/**
* combine weights according to expansion formula
*
* @param queryTerms
* @param relevantDocsTerms
* @param irrelevantDocsTerms
* @return
*/
public List<TermQuery> combine(
Map<String, TermQuery> queryTerms,
Map<String, TermQuery> relevantDocsTerms,
Map<String, TermQuery> irrelevantDocsTerms) {
// Add Terms of the relevant documents
for (Map.Entry<String, TermQuery> e : queryTerms.entrySet()) {
if (relevantDocsTerms.containsKey(e.getKey())) {
TermQuery tq = relevantDocsTerms.get(e.getKey());
tq.setBoost(tq.getBoost() + e.getValue().getBoost());
} else {
relevantDocsTerms.put(e.getKey(), e.getValue());
}
}
// Substract terms of irrelevant documents
for (Map.Entry<String, TermQuery> e : irrelevantDocsTerms.entrySet()) {
if (relevantDocsTerms.containsKey(e.getKey())) {
TermQuery tq = relevantDocsTerms.get(e.getKey());
tq.setBoost(tq.getBoost() - e.getValue().getBoost());
} else {
TermQuery tq = e.getValue();
tq.setBoost(-tq.getBoost());
relevantDocsTerms.put(e.getKey(), tq);
}
}
return new ArrayList<>(relevantDocsTerms.values());
}
/**
* Sets boost of termClaimsDescriptionAbstractTitles. boost = weight = factor(tf*idf)
*
* @param vecsTerms
* @param currentField
* @param factor - adjustment factor ( ex. alpha or beta )
* @param decayFactor
* @return
* @throws java.io.IOException
*/
public Map<String, TermQuery> setBoost(
Map<TermFreqVector, String> vecsTerms, String currentField, float factor, float decayFactor)
throws IOException {
Map<String, TermQuery> terms = new HashMap<>();
// setBoost for each of the terms of each of the docs
int i = 0;
float norm = (float) 1 / vecsTerms.size();
// System.out.println("--------------------------");
for (Map.Entry<TermFreqVector, String> e : vecsTerms.entrySet()) {
// Increase decay
String field = e.getValue();
TermFreqVector docTerms = e.getKey();
float decay = decayFactor * i;
// Populate terms: with TermQuries and set boost
for (String termTxt : docTerms.getTerms()) {
// Create Term
Term term = new Term(currentField, termTxt);
// Calculate weight
float tf = docTerms.getFreq(termTxt);
// float idf = ir.docFreq(termTitle);
int docs;
float idf;
if (sourceField.equals(PatentQuery.all)) {
docs = ir.getDocCount(field);
idf = (float) Math.log10((double) docs / (ir.docFreq(new Term(field, termTxt)) + 1));
} else {
docs = ir.getDocCount(sourceField);
idf =
(float) Math.log10((double) docs / (ir.docFreq(new Term(sourceField, termTxt)) + 1));
}
float weight = tf * idf;
// System.out.println(term.text() + " -> tf= " + tf + " idf= " + idf + "
// tfidf= " + weight);
// Adjust weight by decay factor
weight = weight - (weight * decay);
// Create TermQuery and add it to the collection
TermQuery termQuery = new TermQuery(term);
// Calculate and set boost
float boost;
if (vecsTerms.size() == 1) {
boost = factor * tf;
} else {
boost = factor;
}
if (boost != 0) {
termQuery.setBoost(boost * norm);
if (terms.containsKey(termTxt)) {
TermQuery tq = terms.get(termTxt);
tq.setBoost(tq.getBoost() + termQuery.getBoost());
} else {
terms.put(termTxt, termQuery);
}
}
}
i++;
}
return terms;
}
private static JsonObject convertTermQuery(TermQuery termQuery) {
String field = termQuery.getTerm().field();
JsonObject matchQuery =
JsonBuilder.object()
.add(
"term",
JsonBuilder.object()
.add(
field,
JsonBuilder.object()
.addProperty("value", termQuery.getTerm().text())
.addProperty("boost", termQuery.getBoost())))
.build();
return wrapQueryForNestedIfRequired(field, matchQuery);
}
public Map<String, Float> getRocchioVector(String currentField) throws IOException {
Map<String, Float> out = new HashMap<>();
float beta = parameters.get(RocchioQueryExpansion.ROCCHIO_BETA_FLD);
float gamma = parameters.get(RocchioQueryExpansion.ROCCHIO_GAMMA_FLD);
float decay = parameters.get(RocchioQueryExpansion.DECAY_FLD);
