/*
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
Code to search indices.
Table Of Contents
- Search Basics
- The Query Classes
- Scoring: Introduction
- Scoring: Basics
- Changing the Scoring
- Appendix: Search Algorithm
Search Basics
Lucene offers a wide variety of Query implementations, most of which are in this package, its subpackage (spans, or the queries module. These implementations can be combined in a wide
variety of ways to provide complex querying capabilities along with information about where matches took place in the document
collection. The Query Classes section below highlights some of the more important Query classes. For details
on implementing your own Query class, see Custom Queries -- Expert Level below.
To perform a search, applications usually call IndexSearcher.search(Query, int).
Once a Query has been created and submitted to the IndexSearcher, the scoring process begins. After some infrastructure setup, control finally passes to the Weight implementation and its Scorer or BulkScorer instances. See the Algorithm section for more notes on the process.
Query Classes
TermQuery
Of the various implementations of Query, the TermQuery is the easiest to understand and the most often used in applications. A TermQuery matches all the documents that contain the specified Term, which is a word that occurs in a certain Field. Thus, a TermQuery identifies and scores all Documents that have a Field with the specified string in it. Constructing a TermQuery is as simple as:
TermQuery tq = new TermQuery(new Term("fieldName", "term"));
In this example, the Query identifies all Documents that have the Field named "fieldName"
containing the word "term".
BooleanQuery
Things start to get interesting when one combines multiple TermQuery instances into a BooleanQuery. A BooleanQuery contains multiple BooleanClauses, where each clause contains a sub-query (Query instance) and an operator (from BooleanClause.Occur) describing how that sub-query is combined with the other clauses:
SHOULD — Use this operator when a clause can occur in the result set, but is not required. If a query is made up of all SHOULD clauses, then every document in the result set matches at least one of these clauses.
MUST — Use this operator when a clause is required to occur in the result set and should contribute to the score. Every document in the result set will match all such clauses.
FILTER — Use this operator when a clause is required to occur in the result set but should not contribute to the score. Every document in the result set will match all such clauses.
MUST NOT — Use this operator when a clause must not occur in the result set. No document in the result set will match any such clauses.
Boolean queries are constructed by adding two or more BooleanClause instances. If too many clauses are added, a TooManyClauses exception will be thrown during searching. This most often occurs when a Query is rewritten into a BooleanQuery with many TermQuery clauses, for example by WildcardQuery. The default setting for the maximum number of clauses is 1024, but this can be changed via the static method BooleanQuery.setMaxClauseCount(int). Phrases
Another common search is to find documents containing certain phrases. This
is handled three different ways:
-
PhraseQuery — Matches a sequence of Terms. PhraseQuery uses a slop factor to determine how many positions may occur between any two terms in the phrase and still be considered a match. The slop is 0 by default, meaning the phrase must match exactly.
-
MultiPhraseQuery — A more general form of PhraseQuery that accepts multiple Terms for a position in the phrase. For example, this can be used to perform phrase queries that also incorporate synonyms.
-
SpanNearQuery — Matches a sequence of other SpanQuery instances. SpanNearQuery allows for much more complicated phrase queries since it is constructed from other SpanQuery instances, instead of only TermQuery instances.
PointRangeQuery
The PointRangeQuery matches all documents that occur in a numeric range. For PointRangeQuery to work, you must index the values using a one of the numeric fields (IntPoint, LongPoint, FloatPoint, or DoublePoint).
PrefixQuery, WildcardQuery, RegexpQuery
While the PrefixQuery has a different implementation, it is essentially a special case of the WildcardQuery. The PrefixQuery allows an application to identify all documents with terms that begin with a certain string. The WildcardQuery generalizes this by allowing for the use of * (matches 0 or more characters) and ? (matches exactly one character) wildcards. Note that the WildcardQuery can be quite slow. Also note that WildcardQuery should not start with * and ?, as these are extremely slow.
Some QueryParsers may not allow this by default, but provide a setAllowLeadingWildcard method to remove that protection. The RegexpQuery is even more general than WildcardQuery, allowing an application to identify all documents with terms that match a regular expression pattern.
FuzzyQuery
A FuzzyQuery matches documents that contain terms similar to the specified term. Similarity is determined using Levenshtein distance.
This type of query can be useful when accounting for spelling variations in the collection.
Scoring — Introduction
Lucene scoring is the heart of why we all love Lucene. It is blazingly fast and it hides
almost all of the complexity from the user. In a nutshell, it works. At least, that is,
until it doesn't work, or doesn't work as one would expect it to work. Then we are left
digging into Lucene internals or asking for help on
java-user@lucene.apache.org to figure out
why a document with five of our query terms scores lower than a different document with
only one of the query terms.
While this document won't answer your specific scoring issues, it will, hopefully, point you
to the places that can help you figure out the what and why of Lucene scoring.
Lucene scoring supports a number of pluggable information retrieval
models, including:
These models can be plugged in via the Similarity API, and offer extension hooks and parameters for tuning. In general, Lucene first finds the documents that need to be scored based on boolean logic in the Query specification, and then ranks this subset of matching documents via the retrieval model. For some valuable references on VSM and IR in general refer to Lucene Wiki IR references.
