/****************************** Constructor **********************************/
// Personalized PageRank Algorithm: All node do random jump to Ego node
public Model(Graph graph, double dampingFactor, int targetNode)
{
this.linkMatrix = graph.matrix;
this.nNodes = graph.getCntAllNodes();
this.dampingFactor = dampingFactor;
this.egoNode = targetNode;
rank = new double[nNodes];
nextRank = new double[nNodes];
for (int i = 0; i < nNodes; i++)
{
// Initialize rank score of each node
rank[i] = (i == targetNode) ? nNodes : 0;
nextRank[i] = 0;
}
}
public Model(Graph graph, double dampingFactor, int targetNode)
{
this.graph = graph;
this.nNodes = graph.getCntAllNodes();
this.dampingFactor = dampingFactor;
rank = new double[nNodes];
nextRank = new double[nNodes];
restart = new double[nNodes];
for (int i = 0; i < nNodes; i++)
{
// Initialize rank score of each node
rank[i] = (i == targetNode) ? nNodes : 0;
nextRank[i] = 0;
// Make restart weight
restart[i] = (i == targetNode) ? 1d : 0; // [1, 0, 0, 0, 0, ... , 0, 0, 0]: Personalized PageRank
}
}
public Recommender(Graph graph)
{
this.graph = graph;
}
private void runKfoldCrossValidation(object parameters)
{
try
{
// Setting environment for experiments
kFoldThreadParameter p = (kFoldThreadParameter)parameters;
DataLoader loader = p.loader;
Methodology methodology = p.methodology;
int fold = p.fold;
int nIteration = p.nIteration;
// #1 Core Part: 'EgoNetwork DB' --> 'Graph Strctures(Node, Edge)'
loader.setTrainTestSet(fold);
if (Program.isValidTrainSet == false)
return;
List<Feature> features = loader.getFeaturesOnMethodology(methodology);
loader.setGraphConfiguration(features);
// Graph Elements: Nodes and edges
Dictionary<int, Node> nodes = loader.allNodes;
Dictionary<int, List<ForwardLink>> edges = loader.allLinksFromNodes;
// Incase: Mention Count(O), Friendship(X)
if (methodology == Methodology.INCL_MENTIONCOUNT
|| methodology == Methodology.INCL_AUTHORSHIP_AND_MENTIONCOUNT
|| methodology == Methodology.INCL_FOLLOWSHIP_ON_THIRDPARTY_AND_MENTIONCOUNT
|| methodology == Methodology.EXCL_FRIENDSHIP)
{
foreach (List<ForwardLink> forwardLinks in edges.Values)
{
for (int i = 0; i < forwardLinks.Count; i++)
{
if (forwardLinks[i].type == EdgeType.FRIENDSHIP)
{
ForwardLink revisedForwardLink = forwardLinks[i];
revisedForwardLink.type = EdgeType.UNDEFINED; // FRIENDSHIP --> UNDEFINED
forwardLinks[i] = revisedForwardLink;
}
}
}
}
// #2 Core Part: 'Graph Strcture(Nodes, Edges)' --> 'PageRank Matrix(2D-ragged array)'
Graph graph = new Graph(nodes, edges);
graph.buildGraph();
// #3 Core Part: Recommendation list(Personalized PageRanking Algorithm)
Recommender recommender = new Recommender(graph);
var recommendation = recommender.Recommendation(0, 0.15f, nIteration); // '0': Ego Node's Index, '0.15f': Damping Factor
// #4 Core Part: Validation - AP(Average Precision)
DataSet testSet = loader.getTestSet();
HashSet<long> egoLikedTweets = testSet.getEgoLikedTweets();
int hit = 0, like = 0;
double AP = 0.0, recall = 0.0; // Average Precision
for (int i = 0; i < recommendation.Count; i++)
{
if (egoLikedTweets.Contains(((Tweet)recommendation[i]).ID))
{
hit += 1;
AP += (double)hit / (i + 1);
}
}
// LIKE
like = (int)egoLikedTweets.Count;
// Average Precision & Recall
if (hit != 0)
{
AP /= hit;
recall = (double)hit / like;
}
else
{
AP = 0.0;
recall = 0.0;
}
// Add current result to final one
lock (kFoldLocker)
{
foreach (EvaluationMetric metric in this.metrics)
{
switch (metric)
{
case EvaluationMetric.MAP:
this.finalResult[metric] += AP;
break;
case EvaluationMetric.RECALL:
this.finalResult[metric] += recall;
break;
case EvaluationMetric.LIKE:
this.finalResult[metric] += like;
break;
case EvaluationMetric.HIT:
this.finalResult[metric] += hit;
break;
}
}
}
}
catch (FileNotFoundException e)
{
Console.WriteLine(e);
}
finally
{
kFoldSemaphore.Release();
}
}
public static void runKFoldCrossValidation(object parameters)
{
try {
Program.semaphore.WaitOne();
// Setting environment for experiments
ThreadParams p = (ThreadParams)parameters;
