/** Saves the {@link BasisMapping} created from the {@link OccurrenceCounter}. */
protected void saveSSpace(SemanticSpace sspace, File outputFile) throws IOException {
BasisMapping<String, String> savedTerms = new StringBasisMapping();
for (String term : sspace.getWords()) savedTerms.getDimension(term);
ObjectOutputStream ouStream = new ObjectOutputStream(new FileOutputStream(outputFile));
ouStream.writeObject(savedTerms);
ouStream.close();
}
/** {@inheritDoc} */
public void processSpace(Properties properties) {
SparseMatrix cleanedMatrix = (SparseMatrix) transform.transform(cooccurrenceMatrix);
for (String term : basis.keySet()) {
int index = basis.getDimension(term);
SparseDoubleVector sdv = cleanedMatrix.getRowVector(index);
double score = 0;
for (int i : sdv.getNonZeroIndices()) score += sdv.get(i);
wordScores.put(term, score);
}
}
/** {@inheritDoc} */
public void processDocument(BufferedReader document) throws IOException {
Queue<String> nextWords = new ArrayDeque<String>();
Queue<String> prevWords = new ArrayDeque<String>();
Iterator<String> documentTokens = IteratorFactory.tokenizeOrdered(document);
String focus = null;
// Rather than updating the matrix every time an occurrence is seen,
// keep a thread-local count of what needs to be modified in the
// matrix and update after the document has been processed. This
// saves potential contention from concurrent writes.
Map<Pair<Integer>, Double> matrixEntryToCount = new HashMap<Pair<Integer>, Double>();
// Load the first windowSize words into the Queue
for (int i = 0; i < windowSize && documentTokens.hasNext(); i++)
nextWords.offer(documentTokens.next());
while (!nextWords.isEmpty()) {
// Load the top of the nextWords Queue into the focus word
focus = nextWords.remove();
// Add the next word to nextWords queue (if possible)
if (documentTokens.hasNext()) nextWords.offer(documentTokens.next());
// If the filter does not accept this word, skip the semantic
// processing, continue with the next word
if (focus.equals(IteratorFactory.EMPTY_TOKEN)) {
int focusIndex = basis.getDimension(focus);
countOccurrences(nextWords, focusIndex, 1, matrixEntryToCount);
countOccurrences(prevWords, focusIndex, -prevWords.size(), matrixEntryToCount);
}
// last, put this focus word in the prev words and shift off the
// front if it is larger than the window
prevWords.offer(focus);
if (prevWords.size() > windowSize) prevWords.remove();
}
// Once the document has been processed, update the co-occurrence
// matrix accordingly.
for (Map.Entry<Pair<Integer>, Double> e : matrixEntryToCount.entrySet()) {
Pair<Integer> p = e.getKey();
cooccurrenceMatrix.addAndGet(p.x, p.y, e.getValue());
}
}
/**
* Adds a occurnce count for each term in {@code words} according to it's distance from the
* focus word.
*/
private void countOccurrences(
Queue<String> words,
int focusIndex,
int wordDistance,
Map<Pair<Integer>, Double> entryCounts) {
// Iterate through the words occurring after and add values
for (String term : words) {
// skip adding co-occurence values for words that are not
// accepted by the filter
if (!term.equals(IteratorFactory.EMPTY_TOKEN)) {
int index = basis.getDimension(term);
// Get the current number of times that the focus word has
// co-occurred with this word appearing after it. Weight
// the word appropriately based on distance
Pair<Integer> p = new Pair<Integer>(focusIndex, index);
double value = weighting.weight(wordDistance, windowSize);
Double curCount = entryCounts.get(p);
entryCounts.put(p, (curCount == null) ? value : value + curCount);
}
wordDistance++;
}
}