The rest of this document will cover Scoring basics and explain how to change your Similarity. Next, it will cover ways you can customize the lucene internals in Custom Queries -- Expert Level, which gives details on implementing your own Query class and related functionality. Finally, we will finish up with some reference material in the Appendix.
Scoring — Basics
Scoring is very much dependent on the way documents are indexed, so it is important to understand
indexing. (see Lucene overview before continuing on with this section) Be sure to use the useful IndexSearcher.explain(Query, doc) to understand how the score for a certain matching document was computed.
Generally, the Query determines which documents match (a binary
decision), while the Similarity determines how to assign scores to
the matching documents.
Fields and Documents
In Lucene, the objects we are scoring are Documents. A Document is a collection of Fields. Each Field has semantics about how it is created and stored (tokenized, stored, etc). It is important to note that Lucene scoring works on Fields and then combines the results to return Documents. This is important because two Documents with the exact same content, but one having the content in two Fields and the other in one Field may return different scores for the same query due to length normalization.
Score Boosting
Lucene allows influencing the score contribution of various parts of the query by wrapping with BoostQuery.
Changing Scoring — Similarity
Changing the scoring formula
Changing Similarity is an easy way to influence scoring, this is done at index-time with
IndexWriterConfig.setSimilarity(Similarity) and at query-time with
IndexSearcher.setSimilarity(Similarity). Be sure to use the same Similarity at query-time as at index-time (so that norms are encoded/decoded correctly); Lucene makes no effort to verify this.
You can influence scoring by configuring a different built-in Similarity implementation, or by tweaking its
parameters, subclassing it to override behavior. Some implementations also offer a modular API which you can
extend by plugging in a different component (e.g. term frequency normalizer).
Finally, you can extend the low level Similarity directly to implement a new retrieval model.
See the similarities package documentation for information on the built-in available scoring models and extending or changing Similarity.
Integrating field values into the score
While similarities help score a document relatively to a query, it is also common for documents to hold features that measure the quality of a match. Such features are best integrated into the score by indexing a FeatureField with the document at index-time, and then combining the similarity score and the feature score using a linear combination. For instance the below query matches the same documents as originalQuery and computes scores as similarityScore + 0.7 * featureScore:
Query originalQuery = new BooleanQuery.Builder()
.add(new TermQuery(new Term("body", "apache")), Occur.SHOULD)
.add(new TermQuery(new Term("body", "lucene")), Occur.SHOULD)
.build();
Query featureQuery = FeatureField.newSaturationQuery("features", "pagerank");
Query query = new BooleanQuery.Builder()
.add(originalQuery, Occur.MUST)
.add(new BoostQuery(featureQuery, 0.7f), Occur.SHOULD)
.build();
A less efficient yet more flexible way of modifying scores is to index scoring features into
doc-value fields and then combine them with the similarity score using a
FunctionScoreQuery
from the queries module. For instance the below example shows how to compute scores as similarityScore * Math.log(popularity) using the expressions module and assuming that values for the popularity field have been set in a NumericDocValuesField at index time:
// compile an expression:
Expression expr = JavascriptCompiler.compile("_score * ln(popularity)");
// SimpleBindings just maps variables to SortField instances
SimpleBindings bindings = new SimpleBindings();
bindings.add(new SortField("_score", SortField.Type.SCORE));
bindings.add(new SortField("popularity", SortField.Type.INT));
// create a query that matches based on 'originalQuery' but
// scores using expr
Query query = new FunctionScoreQuery(
originalQuery,
expr.getDoubleValuesSource(bindings));
Custom Queries — Expert Level
Custom queries are an expert level task, so tread carefully and be prepared to share your code if
you want help.
With the warning out of the way, it is possible to change a lot more than just the Similarity
when it comes to matching and scoring in Lucene. Lucene's search is a complex mechanism that is grounded by
three main classes:
-
Query — The abstract object representation of the user's information need.
-
Weight — A specialization of a Query for a given index. This typically associates a Query object with index statistics that are later used to compute document scores. -
Scorer — The core class of the scoring process: for a given segment, scorers return iterators over matches and give a way to compute the score of these matches.
-
BulkScorer — An abstract class that scores a range of documents. A default implementation simply iterates through the hits from Scorer, but some queries such as BooleanQuery have more efficient implementations.
Details on each of these classes, and their children, can be found in the subsections below.
The Query Class
In some sense, the Query class is where it all begins. Without a Query, there would be nothing to score. Furthermore, the Query class is the catalyst for the other scoring classes as it is often responsible for creating them or coordinating the functionality between them. The Query class has several methods that are important for derived classes:
createWeight(IndexSearcher searcher, ScoreMode scoreMode, float boost) — A Weight is the internal representation of the Query, so each Query implementation must provide an implementation of Weight. See the subsection on The Weight Interface below for details on implementing the Weight
interface.rewrite(IndexReader reader) — Rewrites queries into primitive queries. Primitive queries are: TermQuery, BooleanQuery, and other queries that implement createWeight(IndexSearcher searcher,ScoreMode scoreMode, float boost)
The Weight Interface
The Weight interface provides an internal representation of the Query so that it can be reused. Any IndexSearcher dependent state should be stored in the Weight implementation, not in the Query class. The interface defines four main methods:
-
scorer() — Construct a new Scorer for this Weight. See The Scorer Class
below for help defining a Scorer. As the name implies, the Scorer is responsible for doing the actual scoring of documents
given the Query.