Map<String, TermQuery> relevantDocsTerms =
setBoost(docsTermVectorReldocs, currentField, beta, decay);
Map<String, TermQuery> irrrelevantDocsTerms =
setBoost(docsTermVectorIrreldocs, currentField, gamma, decay);
List<TermQuery> expandedQueryTerms =
combine(
new HashMap<String, TermQuery>(), relevantDocsTerms, new HashMap<String, TermQuery>());
for (TermQuery tq : expandedQueryTerms) {
out.put(tq.getTerm().text(), tq.getBoost());
}
return out;
}
private void _includeIfUnique(
BooleanQuery booleanQuery,
boolean like,
QueryParser queryParser,
Query query,
BooleanClause.Occur occur) {
if (query instanceof TermQuery) {
Set<Term> terms = new HashSet<Term>();
TermQuery termQuery = (TermQuery) query;
termQuery.extractTerms(terms);
float boost = termQuery.getBoost();
for (Term term : terms) {
String termValue = term.text();
if (like) {
termValue = termValue.toLowerCase(queryParser.getLocale());
term = term.createTerm(StringPool.STAR.concat(termValue).concat(StringPool.STAR));
query = new WildcardQuery(term);
} else {
query = new TermQuery(term);
}
query.setBoost(boost);
boolean included = false;
for (BooleanClause booleanClause : booleanQuery.getClauses()) {
if (query.equals(booleanClause.getQuery())) {
included = true;
}
}
if (!included) {
booleanQuery.add(query, occur);
}
}
} else if (query instanceof BooleanQuery) {
BooleanQuery curBooleanQuery = (BooleanQuery) query;
BooleanQuery containerBooleanQuery = new BooleanQuery();
for (BooleanClause booleanClause : curBooleanQuery.getClauses()) {
_includeIfUnique(
containerBooleanQuery,
like,
queryParser,
booleanClause.getQuery(),
booleanClause.getOccur());
}
if (containerBooleanQuery.getClauses().length > 0) {
booleanQuery.add(containerBooleanQuery, occur);
}
} else {
boolean included = false;
for (BooleanClause booleanClause : booleanQuery.getClauses()) {
if (query.equals(booleanClause.getQuery())) {
included = true;
}
}
if (!included) {
booleanQuery.add(query, occur);
}
}
}
public void testStarParsing() throws Exception {
final int[] type = new int[1];
QueryParser qp =
new QueryParser(
TEST_VERSION_CURRENT,
"field",
new MockAnalyzer(random, MockTokenizer.WHITESPACE, false)) {
@Override
protected Query getWildcardQuery(String field, String termStr) throws ParseException {
// override error checking of superclass
type[0] = 1;
return new TermQuery(new Term(field, termStr));
}
@Override
protected Query getPrefixQuery(String field, String termStr) throws ParseException {
// override error checking of superclass
type[0] = 2;
return new TermQuery(new Term(field, termStr));
}
@Override
protected Query getFieldQuery(String field, String queryText, boolean quoted)
throws ParseException {
type[0] = 3;
return super.getFieldQuery(field, queryText, quoted);
}
};
TermQuery tq;
tq = (TermQuery) qp.parse("foo:zoo*");
assertEquals("zoo", tq.getTerm().text());
assertEquals(2, type[0]);
tq = (TermQuery) qp.parse("foo:zoo*^2");
assertEquals("zoo", tq.getTerm().text());
assertEquals(2, type[0]);
assertEquals(tq.getBoost(), 2, 0);
tq = (TermQuery) qp.parse("foo:*");
assertEquals("*", tq.getTerm().text());
assertEquals(1, type[0]); // could be a valid prefix query in the future too
tq = (TermQuery) qp.parse("foo:*^2");
assertEquals("*", tq.getTerm().text());
assertEquals(1, type[0]);
assertEquals(tq.getBoost(), 2, 0);
tq = (TermQuery) qp.parse("*:foo");
assertEquals("*", tq.getTerm().field());
assertEquals("foo", tq.getTerm().text());
assertEquals(3, type[0]);
tq = (TermQuery) qp.parse("*:*");
assertEquals("*", tq.getTerm().field());
assertEquals("*", tq.getTerm().text());
assertEquals(1, type[0]); // could be handled as a prefix query in the future
tq = (TermQuery) qp.parse("(*:*)");
assertEquals("*", tq.getTerm().field());
assertEquals("*", tq.getTerm().text());
assertEquals(1, type[0]);
}