/**
* Builds a CooccurenceMap by iterating over the documents of the collection. It counts document
* co-occurence, i.e. it doesn't consider the frequency of two terms in a document. Complexity:
* O(d * t *t/2) = O(d t^2) where d is the number of documents in the collection and t is the
* average number of terms per documents. Note that t = avg doc len
*/
public void build_full_cooccurencemap_docversion() throws IOException {
PostingIndex di = index.getDirectIndex();
DocumentIndex doi = index.getDocumentIndex();
Lexicon<String> lex = index.getLexicon();
for (int docid = 0; docid < doi.getNumberOfDocuments(); docid++) {
if (docid % 1000 == 0)
System.out.println(
"Processing... " + 100.0 * ((double) docid) / doi.getNumberOfDocuments() + "%");
IterablePosting postings = di.getPostings(doi.getDocumentEntry(docid));
Vector<String> seenterms = new Vector<String>();
while (postings.next() != IterablePosting.EOL) {
Map.Entry<String, LexiconEntry> lee = lex.getLexiconEntry(postings.getId());
String termw = lee.getKey();
if (lee.getValue().getFrequency() < this.rarethreshold
|| lee.getValue().getFrequency() > this.topthreshold) continue;
HashMap<String, Integer> w_cooccurence = new HashMap<String, Integer>();
if (this.cooccurencemap.containsKey(termw)) {
w_cooccurence = this.cooccurencemap.get(termw);
this.cooccurencemap.remove(termw);
}
Iterator<String> it = seenterms.iterator();
while (it.hasNext()) {
String termu = it.next();
int count = 1;
if (w_cooccurence.containsKey(termu)) {
count = count + w_cooccurence.get(termu);
w_cooccurence.remove(termu);
}
w_cooccurence.put(termu, count);
// System.out.println(termw + ": " + w_cooccurence);
// and now I need to do the symmetric
HashMap<String, Integer> u_cooccurence = new HashMap<String, Integer>();
if (cooccurencemap.containsKey(termu)) {
u_cooccurence = cooccurencemap.get(termu);
cooccurencemap.remove(termu);
}
int countu = 1;
if (u_cooccurence.containsKey(termw)) {
countu = countu + u_cooccurence.get(termw);
u_cooccurence.remove(termw);
}
u_cooccurence.put(termw, count);
cooccurencemap.put(termu, u_cooccurence);
// System.out.println(termu + ": " + u_cooccurence);
}
cooccurencemap.put(termw, w_cooccurence);
seenterms.add(termw); // I add only the termw that are within the thresholds
}
}
}
/**
* Runs the actual query expansion
*
* @see
* org.terrier.querying.PostProcess#process(org.terrier.querying.Manager,org.terrier.querying.SearchRequest)
*/
public void process(Manager manager, SearchRequest q) {
Index index = getIndex(manager);
lastIndex = index;
documentIndex = index.getDocumentIndex();
invertedIndex = index.getInvertedIndex();
lexicon = index.getLexicon();
collStats = index.getCollectionStatistics();
directIndex = index.getDirectIndex();
metaIndex = index.getMetaIndex();
if (directIndex == null) {
logger.error("This index does not have a direct index. Query expansion disabled!!");
return;
}
logger.debug("Starting query expansion post-processing.");
// get the query expansion model to use
String qeModel = q.getControl("qemodel");
if (qeModel == null || qeModel.length() == 0) {
logger.warn(
"qemodel control not set for QueryExpansion" + " post process. Using default model Bo1");
qeModel = "Bo1";
}
setQueryExpansionModel(getQueryExpansionModel(qeModel));
if (logger.isDebugEnabled()) {
logger.info("query expansion model: " + QEModel.getInfo());
}
MatchingQueryTerms queryTerms = ((Request) q).getMatchingQueryTerms();
if (queryTerms == null) {
logger.warn("No query terms for this query. Skipping QE");
return;
}
// get the expanded query terms
try {
expandQuery(queryTerms, (Request) q);
} catch (IOException ioe) {
logger.error("IOException while expanding query, skipping QE", ioe);
return;
}
if (logger.isDebugEnabled()) {
logger.info("query length after expansion: " + queryTerms.length());
logger.info("Expanded query: ");
}
final String[] newQueryTerms = queryTerms.getTerms();
StringBuilder newQuery = new StringBuilder();
for (int i = 0; i < newQueryTerms.length; i++) {
try {
if (logger.isDebugEnabled()) {
logger.info(
(i + 1)
+ ": "
+ newQueryTerms[i]
+ ", normalisedFrequency: "
+ Rounding.toString(queryTerms.getTermWeight(newQueryTerms[i]), 4));
}
newQuery.append(newQueryTerms[i]);
newQuery.append('^');
newQuery.append(Rounding.toString(queryTerms.getTermWeight(newQueryTerms[i]), 9));
newQuery.append(' ');
} catch (NullPointerException npe) {
logger.error("Nullpointer exception occured in Query Expansion dumping of new Query", npe);
}
}
logger.debug("NEWQUERY " + q.getQueryID() + " " + newQuery.toString());
lastExpandedQuery = newQuery.toString();
q.setControl("QE.ExpandedQuery", newQuery.toString());
final boolean no2ndPass =
Boolean.parseBoolean(ApplicationSetup.getProperty("qe.no.2nd.matching", "false"));
if (no2ndPass) {
return;
}
// run retrieval process again for the expanded query
logger.info("Accessing inverted file for expanded query " + q.getQueryID());
manager.runMatching(q);
}