-
explain(LeafReaderContext context, int doc) — Provide a means for explaining why a given document was scored the way it was. Typically a weight such as TermWeight that scores via a Similarity will make use of the Similarity's implementation: SimScorer#explain(Explanation freq, long norm).
-
extractTerms(Set<Term> terms) — Extract terms that this query operates on. This is typically used to support distributed search: knowing the terms that a query operates on helps merge index statistics of these terms so that scores are computed over a subset of the data like they would if all documents were in the same index.
-
matches(LeafReaderContext context, int doc) — Give information about positions and offsets of matches. This is typically useful to implement highlighting.
The Scorer Class
The Scorer abstract class provides common scoring functionality for all Scorer implementations and is the heart of the Lucene scoring process. The Scorer defines the following methods which must be implemented:
-
iterator() — Return a DocIdSetIterator that can iterate over all document that matches this Query.
-
docID() — Returns the id of the Document that contains the match.
-
score() — Return the score of the current document. This value can be determined in any appropriate way for an application. For instance, the TermScorer simply defers to the configured Similarity: SimScorer.score(float freq, long norm).
-
getChildren() — Returns any child subscorers underneath this scorer. This allows for users to navigate the scorer hierarchy and receive more fine-grained details on the scoring process.
The BulkScorer Class
The BulkScorer scores a range of documents. There is only one abstract method:
-
score(LeafCollector,Bits,int,int) — Score all documents up to but not including the specified max document.
Why would I want to add my own Query?
In a nutshell, you want to add your own custom Query implementation when you think that Lucene's
aren't appropriate for the
task that you want to do. You might be doing some cutting edge research or you need more information
back
out of Lucene (similar to Doug adding SpanQuery functionality).
Appendix: Search Algorithm
This section is mostly notes on stepping through the Scoring process and serves as
fertilizer for the earlier sections.
In the typical search application, a Query is passed to the IndexSearcher, beginning the scoring process.
Once inside the IndexSearcher, a Collector is used for the scoring and sorting of the search results. These important objects are involved in a search:
- The
Weight object of the Query. The Weight object is an internal representation of the Query that allows the Query to be reused by the IndexSearcher.
- The IndexSearcher that initiated the call.
- A
Sort object for specifying how to sort the results if the standard score-based sort method is not desired.
Assuming we are not sorting (since sorting doesn't affect the raw Lucene score), we call one of the search methods of the IndexSearcher, passing in the Weight object created by
IndexSearcher.createWeight(Query,ScoreMode,float) and the number of results we want. This method returns a TopDocs object, which is an internal collection of search results. The IndexSearcher creates a TopScoreDocCollector and passes it along with the Weight to another expert search method (for more on the Collector mechanism, see IndexSearcher). The TopScoreDocCollector uses a PriorityQueue to collect the top results for the search.
At last, we are actually going to score some documents. The score method takes in the Collector (most likely the TopScoreDocCollector or TopFieldCollector) and does its business. Of course, here is where things get involved. The Scorer that is returned by the Weight object depends on what type of Query was submitted. In most real world applications with multiple query terms, the Scorer is going to be a BooleanScorer2 created from BooleanWeight (see the section on custom queries for info on changing this).
Assuming a BooleanScorer2, we get a internal Scorer based on the required, optional and prohibited parts of the query. Using this internal Scorer, the BooleanScorer2 then proceeds into a while loop based on the DocIdSetIterator.nextDoc() method. The nextDoc() method advances to the next document matching the query. This is an abstract method in the Scorer class and is thus overridden by all derived implementations. If you have a simple OR query your internal Scorer is most likely a DisjunctionSumScorer, which essentially combines the scorers from the sub scorers of the OR'd terms.
/**
* Code to search indices.
*
* <h2>Table Of Contents</h2>
* <ol>
* <li><a href="#search">Search Basics</a></li>
* <li><a href="#query">The Query Classes</a></li>
* <li><a href="#scoring">Scoring: Introduction</a></li>
* <li><a href="#scoringBasics">Scoring: Basics</a></li>
* <li><a href="#changingScoring">Changing the Scoring</a></li>
* <li><a href="#algorithm">Appendix: Search Algorithm</a></li>
* </ol>
*
*
* <a name="search"></a>
* <h2>Search Basics</h2>
* <p>
* Lucene offers a wide variety of {@link org.apache.lucene.search.Query} implementations, most of which are in
* this package, its subpackage ({@link org.apache.lucene.search.spans spans},
* or the <a href="{@docRoot}/../queries/overview-summary.html">queries module</a>. These implementations can be combined in a wide
* variety of ways to provide complex querying capabilities along with information about where matches took place in the document
* collection. The <a href="#query">Query Classes</a> section below highlights some of the more important Query classes. For details
* on implementing your own Query class, see <a href="#customQueriesExpert">Custom Queries -- Expert Level</a> below.
* <p>
* To perform a search, applications usually call {@link
* org.apache.lucene.search.IndexSearcher#search(Query,int)}.
* <p>
* Once a Query has been created and submitted to the {@link org.apache.lucene.search.IndexSearcher IndexSearcher}, the scoring
* process begins. After some infrastructure setup, control finally passes to the {@link org.apache.lucene.search.Weight Weight}
* implementation and its {@link org.apache.lucene.search.Scorer Scorer} or {@link org.apache.lucene.search.BulkScorer BulkScorer}
* instances. See the <a href="#algorithm">Algorithm</a> section for more notes on the process.
