LUCENE-3261: Faceting module userguide

git-svn-id: https://svn.apache.org/repos/asf/lucene/dev/trunk@1175341 13f79535-47bb-0310-9956-ffa450edef68
This commit is contained in:
Shai Erera 2011-09-25 09:20:59 +00:00
parent d48960ae05
commit 151c26e7d9
5 changed files with 815 additions and 5 deletions

View File

@ -29,6 +29,14 @@
<property name="build.dir" location="build/" />
<property name="dist.dir" location="dist/" />
<property name="examples.dir" location="src/examples"/>
<path id="classpath">
<path refid="base.classpath" />
<pathelement location="${build.dir}/classes/java" />
<pathelement location="${build.dir}/classes/examples" />
</path>
<path id="examples.classpath">
<path refid="classpath" />
<pathelement location="${build.dir}/classes/java" />
@ -49,7 +57,7 @@
</path>
<target name="compile-examples" description="Compiles Facets examples">
<compile srcdir="src/examples" destdir="${build.dir}/classes/examples">
<compile srcdir="${examples.dir}" destdir="${build.dir}/classes/examples">
<classpath refid="examples.classpath" />
</compile>
</target>
@ -64,4 +72,22 @@
<target name="jar-core" depends="common.jar-core,jar-examples" />
<target name="javadocs" depends="compile-core">
<sequential>
<mkdir dir="${javadoc.dir}/contrib-${name}"/>
<copy todir="${javadoc.dir}/contrib-${name}" file="docs/userguide.html" />
<!-- javadoc core classes -->
<invoke-javadoc
destdir="${javadoc.dir}/contrib-${name}"
title="${Name} ${version} contrib-${name} API">
<sources>
<link href=""/>
<packageset dir="${src.dir}"/>
<packageset dir="${examples.dir}"/>
</sources>
</invoke-javadoc>
<jarify basedir="${javadoc.dir}/contrib-${name}" destfile="${build.dir}/${final.name}-javadoc.jar"/>
</sequential>
</target>
</project>

View File

@ -0,0 +1,784 @@
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<!--
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.
-->
<html>
<title>Facet Userguide</title>
<!-- load stylesheet + javascript in development mode -->
<link rel="stylesheet" type="text/css" href="../../../src/tools/prettify/prettify.css">
<script src="../../../src/tools/prettify/prettify.js" type="text/javascript"></script>
<!-- load stylesheet + javascript in distribution mode -->
<link rel="stylesheet" type="text/css" href="prettify.css">
<script src="prettify.js" type="text/javascript"></script>
<script language="javascript">
window.onload=function() {
prettyPrint();
}
</script>
<style>
body {
margin-left: 20%;
width: 60%;
counter-reset: section;
text-align: left;
}
h1.title {
text-align: center;
margin-top: 30px;
font-size: 5em;
line-height: 150%;
}
h1.section {
margin-top: 50px;
font-size: 2.5em;
counter-reset: subsection;
border: 1px solid black;
background-color: #D8D8D8;
padding-left: 5px;
}
h2.subsection {
font-size: 2em;
border: 1px solid black;
background-color: #D8D8D8;
padding-left: 5px;
}
/* auto-generated heading numbers */
h1.section:before {
counter-increment: section;
content: counter(section) ". ";
}
h2.subsection:before {
counter-increment: subsection;
content: counter(section) "." counter(subsection) " ";
}
/* override from prettify.css - add shadow, padding etc. */
pre.prettyprint {
margin-left: 2%;
width: 80%;
padding: 5px 3px 5px 3px;
/* shadow */
-moz-box-shadow: 5px 5px 2px #888;
-webkit-box-shadow: 5px 5px 2px #888;
box-shadow: 5px 5px 2px #888;
}
/* override from prettify.css - make keywords appear in bold */
span.kwd {
font-weight: bold;
}
ol.toc a {
text-decoration: none;
color: blue;
}
li.toc_first {
// margin-top: 10px;
font-size: 16px;
color: blue;
}
li.toc_second {
font-size: 14px;
margin-left: 15px;
color: blue;
}
/* reset style from prettify.css, so that line numbers appear in each line */
li.L0,li.L1,li.L2,li.L3,li.L5,li.L6,li.L7,li.L8 {
list-style-type:decimal
}
table.code_description td {
vertical-align: top;
}
</style>
<body>
<h1 class="title">
Apache Lucene<br>
Faceted Search<br>
User's Guide</h1>
<div class="toc">
<h1 class="toc">Table of Contents</h1>
<ol class="toc">
<li class="toc_first"><a href="#intro">Introduction</a></li>
<li class="toc_first"><a href="#facet_features">Facet Features</a></li>
<li class="toc_first"><a href="#facet_indexing">Indexing Categories Illustrated</a></li>
<li class="toc_first"><a href="#facet_accumulation">Accumulating Facets Illustrated</a></li>
<li class="toc_first"><a href="#indexed_facet_info">Indexed Facet Information</a></li>
<li class="toc_first"><a href="#taxonomy_index">Taxonomy Index</a></li>
<li class="toc_first"><a href="#facet_params">Facet Parameters</a></li>
<li class="toc_first"><a href="#advanced">Advanced Faceted Examples</a></li>
<li class="toc_first"><a href="#optimizations">Optimizations</a></li>
<li class="toc_first"><a href="#concurrent_indexing_search">Concurrent Indexing and Search</a></li>
</ol>
<h1 class="section"><a name="intro">Introduction</a></h1>
<p>
A category is an aspect of indexed documents which can be used to classify the
documents. For example, in a collection of books at an online bookstore, categories of
a book can be its price, author, publication date, binding type, and so on.
