querqy.lucene.PhraseBoosting Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of querqy-lucene Show documentation
Show all versions of querqy-lucene Show documentation
Querqy library for query rewriting for Lucene
The newest version!
package querqy.lucene;
import org.apache.lucene.analysis.Analyzer;
import org.apache.lucene.search.BooleanQuery;
import org.apache.lucene.search.Query;
import org.apache.lucene.util.QueryBuilder;
import querqy.ComparableCharSequence;
import querqy.model.BooleanClause;
import querqy.model.DisjunctionMaxClause;
import querqy.model.DisjunctionMaxQuery;
import querqy.model.QuerqyQuery;
import querqy.model.Term;
import java.util.ArrayList;
import java.util.LinkedList;
import java.util.List;
import java.util.Optional;
/**
* Facilitate boosting of (sub)phrases created from query terms, similar to Solr's (e)dismax pf, pf2, pf3.
*/
public class PhraseBoosting {
public enum NGramType {
PHRASE(0), BI_GRAM(2), TRI_GRAM(3);
public final int nGramSize;
NGramType(final int nGramSize) {
this.nGramSize = nGramSize;
}
}
/**
* Modified copy of org.apache.solr.search.FieldParams
*/
public static class PhraseBoostFieldParams {
private final NGramType nGramType;
private final int slop;
private final float boost;
private final String field;
public PhraseBoostFieldParams(final String field, final NGramType nGramType, final int slop, final float boost) {
this.nGramType = nGramType;
this.slop = slop;
this.boost = boost;
this.field = field;
}
public NGramType getNGramType() {
return nGramType;
}
public int getSlop() {
return slop;
}
public float getBoost() {
return boost;
}
public String getField() {
return field;
}
}
/**
* Make a phrase boost query.
* Unlike in Solr, the scores of the queries from the (sub-)phrase levels, such as bi-grams, tri-grams and
* complete phrase, are not simply added up but combined using a
* {@link org.apache.lucene.search.DisjunctionMaxQuery}. The phraseBoostTiebreaker parameter defines how
* the scores for these levels are aggregated: a value of 1.0 will add up the scores, a value of 0.0 will pick the
* highest score.
*
* @param userQuery The main query from which to generate phrases
* @param phraseBoostFieldParams A list of phrase boost parameters
* @param phraseBoostTiebreaker The tie breaker for aggregating scores from the different phrase length levels
* @param queryAnalyzer The Lucene query analyzer
* @return An optional query that contains the phrase boost queries if any such query could be created
*/
public static Optional makePhraseFieldsBoostQuery(final QuerqyQuery userQuery,
final List phraseBoostFieldParams,
final float phraseBoostTiebreaker,
final Analyzer queryAnalyzer) {
if (userQuery instanceof querqy.model.Query) {
final List clauses = ((querqy.model.Query) userQuery).getClauses();
if (clauses.size() > 1) {
if (!phraseBoostFieldParams.isEmpty()) {
final List sequence = new LinkedList<>();
for (final querqy.model.BooleanClause clause : clauses) {
if (clause instanceof DisjunctionMaxQuery) {
final DisjunctionMaxQuery dmq = (DisjunctionMaxQuery) clause;
if (dmq.occur != querqy.model.SubQuery.Occur.MUST_NOT) {
for (final DisjunctionMaxClause dmqClause : dmq.getClauses()) {
if (dmqClause instanceof Term) {
final ComparableCharSequence value = ((Term) dmqClause).getValue();
final int length = value.length();
final StringBuilder sb = new StringBuilder(length);
for (int i = 0; i < length; i++) {
sb.append(value.charAt(i));
}
sequence.add(sb.toString());
break;
}
}
}
}
}
if (sequence.size() > 1) {
final List disjuncts = new LinkedList<>();
final QueryBuilder queryBuilder = new QueryBuilder(queryAnalyzer);
final List[] shingles = new List[4];
String pf = null;
for (final PhraseBoosting.PhraseBoostFieldParams fieldParams : phraseBoostFieldParams) {
final PhraseBoosting.NGramType nGramType = fieldParams.getNGramType();
final int slop = fieldParams.getSlop();
final String fieldname = fieldParams.getField();
if (nGramType == PhraseBoosting.NGramType.PHRASE) {
if (pf == null) {
final StringBuilder sb = new StringBuilder(sequence.size() * 7);
for (final String term : sequence) {
if (sb.length() > 0) {
sb.append(' ');
}
sb.append(term);
}
pf = sb.toString();
}
final Query pq = queryBuilder.createPhraseQuery(fieldname, pf, slop);
if (pq != null) {
disjuncts.add(LuceneQueryUtil.boost(pq, fieldParams.getBoost()));
}
} else if (nGramType.nGramSize <= sequence.size()) {
if (shingles[nGramType.nGramSize] == null) {
shingles[nGramType.nGramSize] = new LinkedList<>();
for (int i = 0, lenI = sequence.size() - nGramType.nGramSize + 1; i < lenI; i++) {
final StringBuilder sb = new StringBuilder();
for (int j = i, lenJ = j + nGramType.nGramSize; j < lenJ; j++) {
if (sb.length() > 0) {
sb.append(' ');
}
sb.append(sequence.get(j));
}
shingles[nGramType.nGramSize].add(sb.toString());
}
}
final List nGramQueries = new ArrayList<>(shingles[nGramType.nGramSize].size());
for (final String nGram : shingles[nGramType.nGramSize]) {
final Query pq = queryBuilder.createPhraseQuery(fieldname, nGram, slop);
if (pq != null) {
nGramQueries.add(pq);
}
}
switch (nGramQueries.size()) {
case 0: break;
case 1: {
final Query nGramQuery = nGramQueries.get(0);
disjuncts.add(LuceneQueryUtil.boost(nGramQuery, fieldParams.getBoost()));
break;
}
default:
final BooleanQuery.Builder builder = new BooleanQuery.Builder();
for (final Query nGramQuery : nGramQueries) {
builder.add(nGramQuery, org.apache.lucene.search.BooleanClause.Occur.SHOULD);
}
final BooleanQuery bq = builder.build();
disjuncts.add(LuceneQueryUtil.boost(bq, fieldParams.getBoost()));
}
}
}
switch (disjuncts.size()) {
case 0: break;
case 1: return Optional.of(disjuncts.get(0));
default :
return Optional.of(new org.apache.lucene.search.DisjunctionMaxQuery(disjuncts,
phraseBoostTiebreaker));
}
}
}
}
}
return Optional.empty();
}
}
© 2015 - 2024 Weber Informatics LLC | Privacy Policy