* <!-- FILL IN MORE HERE -->
* <!-- TODO: this page over-links the same things too many times -->
*
*
* <a name="query"></a>
* <h2>Query Classes</h2>
* <h3>
* {@link org.apache.lucene.search.TermQuery TermQuery}
* </h3>
*
* <p>Of the various implementations of
* {@link org.apache.lucene.search.Query Query}, the
* {@link org.apache.lucene.search.TermQuery TermQuery}
* is the easiest to understand and the most often used in applications. A
* {@link org.apache.lucene.search.TermQuery TermQuery} matches all the documents that contain the
* specified
* {@link org.apache.lucene.index.Term Term},
* which is a word that occurs in a certain
* {@link org.apache.lucene.document.Field Field}.
* Thus, a {@link org.apache.lucene.search.TermQuery TermQuery} identifies and scores all
* {@link org.apache.lucene.document.Document Document}s that have a
* {@link org.apache.lucene.document.Field Field} with the specified string in it.
* Constructing a {@link org.apache.lucene.search.TermQuery TermQuery}
* is as simple as:
* <pre class="prettyprint">
* TermQuery tq = new TermQuery(new Term("fieldName", "term"));
* </pre>In this example, the {@link org.apache.lucene.search.Query Query} identifies all
* {@link org.apache.lucene.document.Document Document}s that have the
* {@link org.apache.lucene.document.Field Field} named <tt>"fieldName"</tt>
* containing the word <tt>"term"</tt>.
* <h3>
* {@link org.apache.lucene.search.BooleanQuery BooleanQuery}
* </h3>
*
* <p>Things start to get interesting when one combines multiple
* {@link org.apache.lucene.search.TermQuery TermQuery} instances into a
* {@link org.apache.lucene.search.BooleanQuery BooleanQuery}.
* A {@link org.apache.lucene.search.BooleanQuery BooleanQuery} contains multiple
* {@link org.apache.lucene.search.BooleanClause BooleanClause}s,
* where each clause contains a sub-query ({@link org.apache.lucene.search.Query Query}
* instance) and an operator (from
* {@link org.apache.lucene.search.BooleanClause.Occur BooleanClause.Occur})
* describing how that sub-query is combined with the other clauses:
* <ol>
*
* <li><p>{@link org.apache.lucene.search.BooleanClause.Occur#SHOULD SHOULD} — Use this operator when a clause can occur in the result set, but is not required.
* If a query is made up of all SHOULD clauses, then every document in the result
* set matches at least one of these clauses.</p></li>
*
* <li><p>{@link org.apache.lucene.search.BooleanClause.Occur#MUST MUST} — Use this operator when a clause is required to occur in the result set and should
* contribute to the score. Every document in the result set will match all such clauses.</p></li>
*
* <li><p>{@link org.apache.lucene.search.BooleanClause.Occur#FILTER FILTER} — Use this operator when a clause is required to occur in the result set but
* should not contribute to the score. Every document in the result set will match all such clauses.</p></li>
*
* <li><p>{@link org.apache.lucene.search.BooleanClause.Occur#MUST_NOT MUST NOT} — Use this operator when a
* clause must not occur in the result set. No
* document in the result set will match
* any such clauses.</p></li>
* </ol>
* Boolean queries are constructed by adding two or more
* {@link org.apache.lucene.search.BooleanClause BooleanClause}
* instances. If too many clauses are added, a {@link org.apache.lucene.search.BooleanQuery.TooManyClauses TooManyClauses}
* exception will be thrown during searching. This most often occurs
* when a {@link org.apache.lucene.search.Query Query}
* is rewritten into a {@link org.apache.lucene.search.BooleanQuery BooleanQuery} with many
* {@link org.apache.lucene.search.TermQuery TermQuery} clauses,
* for example by {@link org.apache.lucene.search.WildcardQuery WildcardQuery}.
* The default setting for the maximum number
* of clauses is 1024, but this can be changed via the
* static method {@link org.apache.lucene.search.BooleanQuery#setMaxClauseCount(int)}.
*
* <h3>Phrases</h3>
*
* <p>Another common search is to find documents containing certain phrases. This
* is handled three different ways:
* <ol>
* <li>
* <p>{@link org.apache.lucene.search.PhraseQuery PhraseQuery}
* — Matches a sequence of
* {@link org.apache.lucene.index.Term Term}s.
* {@link org.apache.lucene.search.PhraseQuery PhraseQuery} uses a slop factor to determine
* how many positions may occur between any two terms in the phrase and still be considered a match.
* The slop is 0 by default, meaning the phrase must match exactly.</p>
* </li>
* <li>
* <p>{@link org.apache.lucene.search.MultiPhraseQuery MultiPhraseQuery}
* — A more general form of PhraseQuery that accepts multiple Terms
* for a position in the phrase. For example, this can be used to perform phrase queries that also
* incorporate synonyms.
