6 Apache Lucene - Scoring
10 <author email="gsingers at apache.org">Grant Ingersoll</author>
15 <section id="Introduction"><title>Introduction</title>
16 <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.
17 In a nutshell, it works. At least, that is, until it doesn't work, or doesn't work as one would expect it to
18 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
19 scores lower than a different document with only one of the query terms. </p>
20 <p>While this document won't answer your specific scoring issues, it will, hopefully, point you to the places that can
21 help you figure out the what and why of Lucene scoring.</p>
22 <p>Lucene scoring uses a combination of the
23 <a href="http://en.wikipedia.org/wiki/Vector_Space_Model">Vector Space Model (VSM) of Information
24 Retrieval</a> and the <a href="http://en.wikipedia.org/wiki/Standard_Boolean_model">Boolean model</a>
26 how relevant a given Document is to a User's query. In general, the idea behind the VSM is the more
27 times a query term appears in a document relative to
28 the number of times the term appears in all the documents in the collection, the more relevant that
29 document is to the query. It uses the Boolean model to first narrow down the documents that need to
30 be scored based on the use of boolean logic in the Query specification. Lucene also adds some
31 capabilities and refinements onto this model to support boolean and fuzzy searching, but it
32 essentially remains a VSM based system at the heart.
33 For some valuable references on VSM and IR in general refer to the
34 <a href="http://wiki.apache.org/lucene-java/InformationRetrieval">Lucene Wiki IR references</a>.
36 <p>The rest of this document will cover <a href="#Scoring">Scoring</a> basics and how to change your
37 <a href="api/core/org/apache/lucene/search/Similarity.html">Similarity</a>. Next it will cover ways you can
38 customize the Lucene internals in <a href="#Changing your Scoring -- Expert Level">Changing your Scoring
39 -- Expert Level</a> which gives details on implementing your own
40 <a href="api/core/org/apache/lucene/search/Query.html">Query</a> class and related functionality. Finally, we
41 will finish up with some reference material in the <a href="#Appendix">Appendix</a>.
44 <section id="Scoring"><title>Scoring</title>
45 <p>Scoring is very much dependent on the way documents are indexed,
46 so it is important to understand indexing (see
47 <a href="gettingstarted.html">Apache Lucene - Getting Started Guide</a>
49 <a href="fileformats.html">file formats</a>
50 before continuing on with this section.) It is also assumed that readers know how to use the
51 <a href="api/core/org/apache/lucene/search/Searcher.html#explain(Query query, int doc)">Searcher.explain(Query query, int doc)</a> functionality,
52 which can go a long way in informing why a score is returned.
54 <section id="Fields and Documents"><title>Fields and Documents</title>
55 <p>In Lucene, the objects we are scoring are
56 <a href="api/core/org/apache/lucene/document/Document.html">Documents</a>. A Document is a collection
58 <a href="api/core/org/apache/lucene/document/Field.html">Fields</a>. Each Field has semantics about how
59 it is created and stored (i.e. tokenized, untokenized, raw data, compressed, etc.) It is important to
60 note that Lucene scoring works on Fields and then combines the results to return Documents. This is
61 important because two Documents with the exact same content, but one having the content in two Fields
62 and the other in one Field will return different scores for the same query due to length normalization
64 <a href="api/core/org/apache/lucene/search/DefaultSimilarity.html">DefaultSimilarity</a>
68 <section id="Score Boosting"><title>Score Boosting</title>
69 <p>Lucene allows influencing search results by "boosting" in more than one level:
71 <li><b>Document level boosting</b>
72 - while indexing - by calling
73 <a href="api/core/org/apache/lucene/document/Document.html#setBoost(float)">document.setBoost()</a>
74 before a document is added to the index.
76 <li><b>Document's Field level boosting</b>
77 - while indexing - by calling
78 <a href="api/core/org/apache/lucene/document/Fieldable.html#setBoost(float)">field.setBoost()</a>
79 before adding a field to the document (and before adding the document to the index).
81 <li><b>Query level boosting</b>
82 - during search, by setting a boost on a query clause, calling
83 <a href="api/core/org/apache/lucene/search/Query.html#setBoost(float)">Query.setBoost()</a>.
87 <p>Indexing time boosts are preprocessed for storage efficiency and written to
88 the directory (when writing the document) in a single byte (!) as follows:
89 For each field of a document, all boosts of that field
90 (i.e. all boosts under the same field name in that doc) are multiplied.
91 The result is multiplied by the boost of the document,
92 and also multiplied by a "field length norm" value
93 that represents the length of that field in that doc
94 (so shorter fields are automatically boosted up).
95 The result is decoded as a single byte
96 (with some precision loss of course) and stored in the directory.
