/**
* Generates an extraction template <em>tt</em>, which is the set of non-noisy nodes by
* depth-first traversal of the maximum matching tree <em>tm</em> which only contains nodes that
* matched in both ta and tb trees.
*
* <p>If two pages have the same sub-trees, these fractions are considered noise. As anchor-tags
* are hidden noise text, the anchor-text-ratio of a sub-segment has to exceed a given threshold
* to be excluded.
*
* <p>Moreover punctuation metrics are used to further separate noise from content.
*
* @param i The index of the i-th element of the maximum matching tree tm
* @return The generated extraction template build from the two provided sources
*/
private Deque<Token> treeExtraction(int i) {
Token matchedNode;
if (this.tm.get(i).getMatchedNode() instanceof Tag) {
matchedNode = new Tag(this.tm.get(i).getMatchedNode());
} else {
matchedNode = new Word(this.tm.get(i).getMatchedNode());
}
if (this.tm != null
&& !this.tm.get(i).getSubtreeText().equals(matchedNode.getSubtreeText())
&& this.tm.get(i).getAnchorTextRatio() <= 1.3 * this.ta.get(0).getAnchorTextRatio()
&& !(this.tm.get(i).getPunctNum() == 0 && this.tm.get(i).getSegNum() >= 3)) {
if (this.tm.get(i).getSibNo() == matchedNode.getSibNo()) {
this.tt.add(this.tm.get(i));
for (int j = 1; j < this.tm.get(i).getChildren().length; j++) {
this.treeExtraction(this.tm.get(i).getChildren()[j].getNo());
}
}
// expand extraction range of tt to extract content that did not
// matched in the maximum matching tree tm as they occurred in tb
// but not in ta
else if (this.tm.get(this.tm.get(i).getParentNo()).getSibNo()
== this.tb.get(matchedNode.getParentNo()).getSibNo()) {
while (this.tt.peekLast().getNo() != this.tm.get(i).getParentNo()) {
this.tt.removeLast();
}
}
}
return this.tt;
}
/**
* Uses the extraction template generated by the {@link #templateGeneration} process and utilizes
* some own characteristics of the target page pt to extract the content of it which first needs
* to be parsed to a DOM tree <em>tp</em>.
*
* @param tt The previouslz generated extraction template
* @param tp_i The root DOM element of the document to predict the main content for
*/
public void newsContentExtraction(Deque<Token> tt, Token tp_i) {
// templateNode <-- tt.firstElement
Token templateNode = tt.getFirst();
if (templateNode.getName().equals(tp_i.getName())
&& templateNode.getLevel() == tp_i.getLevel()
&& templateNode.getSibNo() == tp_i.getSibNo()) {
tt.removeFirst();
if (tt.size() > 0) {
Token nextTemplateNode = tt.getFirst();
if (tp_i.getChildren() == null || tp_i.getChildren().length == 0) {
LOG.info(tp_i.getText());
while (nextTemplateNode.getParentNo() == templateNode.getNo()) {
tt.removeFirst();
templateNode = nextTemplateNode;
nextTemplateNode = tt.getFirst();
}
} else {
if (nextTemplateNode.getParentNo() != templateNode.getNo()) {
System.out.println(this.deleteEmbeddedNoise(tp_i.getSubtreeText()));
}
for (int j = 0; j < tp_i.getChildren().length; j++) {
this.newsContentExtraction(tt, tp_i.getChildren()[j]);
}
}
} else {
LOG.info(this.deleteEmbeddedNoise(tp_i.getSubtreeText()));
}
}
}
/**
* Matched children backtracking algorithm
*
* <p>Besides finding the matched children using ta's <em>matchedMatrix</em> field, the main job
* is to modify the initial <em>matchedMatrix</em> of each matched child.
*
* @param p The index of the p-th element of the DOM-tree ta
* @param matchedMatrix The matched tree path matrix of ta for node p
* @param i The x-position in the matchedMatrix to look for the value
* @param j The y-position in the matchedMatrix to look for the value
*/
private void matchedChildrenBacktracking(int p, int[][] matchedMatrix, int i, int j) {
// slight modification as if the 1st page is larger than the second or a
// segment of the 1st page contains more children than the 2nd page,
// j might become negative while i is 0.
// To my understanding if either of i or j gets 0 the border of the
// matrix was reached.
if (i == 0 && j == 0) {
return;
}
if (matchedMatrix[i][j] == MatchedMatrixValue.UP_LEFT.getValue()) {
this.matchedChildrenBacktracking(p, matchedMatrix, i - 1, j - 1);
// contains elements from 1st page
Token child = this.ta.get(p).getChildren()[i - 1];
// contains elements from 2nd page
Token comparedNode = this.tb.get(child.getComparedNodes().peek().getParentNo());
while (this.ta.get(p).getMatchedNode() != comparedNode) {
int k = comparedNode.getChildren().length;
int n = this.ta.get(p).getChildren().length;
for (int _i = 0; _i < k - 1; _i++)
// Delete child.comparedNodes.firstElement
{
child.getComparedNodes().poll();
}
for (int h = 0; h < n - 1; h++) {
this.ta.get(p).getChildren()[h].setComparedNodes(child.getComparedNodes());
for (int _i = 0; _i < k - 1; _i++)
// Delete ta[p].children[h].comparedMatrix.firstElement
{
this.ta.get(p).getChildren()[h].getComparedMatrix().poll();
}
}
if (child.getComparedNodes().size() > 0) {
comparedNode = this.tb.get(child.getComparedNodes().peek().getParentNo());
} else {
break;
}
}
// child.setMatchedNode(child.getComparedNodes().get(j-1));
child.setMatchedNode(comparedNode.getChildren()[j - 1]);
child.setMatchedMatrix(null);
if (child.getComparedMatrix().size() > 0) {
child.setMatchedMatrix(child.getComparedMatrix().get(j - 1));
}
// Add child to matchedChildren, tm[p].children
// matchedChildren.add(new HTMLNode(child));
this.matchedChildren.add(child);
// this.tm[p].addChild(new HTMLNode(child));
this.tm.get(p).addChild(child);
} else if (matchedMatrix[i][j] == MatchedMatrixValue.UP.getValue()) {
this.matchedChildrenBacktracking(p, matchedMatrix, i - 1, j);
} else {
this.matchedChildrenBacktracking(p, matchedMatrix, i, j - 1);
}
}