* </li>
* <li>
* <p>{@link org.apache.lucene.search.spans.SpanNearQuery SpanNearQuery}
* — Matches a sequence of other
* {@link org.apache.lucene.search.spans.SpanQuery SpanQuery}
* instances. {@link org.apache.lucene.search.spans.SpanNearQuery SpanNearQuery} allows for
* much more
* complicated phrase queries since it is constructed from other
* {@link org.apache.lucene.search.spans.SpanQuery SpanQuery}
* instances, instead of only {@link org.apache.lucene.search.TermQuery TermQuery}
* instances.</p>
* </li>
* </ol>
*
* <h3>
* {@link org.apache.lucene.search.PointRangeQuery PointRangeQuery}
* </h3>
*
* <p>The
* {@link org.apache.lucene.search.PointRangeQuery PointRangeQuery}
* matches all documents that occur in a numeric range.
* For PointRangeQuery to work, you must index the values
* using a one of the numeric fields ({@link org.apache.lucene.document.IntPoint IntPoint},
* {@link org.apache.lucene.document.LongPoint LongPoint}, {@link org.apache.lucene.document.FloatPoint FloatPoint},
* or {@link org.apache.lucene.document.DoublePoint DoublePoint}).
*
* <h3>
* {@link org.apache.lucene.search.PrefixQuery PrefixQuery},
* {@link org.apache.lucene.search.WildcardQuery WildcardQuery},
* {@link org.apache.lucene.search.RegexpQuery RegexpQuery}
* </h3>
*
* <p>While the
* {@link org.apache.lucene.search.PrefixQuery PrefixQuery}
* has a different implementation, it is essentially a special case of the
* {@link org.apache.lucene.search.WildcardQuery WildcardQuery}.
* The {@link org.apache.lucene.search.PrefixQuery PrefixQuery} allows an application
* to identify all documents with terms that begin with a certain string. The
* {@link org.apache.lucene.search.WildcardQuery WildcardQuery} generalizes this by allowing
* for the use of <tt>*</tt> (matches 0 or more characters) and <tt>?</tt> (matches exactly one character) wildcards.
* Note that the {@link org.apache.lucene.search.WildcardQuery WildcardQuery} can be quite slow. Also
* note that
* {@link org.apache.lucene.search.WildcardQuery WildcardQuery} should
* not start with <tt>*</tt> and <tt>?</tt>, as these are extremely slow.
* Some QueryParsers may not allow this by default, but provide a <code>setAllowLeadingWildcard</code> method
* to remove that protection.
* The {@link org.apache.lucene.search.RegexpQuery RegexpQuery} is even more general than WildcardQuery,
* allowing an application to identify all documents with terms that match a regular expression pattern.
* <h3>
* {@link org.apache.lucene.search.FuzzyQuery FuzzyQuery}
* </h3>
*
* <p>A
* {@link org.apache.lucene.search.FuzzyQuery FuzzyQuery}
* matches documents that contain terms similar to the specified term. Similarity is
* determined using
* <a href="http://en.wikipedia.org/wiki/Levenshtein_distance">Levenshtein distance</a>.
* This type of query can be useful when accounting for spelling variations in the collection.
*
*
* <a name="scoring"></a>
* <h2>Scoring — Introduction</h2>
* <p>Lucene scoring is the heart of why we all love Lucene. It is blazingly fast and it hides
* almost all of the complexity from the user. In a nutshell, it works. At least, that is,
* until it doesn't work, or doesn't work as one would expect it to work. Then we are left
* digging into Lucene internals or asking for help on
* <a href="mailto:java-user@lucene.apache.org">java-user@lucene.apache.org</a> to figure out
* why a document with five of our query terms scores lower than a different document with
* only one of the query terms.
* <p>While this document won't answer your specific scoring issues, it will, hopefully, point you
* to the places that can help you figure out the <i>what</i> and <i>why</i> of Lucene scoring.
* <p>Lucene scoring supports a number of pluggable information retrieval
* <a href="http://en.wikipedia.org/wiki/Information_retrieval#Model_types">models</a>, including:
* <ul>
* <li><a href="http://en.wikipedia.org/wiki/Vector_Space_Model">Vector Space Model (VSM)</a></li>
* <li><a href="http://en.wikipedia.org/wiki/Probabilistic_relevance_model">Probabilistic Models</a> such as
* <a href="http://en.wikipedia.org/wiki/Probabilistic_relevance_model_(BM25)">Okapi BM25</a> and
* <a href="http://en.wikipedia.org/wiki/Divergence-from-randomness_model">DFR</a></li>
* <li><a href="http://en.wikipedia.org/wiki/Language_model">Language models</a></li>
* </ul>
* These models can be plugged in via the {@link org.apache.lucene.search.similarities Similarity API},
* and offer extension hooks and parameters for tuning. In general, Lucene first finds the documents
* that need to be scored based on boolean logic in the Query specification, and then ranks this subset of
* matching documents via the retrieval model. For some valuable references on VSM and IR in general refer to
* <a href="http://wiki.apache.org/lucene-java/InformationRetrieval">Lucene Wiki IR references</a>.
* <p>The rest of this document will cover <a href="#scoringBasics">Scoring basics</a> and explain how to
* change your {@link org.apache.lucene.search.similarities.Similarity Similarity}. Next, it will cover
* ways you can customize the lucene internals in
* <a href="#customQueriesExpert">Custom Queries -- Expert Level</a>, which gives details on
* implementing your own {@link org.apache.lucene.search.Query Query} class and related functionality.