97 The similarity object in effect at indexing computes the length-norm of the field.
99 <p>This composition of 1-byte representation of norms
100 (that is, indexing time multiplication of field boosts & doc boost & field-length-norm)
101 is nicely described in
102 <a href="api/core/org/apache/lucene/document/Fieldable.html#setBoost(float)">Fieldable.setBoost()</a>.
104 <p>Encoding and decoding of the resulted float norm in a single byte are done by the
105 static methods of the class Similarity:
106 <a href="api/core/org/apache/lucene/search/Similarity.html#encodeNorm(float)">encodeNorm()</a> and
107 <a href="api/core/org/apache/lucene/search/Similarity.html#decodeNorm(byte)">decodeNorm()</a>.
108 Due to loss of precision, it is not guaranteed that decode(encode(x)) = x,
109 e.g. decode(encode(0.89)) = 0.75.
110 At scoring (search) time, this norm is brought into the score of document
111 as <b>norm(t, d)</b>, as shown by the formula in
112 <a href="api/core/org/apache/lucene/search/Similarity.html">Similarity</a>.
115 <section id="Understanding the Scoring Formula"><title>Understanding the Scoring Formula</title>
118 This scoring formula is described in the
119 <a href="api/core/org/apache/lucene/search/Similarity.html">Similarity</a> class. Please take the time to study this formula, as it contains much of the information about how the
120 basics of Lucene scoring work, especially the
121 <a href="api/core/org/apache/lucene/search/TermQuery.html">TermQuery</a>.
124 <section id="The Big Picture"><title>The Big Picture</title>
125 <p>OK, so the tf-idf formula and the
126 <a href="api/core/org/apache/lucene/search/Similarity.html">Similarity</a>
127 is great for understanding the basics of Lucene scoring, but what really drives Lucene scoring are
128 the use and interactions between the
129 <a href="api/core/org/apache/lucene/search/Query.html">Query</a> classes, as created by each application in
130 response to a user's information need.
132 <p>In this regard, Lucene offers a wide variety of <a href="api/core/org/apache/lucene/search/Query.html">Query</a> implementations, most of which are in the
133 <a href="api/core/org/apache/lucene/search/package-summary.html">org.apache.lucene.search</a> package.
134 These implementations can be combined in a wide variety of ways to provide complex querying
135 capabilities along with
136 information about where matches took place in the document collection. The <a href="#Query Classes">Query</a>
138 highlights some of the more important Query classes. For information on the other ones, see the
139 <a href="api/core/org/apache/lucene/search/package-summary.html">package summary</a>. For details on implementing
140 your own Query class, see <a href="#Changing your Scoring -- Expert Level">Changing your Scoring --
141 Expert Level</a> below.
143 <p>Once a Query has been created and submitted to the
144 <a href="api/core/org/apache/lucene/search/IndexSearcher.html">IndexSearcher</a>, the scoring process
146 href="#Appendix">Appendix</a> Algorithm section for more notes on the process.) After some infrastructure setup,
147 control finally passes to the <a href="api/core/org/apache/lucene/search/Weight.html">Weight</a> implementation and its
148 <a href="api/core/org/apache/lucene/search/Scorer.html">Scorer</a> instance. In the case of any type of
149 <a href="api/core/org/apache/lucene/search/BooleanQuery.html">BooleanQuery</a>, scoring is handled by the
150 <a href="http://svn.apache.org/viewvc/lucene/dev/trunk/lucene/src/java/org/apache/lucene/search/BooleanQuery.java?view=log">BooleanWeight2</a>
151 (link goes to ViewVC BooleanQuery java code which contains the BooleanWeight2 inner class) or
152 <a href="http://svn.apache.org/viewvc/lucene/dev/trunk/lucene/src/java/org/apache/lucene/search/BooleanQuery.java?view=log">BooleanWeight</a>
153 (link goes to ViewVC BooleanQuery java code, which contains the BooleanWeight inner class).
156 Assuming the use of the BooleanWeight2, a
157 BooleanScorer2 is created by bringing together
159 <a href="api/core/org/apache/lucene/search/Scorer.html">Scorer</a>s from the sub-clauses of the BooleanQuery.
160 When the BooleanScorer2 is asked to score it delegates its work to an internal Scorer based on the type
161 of clauses in the Query. This internal Scorer essentially loops over the sub scorers and sums the scores
162 provided by each scorer while factoring in the coord() score.