<p>
In faceted search, in addition to the standard set of search results, we also get facet
results, which are lists of subcategories for certain categories. For example, for the
price facet, we get a list of relevant price ranges; for the author facet, we get a list of
relevant authors; and so on. In most UIs, when users click one of these subcategories,
the search is narrowed, or drilled down, and a new search limited to this subcategory
(e.g., to a specific price range or author) is performed.
<p>
Note that faceted search is more than just the ordinary fielded search. In fielded
search, users can add search keywords like price:10 or author:"Mark
Twain" to the query to narrow the search, but this requires knowledge of which
fields are available, and which values are worth trying. This is where faceted search
comes in: it provides a list of useful subcategories, which ensures that the user only
drills down into useful subcategories and never into a category for which there are no
results. In essence, faceted search makes it easy to navigate through the search results.
The list of subcategories provided for each facet is also useful to the user in itself,
even when the user never drills down. This list allows the user to see at one glance
some statistics on the search results, e.g., what price ranges and which authors are
most relevant to the given query.
<p>
In recent years, faceted search has become a very common UI feature in search
engines, especially in e-commerce websites. Faceted search makes it easy for
untrained users to find the specific item they are interested in, whereas manually
adding search keywords (as in the examples above) proved too cumbersome for
ordinary users, and required too much guesswork, trial-and-error, or the reading of
lengthy help pages.
<p>
See <a href="http://en.wikipedia.org/wiki/Faceted_search">http://en.wikipedia.org/wiki/Faceted_search</a> for more information on faceted
search.
<h1 class="section"><a name="facet_features">Facet Features</a></h1>
First and main faceted search capability that comes to mind is counting, but in fact
faceted search is more than facet counting. We now briefly discuss the available
faceted search features.
<h2 class="subsection">Facet Counting</h2>
<p>
Which of the available subcategories of a facet should a UI display? A query in a
book store might yield books by a hundred different authors, but normally we'd want
do display only, say, ten of those.
<p>
Most available faceted search implementations use counts to determine the
importance of each subcategory. These implementations go over all search results for
the given query, and count how many results are in each subcategory. Finally, the
subcategories with the most results can be displayed. So the user sees the price ranges,
authors, and so on, for which there are most results. Often, the count is displayed next
to the subcategory name, in parentheses, telling the user how many results he can
expect to see if he drills down into this subcategory.
<p>
The main API for obtaining facet counting is <code>CountFacetRequest</code>, as in the
following code snippet:
<pre class="prettyprint lang-java">
new CountFacetRequest(new CategoryPath("author"), 10));
</pre>
A detailed code example using count facet requests is shown below - see
<a href="#facet_accumulation">Accumulating Facets</a>.
<h2 class="subsection"><a name="facet_association">Facet Associations</a></h2>
<p>
So far we've discussed categories as binary features, where a document either belongs
to a category, or not.
<p>
While counts are useful in most situations, they are sometimes not sufficiently
informative for the user, with respect to deciding which subcategory is more
important to display.
<p>
For this, the facets package allows to associate a value with a category. The search
time interpretation of the associated value is application dependent. For example, a
possible interpretation is as a <i>match level</i> (e.g., confidence level). This value can
then be used so that a document that is very weakly associated with a certain category
will only contribute little to this category's aggregated weight.
<h2 class="subsection"><a name="multiple_requests">Multiple Facet Requests</a></h2>
<p>
A single faceted accumulation is capable of servicing multiple facet requests.