* Finally, we will finish up with some reference material in the <a href="#algorithm">Appendix</a>.
*
*
* <a name="scoringBasics"></a>
* <h2>Scoring — Basics</h2>
* <p>Scoring is very much dependent on the way documents are indexed, so it is important to understand
* indexing. (see <a href="{@docRoot}/overview-summary.html#overview_description">Lucene overview</a>
* before continuing on with this section) Be sure to use the useful
* {@link org.apache.lucene.search.IndexSearcher#explain(org.apache.lucene.search.Query, int) IndexSearcher.explain(Query, doc)}
* to understand how the score for a certain matching document was
* computed.
*
* <p>Generally, the Query determines which documents match (a binary
* decision), while the Similarity determines how to assign scores to
* the matching documents.
*
* </p>
* <h3>Fields and Documents</h3>
* <p>In Lucene, the objects we are scoring are {@link org.apache.lucene.document.Document Document}s.
* A Document is a collection of {@link org.apache.lucene.document.Field Field}s. Each Field has
* {@link org.apache.lucene.document.FieldType semantics} about how it is created and stored
* ({@link org.apache.lucene.document.FieldType#tokenized() tokenized},
* {@link org.apache.lucene.document.FieldType#stored() stored}, etc). It is important to note that
* Lucene scoring works on Fields and then combines the results to return Documents. This is
* important because two Documents with the exact same content, but one having the content in two
* Fields and the other in one Field may return different scores for the same query due to length
* normalization.
* <h3>Score Boosting</h3>
* <p>Lucene allows influencing the score contribution of various parts of the query by wrapping with
* {@link org.apache.lucene.search.BoostQuery}.</p>
*
* <a name="changingScoring"></a>
* <h2>Changing Scoring — Similarity</h2>
* <h3>Changing the scoring formula</h3>
* <p>
* Changing {@link org.apache.lucene.search.similarities.Similarity Similarity} is an easy way to
* influence scoring, this is done at index-time with
* {@link org.apache.lucene.index.IndexWriterConfig#setSimilarity(org.apache.lucene.search.similarities.Similarity)
* IndexWriterConfig.setSimilarity(Similarity)} and at query-time with
* {@link org.apache.lucene.search.IndexSearcher#setSimilarity(org.apache.lucene.search.similarities.Similarity)
* IndexSearcher.setSimilarity(Similarity)}. Be sure to use the same
* Similarity at query-time as at index-time (so that norms are
* encoded/decoded correctly); Lucene makes no effort to verify this.
* <p>
* You can influence scoring by configuring a different built-in Similarity implementation, or by tweaking its
* parameters, subclassing it to override behavior. Some implementations also offer a modular API which you can
* extend by plugging in a different component (e.g. term frequency normalizer).
* <p>
* Finally, you can extend the low level {@link org.apache.lucene.search.similarities.Similarity Similarity} directly
* to implement a new retrieval model.
* <p>
* See the {@link org.apache.lucene.search.similarities} package documentation for information
* on the built-in available scoring models and extending or changing Similarity.
*
* <h3>Integrating field values into the score</h3>
* <p>While similarities help score a document relatively to a query, it is also common for documents to hold
* features that measure the quality of a match. Such features are best integrated into the score by indexing
* a {@link org.apache.lucene.document.FeatureField FeatureField} with the document at index-time, and then
* combining the similarity score and the feature score using a linear combination. For instance the below
* query matches the same documents as {@code originalQuery} and computes scores as
* {@code similarityScore + 0.7 * featureScore}:
* <pre class="prettyprint">
* Query originalQuery = new BooleanQuery.Builder()
* .add(new TermQuery(new Term("body", "apache")), Occur.SHOULD)
* .add(new TermQuery(new Term("body", "lucene")), Occur.SHOULD)
* .build();
* Query featureQuery = FeatureField.newSaturationQuery("features", "pagerank");
* Query query = new BooleanQuery.Builder()
* .add(originalQuery, Occur.MUST)
* .add(new BoostQuery(featureQuery, 0.7f), Occur.SHOULD)
* .build();
* </pre>
*
* <p>A less efficient yet more flexible way of modifying scores is to index scoring features into
* doc-value fields and then combine them with the similarity score using a
* <a href="{@docRoot}/../queries/org/apache/lucene/queries/function/FunctionScoreQuery.html">FunctionScoreQuery</a>
* from the <a href="{@docRoot}/../queries/overview-summary.html">queries module</a>. For instance
* the below example shows how to compute scores as {@code similarityScore * Math.log(popularity)}
* using the <a href="{@docRoot}/../expressions/overview-summary.html">expressions module</a> and
* assuming that values for the {@code popularity} field have been set in a
* {@link org.apache.lucene.document.NumericDocValuesField NumericDocValuesField} at index time:
* <pre class="prettyprint">
* // compile an expression:
* Expression expr = JavascriptCompiler.compile("_score * ln(popularity)");
*
* // SimpleBindings just maps variables to SortField instances
* SimpleBindings bindings = new SimpleBindings();
* bindings.add(new SortField("_score", SortField.Type.SCORE));
* bindings.add(new SortField("popularity", SortField.Type.INT));
*
* // create a query that matches based on 'originalQuery' but
* // scores using expr
* Query query = new FunctionScoreQuery(
* originalQuery,
* expr.getDoubleValuesSource(bindings));
* </pre>
*
* <a name="customQueriesExpert"></a>
* <h2>Custom Queries — Expert Level</h2>
*
* <p>Custom queries are an expert level task, so tread carefully and be prepared to share your code if
* you want help.
