set<String> interconnectedNodeIDs = new HashSet<String>();
// Write all nodeIDs included in at least one of the path into the interconnectedNodeIDs Set
for (Path p : paths) {
System.out.println("Path: " + p.toString());
for (Node n : p) {
interconnectedNodeIDs.add(n.getId());
}
}
System.out.println(
"Count of Nodes before interconnectionFiltering: " + graph.getGraph().getNodeCount());
// Nodes whose nodeID is not included in interconnectedNodeIDs and which should be deleted,
// because they are not interconnected
List<Node> notInterconnectedNodes = new LinkedList<Node>();
for (Node n : graph.getGraph().getNodeSet()) {
if (!interconnectedNodeIDs.contains(n.getId())) notInterconnectedNodes.add(n);
}
// remove all of the not interconnected Nodes
graph.getGraph().getNodeSet().removeAll(notInterconnectedNodes);
System.out.println(
"Count of Nodes after interconnectionFiltering:" + graph.getGraph().getNodeCount());
}
/**
* @param graph the graph
* @param relevanceThreshold atleast number of including paths to be relevant
*/
private void removeNodesWithLessThanNOccurrences(WTPGraph graph, int minNumOfOccurrences) {
if (countOfNodeOccurrencesInPaths != null) {
LinkedList<Node> nodesToDelete = new LinkedList<Node>();
for (Node node : graph.getGraph().getNodeSet()) {
Integer occurrenceCount = countOfNodeOccurrencesInPaths.get(node.getId());
if (occurrenceCount == null || occurrenceCount < minNumOfOccurrences)
nodesToDelete.add(node);
}
graph.getGraph().getNodeSet().removeAll(nodesToDelete);
// System.out.println("nodes filtered by number of including paths heuristic");
}
}
public void filterClusterBySize(WTPGraph graph) {
Set<Set<String>> clusters = getClusters(graph);
printClusters(clusters);
Set<String> survivingNodes = new HashSet<String>();
int maxClusterSize = 0;
for (Set<String> cluster : clusters) {
System.out.println("Size of Cluster: " + cluster.size());
if (cluster.size() > maxClusterSize) {
maxClusterSize = cluster.size();
survivingNodes = new HashSet<String>();
survivingNodes.addAll(cluster);
} else if (cluster.size() == maxClusterSize) {
survivingNodes.addAll(cluster);
}
}
Set<Node> dyingNodes = new HashSet<Node>();
for (Node n : graph.getGraph().getNodeSet()) {
if (!survivingNodes.contains(n.getId())) dyingNodes.add(n);
}
graph.getGraph().getNodeSet().removeAll(dyingNodes);
}
public void filterClusterByNodeOccurrencesInPaths(WTPGraph graph, List<Path> paths) {
Set<Set<String>> clusters = getClusters(graph);
printClusters(clusters);
Set<String> survivingNodes = new HashSet<String>();
int maxOccurrance = 0;
for (Set<String> cluster : clusters) {
int count = getNumberOfClusterOccurrencesInPaths(cluster, paths);
System.out.println("Number of cluster occurrences in paths: " + count);
if (count > maxOccurrance) {
maxOccurrance = count;
survivingNodes = new HashSet<String>();
survivingNodes.addAll(cluster);
} else if (count == maxOccurrance) {
survivingNodes.addAll(cluster);
}
}
Set<Node> dyingNodes = new HashSet<Node>();
for (Node n : graph.getGraph().getNodeSet()) {
if (!survivingNodes.contains(n.getId())) dyingNodes.add(n);
}
graph.getGraph().getNodeSet().removeAll(dyingNodes);
}
private Set<Set<String>> getClusters(WTPGraph graph) {
visitedWhileClustering = new HashSet<String>();
Set<Set<String>> clusters = new HashSet<Set<String>>();
// for each node of the graph: if not yet in a cluster, try to generate the cluster starting
// with this node
for (Node n : graph.getGraph().getNodeSet()) {
String id = n.getId();
if (!visitedWhileClustering.contains(id)) {
clusters.add(getCluster(graph, id));
}
}
return clusters;
}
/**
* this method creates a set of all nodes that are connected by at least one path (cluster) it
* starts with on nodeID and uses a breadth first search
*
* @param graph the graph
* @param id of one node of the cluster
* @return a set of node ids, which are all in the same cluster
*/
private Set<String> getCluster(WTPGraph graph, String id) {
Queue<String> queue = new LinkedList<String>();
Set<String> cluster = new HashSet<String>();
queue.add(id);
while (!queue.isEmpty()) {
String temp = queue.poll();
if (!visitedWhileClustering.contains(temp)) {
visitedWhileClustering.add(temp);