Programmatic, this is quite simple - wrap all the facet requests of interest into the
facet-search-parameters which are passed to a facets accumulator/collector (more on
these objects below). The results would be comprised of as many facet results as there
were facet requests.
<p>
However there is a delicate <b>limitation</b>: all facets maintained in the same location in
the index are required to be treated the same. See the section on <a href="#indexing_params">Indexing Parameters</a>
for an explanation on maintaining certain facets at certain locations.
<h2 class="subsection"><a name="facet_labels">Facet Labels at Search Time</a></h2>
<p>
Facets results always contain the facet (internal) ID and (accumulated) value. Some of
the results also contain the facet label, AKA the category name. We mention this here
since computing the label is a time consuming task, and hence applications can
specify with a facet request to return top 1000 facets but to compute the label only for
the top 10 facets. In order to compute labels for more of the facet results it is not
required to perform accumulation again.
<p>
See <code>FacetRequest.getNumResults()</code>, <code>FacetRequest.getNumLabel()</code> and
<code>FacetResultNode.getLabel(TaxonomyReader)</code>.
<h1 class="section"><a name="facet_indexing">Indexing Categories Illustrated</a></h1>
<p>
In order to find facets at search time they must first be added to the index at indexing
time. Recall that Lucene documents are made of fields for textual search. The addition
of categories is performed by an appropriate <code>DocumentBuilder</code> - or
<code>CategoryDocumentBuilder</code> in our case.
<p>
Indexing therefore usually goes like this:
<ul>
<li>For each input document:
<ul>
<li>Create a fresh (empty) Lucene Document</li>
<li>Parse input text and add appropriate text search fields</li>
<li><b>Gather all input categories associated with the document and create
a CategoryDocumentBuilder with the list of categories</b></li>
<li><b>Build the document - this actually adds the categories to the
Lucene document.</b></li>
<li>Add the document to the index</li>
</ul></li>
</ul>
Following is a code snippet for indexing categories. The complete example can be
found in package <code>org.apache.lucene.facet.example.simple.SimpleIndexer</code>.
<pre class="prettyprint lang-java linenums">
IndexWriter writer = ...
TaxonomyWriter taxo = new LuceneTaxonomyWriter(taxoDir, OpenMode.CREATE);
...
Document doc = new Document();
doc.add(new Field("title", titleText, Store.YES, Index.ANALYZED));
...
List&lt;CategoryPath&gt; categories = new ArrayList&lt;CategoryPath&gt;();
categories.add(new CategoryPath("author", "Mark Twain"));
categories.add(new CategoryPath("year", "2010"));
...
DocumentBuilder categoryDocBuilder = new CategoryDocumentBuilder(taxo);
categoryDocBuilder.setCategoryPaths(categories);
categoryDocBuilder.build(doc);
writer.addDocument(doc);
</pre>
<p>
We now explain the steps above, following the code line numbers:
<table class="code_description">
<tr>
<td>(4)</td>
<td>Document contains not only text search fields but also facet search
information.</td>
</tr>
<tr>
<td>(7)</td>
<td>Prepare a container for document categories.</td>
</tr>
<tr>
<td>(8)</td>
<td>Categories that should be added to the document are accumulated in the
categories list.</td>
</tr>
<tr>
<td>(11)</td>
<td>A <code>CategoryDocumentBuilder</code> is created, set with the appropriate list
of categories, and invoked to build - that is, to populate the document
with categories. It is in this step that the taxonomy is updated to contain the
newly added categories (if not already there) - see more on this in the
section about the <a href="#taxonomy_index">Taxonomy Index</a> below. This line could be made more
compact: one can create a single <code>CategoryDocumentBuilder cBuilder</code> and reuse it like this:
<pre class="prettyprint lang-java linenums">
DocumentBuilder cBuilder = new CategoryDocumentBuilder(taxo);
cBuilder.setCategoryPaths(categories).build(doc);
</pre>
</td>
</tr>
<tr>
<td>(14)</td>
<td>Add the document to the index. As a result, category info is saved also in
the regular search index, for supporting facet aggregation at search time
(e.g. facet counting) as well as facet drill-down. For more information on
indexed facet information see below the section <a href="#indexed_facet_info">Indexed Facet Information</a>.</td>
</tr>
</table>
<h1 class="section"><a name="facet_accumulation">Accumulating Facets Illustrated</a></h1>
<p>
Facets accumulation reflects a set of documents over some facet requests:
<ul>
<li><code>Document set</code> - a subset of the index documents, usually documents
matching a user query.</li>
<li><code>Facet requests</code> - facet accumulation specification, e.g. count a certain facet
<i>dimension</i>.</li>
</ul>
<p>
<code>FacetRequest</code> is a basic component in faceted search - it describes the facet
information need. Every facet <b>request</b> is made of at least two fields:
<ul>
<li><code>CategoryPath</code> - root category of the facet request. The categories that
are returned as a result of the request will all be descendants of this root</li>
<li><code>Number of Results</code> - number of sub-categories to return (at most).</li>
</ul>
<p>
There are other parameters to a facet request, such as -how many facet results to
label-, -how <b>deep</b> to go from the request root when serving the facet request- and
more - see the API Javadocs for <code>FacetRequest</code> and its subclasses for more
information on these parameters. For labels in particular, see the section <a href="#facet_labels">Facet Labels
at Search Time</a>.