*
* <p>With the warning out of the way, it is possible to change a lot more than just the Similarity
* when it comes to matching and scoring in Lucene. Lucene's search is a complex mechanism that is grounded by
* <span>three main classes</span>:
* <ol>
* <li>
* {@link org.apache.lucene.search.Query Query} — The abstract object representation of the
* user's information need.</li>
* <li>
* {@link org.apache.lucene.search.Weight Weight} — A specialization of a Query for a given
* index. This typically associates a Query object with index statistics that are later used to
* compute document scores.
* <li>
* {@link org.apache.lucene.search.Scorer Scorer} — The core class of the scoring process:
* for a given segment, scorers return {@link org.apache.lucene.search.Scorer#iterator iterators}
* over matches and give a way to compute the {@link org.apache.lucene.search.Scorer#score score}
* of these matches.</li>
* <li>
* {@link org.apache.lucene.search.BulkScorer BulkScorer} — An abstract class that scores
* a range of documents. A default implementation simply iterates through the hits from
* {@link org.apache.lucene.search.Scorer Scorer}, but some queries such as
* {@link org.apache.lucene.search.BooleanQuery BooleanQuery} have more efficient
* implementations.</li>
* </ol>
* Details on each of these classes, and their children, can be found in the subsections below.
* <h3>The Query Class</h3>
* <p>In some sense, the
* {@link org.apache.lucene.search.Query Query}
* class is where it all begins. Without a Query, there would be
* nothing to score. Furthermore, the Query class is the catalyst for the other scoring classes as it
* is often responsible
* for creating them or coordinating the functionality between them. The
* {@link org.apache.lucene.search.Query Query} class has several methods that are important for
* derived classes:
* <ol>
* <li>{@link org.apache.lucene.search.Query#createWeight(IndexSearcher,ScoreMode,float) createWeight(IndexSearcher searcher, ScoreMode scoreMode, float boost)} — A
* {@link org.apache.lucene.search.Weight Weight} is the internal representation of the
* Query, so each Query implementation must
* provide an implementation of Weight. See the subsection on <a
* href="#weightClass">The Weight Interface</a> below for details on implementing the Weight
* interface.</li>
* <li>{@link org.apache.lucene.search.Query#rewrite(org.apache.lucene.index.IndexReader) rewrite(IndexReader reader)} — Rewrites queries into primitive queries. Primitive queries are:
* {@link org.apache.lucene.search.TermQuery TermQuery},
* {@link org.apache.lucene.search.BooleanQuery BooleanQuery}, <span
* >and other queries that implement {@link org.apache.lucene.search.Query#createWeight(IndexSearcher,ScoreMode,float) createWeight(IndexSearcher searcher,ScoreMode scoreMode, float boost)}</span></li>
* </ol>
* <a name="weightClass"></a>
* <h3>The Weight Interface</h3>
* <p>The
* {@link org.apache.lucene.search.Weight Weight}
* interface provides an internal representation of the Query so that it can be reused. Any
* {@link org.apache.lucene.search.IndexSearcher IndexSearcher}
* dependent state should be stored in the Weight implementation,
* not in the Query class. The interface defines four main methods:
* <ol>
* <li>
* {@link org.apache.lucene.search.Weight#scorer scorer()} —
* Construct a new {@link org.apache.lucene.search.Scorer Scorer} for this Weight. See <a href="#scorerClass">The Scorer Class</a>
* below for help defining a Scorer. As the name implies, the Scorer is responsible for doing the actual scoring of documents
* given the Query.
* </li>
* <li>
* {@link org.apache.lucene.search.Weight#explain(org.apache.lucene.index.LeafReaderContext, int)
* explain(LeafReaderContext context, int doc)} — Provide a means for explaining why a given document was
* scored the way it was.
* Typically a weight such as TermWeight
* that scores via a {@link org.apache.lucene.search.similarities.Similarity Similarity} will make use of the Similarity's implementation:
* {@link org.apache.lucene.search.similarities.Similarity.SimScorer#explain(Explanation, long) SimScorer#explain(Explanation freq, long norm)}.
* </li>
* <li>
* {@link org.apache.lucene.search.Weight#extractTerms(java.util.Set) extractTerms(Set<Term> terms)} — Extract terms that
* this query operates on. This is typically used to support distributed search: knowing the terms that a query operates on helps
* merge index statistics of these terms so that scores are computed over a subset of the data like they would if all documents
* were in the same index.
* </li>
* <li>
* {@link org.apache.lucene.search.Weight#matches matches(LeafReaderContext context, int doc)} — Give information about positions
* and offsets of matches. This is typically useful to implement highlighting.
* </li>
* </ol>
* <a name="scorerClass"></a>
* <h3>The Scorer Class</h3>
* <p>The
* {@link org.apache.lucene.search.Scorer Scorer}
* abstract class provides common scoring functionality for all Scorer implementations and
* is the heart of the Lucene scoring process. The Scorer defines the following methods which
* must be implemented:
* <ol>
* <li>
* {@link org.apache.lucene.search.Scorer#iterator iterator()} — Return a
* {@link org.apache.lucene.search.DocIdSetIterator DocIdSetIterator} that can iterate over all
* document that matches this Query.