163 <!-- Do we want to fill in the details of the counting sum scorer, disjunction scorer, etc.? -->
166 <section id="Query Classes"><title>Query Classes</title>
167 <p>For information on the Query Classes, refer to the
168 <a href="api/core/org/apache/lucene/search/package-summary.html#query">search package javadocs</a>
171 <section id="Changing Similarity"><title>Changing Similarity</title>
172 <p>One of the ways of changing the scoring characteristics of Lucene is to change the similarity factors. For information on
173 how to do this, see the
174 <a href="api/core/org/apache/lucene/search/package-summary.html#changingSimilarity">search package javadocs</a></p>
178 <section id="Changing your Scoring -- Expert Level"><title>Changing your Scoring -- Expert Level</title>
179 <p>At a much deeper level, one can affect scoring by implementing their own Query classes (and related scoring classes.) To learn more
180 about how to do this, refer to the
181 <a href="api/core/org/apache/lucene/search/package-summary.html#scoring">search package javadocs</a>
185 <section id="Appendix"><title>Appendix</title>
186 <section id="Algorithm"><title>Algorithm</title>
187 <p>This section is mostly notes on stepping through the Scoring process and serves as
188 fertilizer for the earlier sections.</p>
189 <p>In the typical search application, a
190 <a href="api/core/org/apache/lucene/search/Query.html">Query</a>
193 href="api/core/org/apache/lucene/search/Searcher.html">Searcher</a>
194 , beginning the scoring process.
196 <p>Once inside the Searcher, a
197 <a href="api/core/org/apache/lucene/search/Collector.html">Collector</a>
198 is used for the scoring and sorting of the search results.
199 These important objects are involved in a search:
202 <a href="api/core/org/apache/lucene/search/Weight.html">Weight</a>
203 object of the Query. The Weight object is an internal representation of the Query that
204 allows the Query to be reused by the Searcher.
206 <li>The Searcher that initiated the call.</li>
208 <a href="api/core/org/apache/lucene/search/Filter.html">Filter</a>
209 for limiting the result set. Note, the Filter may be null.
212 <a href="api/core/org/apache/lucene/search/Sort.html">Sort</a>
213 object for specifying how to sort the results if the standard score based sort method is not
218 <p> Assuming we are not sorting (since sorting doesn't
219 effect the raw Lucene score),
220 we call one of the search methods of the Searcher, passing in the
221 <a href="api/core/org/apache/lucene/search/Weight.html">Weight</a>
222 object created by Searcher.createWeight(Query),
223 <a href="api/core/org/apache/lucene/search/Filter.html">Filter</a>
224 and the number of results we want. This method
226 <a href="api/core/org/apache/lucene/search/TopDocs.html">TopDocs</a>
227 object, which is an internal collection of search results.
228 The Searcher creates a
229 <a href="api/core/org/apache/lucene/search/TopScoreDocCollector.html">TopScoreDocCollector</a>
230 and passes it along with the Weight, Filter to another expert search method (for more on the
231 <a href="api/core/org/apache/lucene/search/Collector.html">Collector</a>
233 <a href="api/core/org/apache/lucene/search/Searcher.html">Searcher</a>
234 .) The TopDocCollector uses a
235 <a href="api/core/org/apache/lucene/util/PriorityQueue.html">PriorityQueue</a>
236 to collect the top results for the search.
238 <p>If a Filter is being used, some initial setup is done to determine which docs to include. Otherwise,
239 we ask the Weight for
241 <a href="api/core/org/apache/lucene/search/Scorer.html">Scorer</a>
243 <a href="api/core/org/apache/lucene/index/IndexReader.html">IndexReader</a>
244 of the current searcher and we proceed by
245 calling the score method on the
246 <a href="api/core/org/apache/lucene/search/Scorer.html">Scorer</a>
249 <p>At last, we are actually going to score some documents. The score method takes in the Collector
250 (most likely the TopScoreDocCollector or TopFieldCollector) and does its business.
251 Of course, here is where things get involved. The
252 <a href="api/core/org/apache/lucene/search/Scorer.html">Scorer</a>
253 that is returned by the
254 <a href="api/core/org/apache/lucene/search/Weight.html">Weight</a>
255 object depends on what type of Query was submitted. In most real world applications with multiple
258 <a href="api/core/org/apache/lucene/search/Scorer.html">Scorer</a>
260 <a href="http://svn.apache.org/viewvc/lucene/dev/trunk/lucene/src/java/org/apache/lucene/search/BooleanScorer2.java?view=log">BooleanScorer2</a>
261 (see the section on customizing your scoring for info on changing this.)
264 <p>Assuming a BooleanScorer2 scorer, we first initialize the Coordinator, which is used to apply the
265 coord() factor. We then
266 get a internal Scorer based on the required, optional and prohibited parts of the query.
267 Using this internal Scorer, the BooleanScorer2 then proceeds
268 into a while loop based on the Scorer#next() method. The next() method advances to the next document
269 matching the query. This is an
270 abstract method in the Scorer class and is thus overriden by all derived
271 implementations. <!-- DOUBLE CHECK THIS -->If you have a simple OR query
272 your internal Scorer is most likely a DisjunctionSumScorer, which essentially combines the scorers
273 from the sub scorers of the OR'd terms.</p>