public static double idfModifiedCos(IDF idf, Sentence firstSentence, Sentence secondSentence)
{
double idfModifiedCosine = 0;
HashSet<string> commonWords = new HashSet<string>();
foreach (string aWord in firstSentence.words)
{
if (secondSentence.words.Contains(aWord))
commonWords.Add(aWord);
}
double numerator = 0;
foreach (string aWord in commonWords)
{
numerator += (termFrequency(firstSentence, aWord) * termFrequency(secondSentence, aWord) * Math.Pow(idf.get(aWord), 2));
}
double denominator1 = 0;
foreach (string aWord in firstSentence.words)
{
//if (docStats.wordRefs[aWord] != null)
denominator1 += Math.Pow(termFrequency(firstSentence, aWord) * idf.get(aWord), 2);
}
denominator1 = Math.Sqrt(denominator1);
double denominator2 = 0;
foreach (string aWord in secondSentence.words)
{
//if (docStats.wordRefs[aWord] != null)
denominator2 += Math.Pow(termFrequency(secondSentence, aWord) * idf.get(aWord), 2);
}
denominator2 = Math.Sqrt(denominator2);
idfModifiedCosine = numerator / (denominator1 * denominator2);
return (idfModifiedCosine);
}
public static IDF fromFiles(string[] files)
{
DocsStatistics docStats = new DocsStatistics();
DocumentProcessor docProcessor = new DocumentProcessor();
int i = 0;
foreach (string file in files)
{
++i;
//processFile(docStats, file);
//*
string fileContent = File.ReadAllText(file, Encoding.Default);
Document doc = docProcessor.process(fileContent);
docStats.addDocument(doc);
/*
* if ((i % 1000) == 0)
* {
* System.GC.Collect();
* Trace.write("Done for : " + i);
* }
* //*/
//*/
//doc = null;
}
IDF idf = new IDF();
foreach (string word in docStats.wordsCount.Keys)
{
//double wordRefCount = docStats.wordRefs[firstWord] == null ? 0 : ((HashSet<Document>)docStats.wordRefs[firstWord]).Count;
double wordRefCount = docStats.wordRefsCount[word] == null ? 0 : ((int)docStats.wordRefsCount[word]);
double wordIdf = Math.Log(docStats.docCount / (wordRefCount));
idf.idf[word] = wordIdf;
}
return(idf);
}
public static double[][] generateIdfModifiedCosineMatrix(IDF idf, ArrayList sentences)
{
double[][] idfModifiedCosine = new double[sentences.Count][];
for (int i = 0; i < sentences.Count; i++)
{
idfModifiedCosine[i] = new double[sentences.Count];
}
for (int i = 0; i < sentences.Count; i++)
{
Sentence firstSent = (Sentence)sentences[i];
for (int j = 0; j < sentences.Count; j++)
{
// same sentence then 1
//*
if (i == j)
{
idfModifiedCosine[i][j] = 1;
continue;
}
//*/
// has been processed before
if (idfModifiedCosine[i][j] != 0)
continue;
Sentence secondSent = (Sentence)sentences[j];
idfModifiedCosine[i][j] = idfModifiedCos(idf, firstSent, secondSent);
idfModifiedCosine[j][i] = idfModifiedCosine[i][j];
}
}
return (idfModifiedCosine);
}
public static double sim(IDF idf, Hashtable first, Hashtable second)
{
double similarity = 0;
HashSet<string> commonWords = SummaryUtil.getCommonWords(new ArrayList(first.Keys), new ArrayList(second.Keys));
double numerator = 0;
foreach (string aWord in commonWords)
{
numerator += ((double)first[aWord] * (double)second[aWord] * idf.get(aWord));
}
double denominator1 = 0;
foreach (string aWord in first.Keys)
{
//if (docStats.wordRefs[aWord] != null)
denominator1 += Math.Pow((double)first[aWord], 2);
}
denominator1 = Math.Sqrt(denominator1);
double denominator2 = 0;
foreach (string aWord in second.Keys)
{
//if (docStats.wordRefs[aWord] != null)
denominator2 += Math.Pow((double)second[aWord], 2);
}
denominator2 = Math.Sqrt(denominator2);
similarity = numerator / (denominator1 * denominator2);
return (similarity);
}
public static void setInstance(IDF idf)
{
_idf = idf;
}
public static IDF fromFiles(string[] files)
{
DocsStatistics docStats = new DocsStatistics();
DocumentProcessor docProcessor = new DocumentProcessor();
int i = 0;
foreach (string file in files)
{
++i;
//processFile(docStats, file);
//*
string fileContent = File.ReadAllText(file, Encoding.Default);
Document doc = docProcessor.process(fileContent);
docStats.addDocument(doc);
/*
if ((i % 1000) == 0)
{
System.GC.Collect();
Trace.write("Done for : " + i);
}
//*/
//*/
//doc = null;
}
IDF idf = new IDF();
foreach (string word in docStats.wordsCount.Keys)
{
//double wordRefCount = docStats.wordRefs[firstWord] == null ? 0 : ((HashSet<Document>)docStats.wordRefs[firstWord]).Count;
double wordRefCount = docStats.wordRefsCount[word] == null ? 0 : ((int)docStats.wordRefsCount[word]);
double wordIdf = Math.Log(docStats.docCount / (wordRefCount));
idf.idf[word] = wordIdf;
}
return (idf);
}
public ArrayList buildCentroids(ArrayList docs, IDF idfdb)
{
ArrayList centroids = new ArrayList();