* </li>
* <li>
* {@link org.apache.lucene.search.Scorer#docID docID()} — Returns the id of the
* {@link org.apache.lucene.document.Document Document} that contains the match.
* </li>
* <li>
* {@link org.apache.lucene.search.Scorer#score score()} — Return the score of the
* current document. This value can be determined in any appropriate way for an application. For instance, the
* {@link org.apache.lucene.search.TermScorer TermScorer} simply defers to the configured Similarity:
* {@link org.apache.lucene.search.similarities.Similarity.SimScorer#score(float, long) SimScorer.score(float freq, long norm)}.
* </li>
* <li>
* {@link org.apache.lucene.search.Scorer#getChildren getChildren()} — Returns any child subscorers
* underneath this scorer. This allows for users to navigate the scorer hierarchy and receive more fine-grained
* details on the scoring process.
* </li>
* </ol>
* <a name="bulkScorerClass"></a>
* <h3>The BulkScorer Class</h3>
* <p>The
* {@link org.apache.lucene.search.BulkScorer BulkScorer} scores a range of documents. There is only one
* abstract method:
* <ol>
* <li>
* {@link org.apache.lucene.search.BulkScorer#score(org.apache.lucene.search.LeafCollector,org.apache.lucene.util.Bits,int,int) score(LeafCollector,Bits,int,int)} —
* Score all documents up to but not including the specified max document.
* </li>
* </ol>
* <h3>Why would I want to add my own Query?</h3>
*
* <p>In a nutshell, you want to add your own custom Query implementation when you think that Lucene's
* aren't appropriate for the
* task that you want to do. You might be doing some cutting edge research or you need more information
* back
* out of Lucene (similar to Doug adding SpanQuery functionality).
*
* <!-- TODO: integrate this better, it's better served as an intro than an appendix -->
*
*
* <a name="algorithm"></a>
* <h2>Appendix: Search Algorithm</h2>
* <p>This section is mostly notes on stepping through the Scoring process and serves as
* fertilizer for the earlier sections.
* <p>In the typical search application, a {@link org.apache.lucene.search.Query Query}
* is passed to the {@link org.apache.lucene.search.IndexSearcher IndexSearcher},
* beginning the scoring process.
* <p>Once inside the IndexSearcher, a {@link org.apache.lucene.search.Collector Collector}
* is used for the scoring and sorting of the search results.
* These important objects are involved in a search:
* <ol>
* <li>The {@link org.apache.lucene.search.Weight Weight} object of the Query. The
* Weight object is an internal representation of the Query that allows the Query
* to be reused by the IndexSearcher.</li>
* <li>The IndexSearcher that initiated the call.</li>
* <li>A {@link org.apache.lucene.search.Sort Sort} object for specifying how to sort
* the results if the standard score-based sort method is not desired.</li>
* </ol>
* <p>Assuming we are not sorting (since sorting doesn't affect the raw Lucene score),
* we call one of the search methods of the IndexSearcher, passing in the
* {@link org.apache.lucene.search.Weight Weight} object created by
* {@link org.apache.lucene.search.IndexSearcher#createWeight(org.apache.lucene.search.Query,ScoreMode,float)
* IndexSearcher.createWeight(Query,ScoreMode,float)} and the number of results we want.
* This method returns a {@link org.apache.lucene.search.TopDocs TopDocs} object,
* which is an internal collection of search results. The IndexSearcher creates
* a {@link org.apache.lucene.search.TopScoreDocCollector TopScoreDocCollector} and
* passes it along with the Weight to another expert search method (for
* more on the {@link org.apache.lucene.search.Collector Collector} mechanism,
* see {@link org.apache.lucene.search.IndexSearcher IndexSearcher}). The TopScoreDocCollector
* uses a {@link org.apache.lucene.util.PriorityQueue PriorityQueue} to collect the
* top results for the search.
* <p>At last, we are actually going to score some documents. The score method takes in the Collector
* (most likely the TopScoreDocCollector or TopFieldCollector) and does its business. Of course, here
* is where things get involved. The {@link org.apache.lucene.search.Scorer Scorer} that is returned
* by the {@link org.apache.lucene.search.Weight Weight} object depends on what type of Query was
* submitted. In most real world applications with multiple query terms, the
* {@link org.apache.lucene.search.Scorer Scorer} is going to be a <code>BooleanScorer2</code> created
* from {@link org.apache.lucene.search.BooleanWeight BooleanWeight} (see the section on
* <a href="#customQueriesExpert">custom queries</a> for info on changing this).
* <p>Assuming a BooleanScorer2, we get a internal Scorer based on the required, optional and prohibited parts of the query.
* Using this internal Scorer, the BooleanScorer2 then proceeds into a while loop based on the
* {@link org.apache.lucene.search.DocIdSetIterator#nextDoc DocIdSetIterator.nextDoc()} method. The nextDoc() method advances
* to the next document matching the query. This is an abstract method in the Scorer class and is thus
* overridden by all derived implementations. If you have a simple OR query your internal Scorer is most
* likely a DisjunctionSumScorer, which essentially combines the scorers from the sub scorers of the OR'd terms.
*/
package org.apache.lucene.search;