string dbFile = p.dbFile;
int nFolds = p.nFolds;
int nIterations = p.nIterations;
// Check if the DB file exists
if (!File.Exists(dbFile))
throw new FileNotFoundException(dbFile);
// Do experiments for each methodology
foreach (Methodology methodology in Program.methodologies) {
// Get ego user's ID and his like count
long egoUser = long.Parse(Path.GetFileNameWithoutExtension(dbFile));
int cntLikes = 0;
// Check if this experiment has ever been performed earlier
int m = (int)methodology;
if (Program.existingResults.ContainsKey(egoUser) && Program.existingResults[egoUser].Contains(m)) {
lock (Program.locker) {
Console.WriteLine("Ego network(" + egoUser + "): done on experiment #" + m);
}
continue;
}
// Final result to put the experimental result per fold together
var finalResult = new Dictionary<EvaluationMetric, double>();
foreach (EvaluationMetric metric in Enum.GetValues(typeof(EvaluationMetric)))
finalResult.Add(metric, 0d);
// Need to avoid the following error: "Collection was modified; enumeration operation may not execute"
List<EvaluationMetric> metrics = new List<EvaluationMetric>(finalResult.Keys);
// K-Fold Cross Validation
for (int fold = 0; fold < nFolds; fold++) {
// Load graph information from database and then configurate the graph
DataLoader loader = new DataLoader(dbFile, nFolds);
if (fold == 0) {
if (loader.checkEgoNetworkValidation() == false)
return;
cntLikes = loader.cntLikesOfEgoUser;
}
loader.graphConfiguration(methodology, fold);
// Nodes and edges of graph
Dictionary<int, Node> nodes = loader.allNodes;
Dictionary<int, List<ForwardLink>> edges = loader.allLinks;
// Exeption: for the case that mention count is included when the friendship is none
if (methodology == Methodology.INCL_MENTIONCOUNT
|| methodology == Methodology.EXCL_FRIENDSHIP
|| methodology == Methodology.INCL_FOLLOWSHIP_ON_THIRDPARTY_AND_MENTIONCOUNT) {
foreach (List<ForwardLink> forwardLinks in edges.Values) {
List<int> indFriendshipLinks = new List<int>();
for (int i = 0; i < forwardLinks.Count; i++) {
if (forwardLinks[i].type == EdgeType.FRIENDSHIP)
indFriendshipLinks.Add(i);
}
foreach (int i in indFriendshipLinks) {
ForwardLink forwardLink = forwardLinks[i];
if (forwardLink.type == EdgeType.FRIENDSHIP) {
forwardLink.type = EdgeType.UNDEFINED;
forwardLinks[i] = forwardLink;
}
}
}
}
// Make a graph structure to run Random Walk with Restart algorithm
Graph graph = new Graph(nodes, edges);
graph.buildGraph();
// Get recommendation list
Recommender recommender = new Recommender(graph);
var recommendation = recommender.Recommendation(0, 0.15f, nIterations);
//// temp
//lock (Program.locker) {
// StreamWriter logger = new StreamWriter(Program.dirData + "rank.dat", true);
// logger.WriteLine(methodology);
// for (int i = 0; i < recommendation.Count; i++)
// logger.WriteLine(recommendation[i].Key + ":\t" + recommendation[i].Value);
// logger.Close();
//}
// Get evaluation result
int nHits = 0;
double sumPrecision = 0;
for (int i = 0; i < recommendation.Count; i++) {
if (loader.testSet.Contains(recommendation[i].Key)) {
nHits += 1;
sumPrecision += (double)nHits / (i + 1);
}
}
// Add current result to final one
foreach (EvaluationMetric metric in metrics) {
switch (metric) {
case EvaluationMetric.HIT:
finalResult[metric] += nHits; break;
case EvaluationMetric.AVGPRECISION:
finalResult[metric] += (nHits == 0) ? 0 : sumPrecision / nHits; break;
}
}
}
lock (Program.locker) {
// Write the result of this ego network to file
StreamWriter logger = new StreamWriter(Program.dirData + "result.dat", true);
logger.Write(egoUser + "\t" + (int)methodology + "\t" + nFolds + "\t" + nIterations);
foreach (EvaluationMetric metric in metrics) {
switch (metric) {
case EvaluationMetric.HIT:
logger.Write("\t" + (int)finalResult[metric] + "\t" + cntLikes); break;
case EvaluationMetric.AVGPRECISION:
logger.Write("\t" + (finalResult[metric] / nFolds)); break;
}
}
logger.WriteLine();
logger.Close();
}
}
} catch (FileNotFoundException e) {
Console.WriteLine(e);
} finally {
Program.semaphore.Release();
}
}