foreach (Document doc in docs)
{
ArrayList currDoc = new ArrayList();
currDoc.Add(doc);
DocsStatistics currDocStats = DocsStatistics.generateStatistics(currDoc);
Hashtable docVector = new Hashtable();
foreach (DictionaryEntry entry in currDocStats.wordsCount)
{
string word = (string)entry.Key;
int count = (int)entry.Value;
//double idf = CentroidAlgorithm2.idf(allDocStats, firstWord);
double idf = idfdb.get(word);
if (idf < this.idfThreshold)
continue;
double tfidf = ((double)count) * idf;
docVector[word] = tfidf;
}
if (centroids.Count == 0)
{
Centroid centroid = new Centroid(docVector, this.keepWords);
centroid.noOfDocuments = 1;
centroids.Add(centroid);
}
else
{
Centroid nearestCentroid = null;
double maxSimilarity = double.MinValue;
foreach (Centroid centroid in centroids)
{
double similarity = sim(IDF.getInstance(), centroid.values, docVector);
if (similarity > simThreshold)
{
if (similarity > maxSimilarity)
{
maxSimilarity = similarity;
nearestCentroid = centroid;
}
}
}
if (nearestCentroid == null)
{
nearestCentroid = new Centroid(docVector, this.keepWords);
centroids.Add(nearestCentroid);
}
else
{
nearestCentroid.addDocument(docVector);
}
}
}
// Apply the KEEP_WORDS parameter for each centroid
/*
foreach (Centroid centroid in centroids)
{
Hashtable centroidValues = centroid.values;
DictionaryEntry[] centValuesArr = new DictionaryEntry[centroids.Count];
centroidValues.CopyTo(centValuesArr, 0);
Array.Sort(centValuesArr, new DictionaryEntryValueComparer());
Array.Reverse(centValuesArr);
DictionaryEntry[] finalCentroidValuesArr = new DictionaryEntry[this.keepWords];
Array.Copy(centValuesArr, finalCentroidValuesArr, this.keepWords);
Hashtable finalCentroidValues = new Hashtable();
foreach (DictionaryEntry entry in finalCentroidValuesArr)
{
finalCentroidValues.Add(entry.Key, entry.Value);
}
centroid.values = finalCentroidValues;
}
//*/
//*
foreach (Centroid centroid in centroids)
{
centroid.applyKeepWords();
}
//*/
// Trace
/*
int i = 0;
foreach (Centroid centroid in centroids)
{
Trace.write("Centroid #" + (++i));
foreach (DictionaryEntry entry in centroid.values)
{
Trace.write(entry.Key + " : " + entry.Value);
}
}
//*/
return (centroids);
}
public static double calcSentenceWeight(IDF idf, Document doc, Sentence sent)
{
Trace.write(sent.fullText);
double weight = 0;
// 1: ScLead
double sclead = 0;
if (sent == doc.sentences[0])
sclead = 2;
else
sclead = 1;
Trace.write("SCLead : " + sclead);
// 2: ScTitle
double sctitle = 0;
foreach (string aWord in sent.words)
{
//double tf = docStats.wordsCount[aWord] == null ? 0 : (((int)docStats.wordsCount[aWord]) / docStats.wordTotal);
//double tf = termFrequency(docStats, aWord);
double tf = termFrequency(sent, aWord);
if (doc.title != null)
{
if (doc.title.words.ToArray().Contains(aWord))
sctitle += (2 * tf);
}
}
Trace.write("SCTitle : " + sctitle);
// 3: sccue
double sccue = 0;
foreach (string aWord in sent.words)
{
if (CueWords.getInstance(Conf.CUE_WORDS_PATH).contains(aWord))
{
double tf = termFrequency(sent, aWord);
sccue += tf;
}
}
Trace.write("SCCue : " + sccue);
// 4: sctfidf
double sctfidf = 0;
foreach (string aWord in sent.words)
{
//double tf = termFrequency(docStats, aWord);
double tf = termFrequency(sent, aWord);
//if (docStats.wordRefs[aWord] != null && tf != 0)
if (tf != 0)
//sctfidf += (((tf - 1) / tf) * Math.Log(docStats.docCount / ((HashSet<Document>)docStats.wordRefs[aWord]).Count));
sctfidf += (((tf - 1) / tf) * idf.get(aWord));
}
//sctfidf = sctfidf / docStats.sentCount;
//sctfidf = sctfidf / doc.sentences.Count;
//sctfidf = sctfidf / sent.words.Length;
sctfidf = sctfidf / sent.words.Count;
Trace.write("SCTFIDF : " + sctfidf);
weight = sclead + sctitle + sccue + sctfidf;
sent.weight = weight;
Trace.write("Weight : " + weight);
return (weight);
}
public static void setInstance(IDF idf)
{
_idf = idf;
}
private void IDFPreprocessDocsDialog_FileOk(object sender, CancelEventArgs e)
{
string[] fileNames = this.IDFPreprocessDocsDialog.FileNames;
/*
ArrayList docs = new ArrayList();
this.progressBar.Show();
this.progressBar.Minimum = 0;
this.progressBar.Maximum = fileNames.Length + (fileNames.Length / 4);
this.progressBar.Value = 0;
foreach (string fileName in fileNames)
{
string fileText = File.ReadAllText(fileName, Encoding.Default);
Document doc = docProcessor.process(fileText);
docs.Add(doc);
this.progressBar.Increment(1);
}
this.idf = IDF.fromDocuments(docs);
this.progressBar.Value = this.progressBar.Maximum;
this.progressBar.Hide();
//*/
this.idf = IDF.IDFGenerator.fromFiles(fileNames);
this.IDFSaveDialog.ShowDialog();
}