cluster.add(temp);
Iterator<Node> neighbourNodesIterator =
graph.getGraph().getNode(temp).getNeighborNodeIterator();
while (neighbourNodesIterator.hasNext()) {
queue.add(neighbourNodesIterator.next().getId());
}
}
}
return cluster;
}
private void _genSFPinQueue(List<String> keywords, UUID jobID) {
// create timestamp
Date now = Calendar.getInstance().getTime();
// create jobqueue entry
jobqueue.put(jobID, new SFPGenJob(SFPGenJob.PROCESSING, now));
System.out.println("[_genSFPinQueue] size of job queue: " + jobqueue.size());
System.out.println("Job accessible @ " + this.jobqueue.toString());
// map for the semantic concepts found in the ontology and their
// corresponding keyword, used for searching them
Map<String, String> correspondingKeywords = new HashMap<String, String>();
KeyWordSearch s = new KeyWordSearch();
List<SearchResult> res = s.search(keywords, maxSearchResults, correspondingKeywords);
System.out.println("Resultlist from KW search: " + res);
List<String> request = KeyWordSearch.toUriList(res);
System.out.println("Starting BFS...");
BreadthFirstSearch lc = new BreadthFirstSearch();
ResultSet result = lc.getConnections(request, maxSearchDepth);
System.out.println("...Done");
// -- 2) create the graph
System.out.println("Creating the initial graph...");
WTPGraph graph = WTPGraph.createFromResultSet(result, "Semantic Fingerprint");
System.out.println("...Done");
// -- 3) remove specific edges
// graph.removeEdgesByName("ject");
// graph.removeEdgesByName("paradigm");
// graph.removeEdgesByName("influencedBy");
// graph.removeEdgesByName("influenced");
// graph.removeEdgesByName("typing");
// graph.removeEdgesByName("license");
// -- 4) tidy graph
System.out.print(
"Tidying graph ("
+ graph.getNodeCount()
+ " Nodes, "
+ graph.getEdgeCount()
+ " Edges) ...");
GraphCleaner c = new GraphCleaner(graph.getGraph(), result.requestNodes);
LinkedList<graph.GraphCleaner.Path> paths = c.clean(maxPathLength, maxPathExtensionLength);
System.out.println(
" Done ("
+ graph.getNodeCount()
+ " Nodes, "
+ graph.getEdgeCount()
+ " Edges, "
+ paths.size()
+ " Paths)");
// --4.2) heuristics finger print selection
InterConceptConntecting heuristic = new InterConceptConntecting();
/** Filters all Nodes that have paths to other Nodes which correspond to a different keyword */
// heuristic.filterInterconntection(graph, paths,
// correspondingKeywords);
/** Filters the n Nodes which occur most frequently in the paths */
heuristic.filterNMostFrequentlyOccuring(
graph, paths, numRelevantNodesFilter, correspondingKeywords);
/** Selects the cluster which corresponds to the most different keywords */
heuristic.filterClusterByInterconnectionLevel(graph, correspondingKeywords);
/** Selects the biggest cluster */
heuristic.filterClusterBySize(graph);
/** Selects the cluster whose nodes occur most frequently in the paths */
// ArrayList<ArrayList<String>> graph = new ArrayList<ArrayString>();
// convert WTP graph to RDF
Model rdfgraph = WTPGraph.getRDFGraph(graph);
rdfgraph.write(System.out);
/*
* ObjectMapper mapper = new ObjectMapper();
*
*
* try { return
* makeCORS(Response.status(Status.OK).entity(mapper.writeValueAsString
* (rdfgraph.write(System.out))), ""); } catch (JsonGenerationException
* e) { // TODO Auto-generated catch block e.printStackTrace(); } catch
* (JsonMappingException e) { // TODO Auto-generated catch block
* e.printStackTrace(); } catch (IOException e) { // TODO Auto-generated
* catch block e.printStackTrace();
*
* } return makeCORS(Response.status(Status.OK), "");
*/
OutputStream output =
new OutputStream() {
private StringBuilder string = new StringBuilder();
@Override
public void write(int b) throws IOException {
this.string.append((char) b);
}
public String toString() {
return this.string.toString();
}
};
rdfgraph.write((OutputStream) output);
// put result in sfplist
this.sfplist.put(jobID, output.toString());
// get the job object of current jobid and update it
SFPGenJob currJob = this.jobqueue.get(jobID);
currJob.updateStatus(SFPGenJob.FINISHED);
// update timestamp
now = Calendar.getInstance().getTime();
currJob.updateTimestamp(now);
this.jobqueue.put(jobID, currJob);
}