<p>
<code>FacetRequest</code> in an abstract class, open for extensions, and users may add their
own requests. The most often used request is <code>CountFacetRequest</code> - used for
counting facets.
<p>
Facets accumulation is - not surprisingly - driven by a <code>FacetsAccumulator</code>. The
most used one is <code>StandardFacetsAccumulator</code>, however there are also accumulators
that support sampling - to be used in huge collections, and there's an adaptive facets
accumulator which applies sampling conditionally on the statistics of the data. While
facets accumulators are very extendible and powerful, they might be too
overwhelming for beginners. For this reason, the code offers a higher level interface
for facets accumulating: the <code>FacetsCollector</code>. It extends <code>Collector</code>, and as such
can be passed to the search() method of Lucene's <code>IndexSearcher</code>. In case the
application also needs to collect documents (in addition to accumulating/collecting
facets), it can wrap multiple collectors with <code>MultiCollector</code>. Most code samples
below use <code>FacetsCollector</code> due to its simple interface. It is quite likely that
<code>FacetsCollector</code> should suffice the needs of most applications, therefore we
recommend to start with it, and only when needing more flexibility turn to directly
use facets accumulators.
<p>
Following is a code snippet from the example code - the complete example can be
found under <code>org.apache.lucene.facet.example.simple.Searcher</code>:
<pre class="prettyprint lang-java linenums">
IndexReader indexReader = IndexReader.open(indexDir);
Searcher searcher = new IndexSearcher(indexReader);
TaxonomyReader taxo = new LuceneTaxonomyReader(taxoDir);
...
Query q = new TermQuery(new Term(SimpleUtils.TEXT, "white"));
TopScoreDocCollector tdc = TopScoreDocCollector.create(10, true);
...
FacetSearchParams facetSearchParams = new FacetSearchParams();
facetSearchParams.addFacetRequest(new CountFacetRequest(
new CategoryPath("author"), 10));
...
FacetsCollector facetsCollector = new FacetsCollector(facetSearchParams, indexReader, taxo);
searcher.search(q, MultiCollector.wrap(topDocsCollector, facetsCollector));
List&lt;FacetResult&gt; res = facetsCollector.getFacetResults();
</pre>
<p>
We now explain the steps above, following the code line numbers:
<table class="code_description">
<tr>
<td>(1)</td>
<td>Index reader and Searcher are initialized as usual.</td>
</tr>
<tr>
<td>(3)</td>
<td>A taxonomy reader is opened - it provides access to the facet information
which was stored by the Taxonomy Writer at indexing time.</td>
</tr>
<tr>
<td>(5)</td>
<td>Regular text query is created to find the documents matching user need, and
a collector for collecting the top matching documents is created.</td>
</tr>
<tr>
<td>(8)</td>
<td>Facet-search-params is a container for facet requests.</td>
</tr>
<tr>
<td>(10)</td>
<td>A single facet-request - namely a count facet request - is created and added
to the facet search params. The request should return top 10 Author
subcategory counts.</td>
</tr>
<tr>
<td>(12)</td>
<td>Facets-Collector is the simplest interface for facets accumulation (counting
in this example).</td>
</tr>
<tr>
<td>(13)</td>
<td>Lucene search takes both collectors - facets-collector and top-doccollector,
both wrapped by a multi-collector. This way, a single search
operation finds both top documents and top facets. Note however that facets
aggregation takes place not only over the top documents, but rather over all
documents matching the query.</td>
</tr>
<tr>
<td>(14)</td>
<td>Once search completes, facet-results can be obtained from the facetscollector.</td>
</tr>
</table>
<p>
Returned facet results are organized in a list, conveniently ordered the same as the
facet-requests in the facet-search-params. Each result however contains the request
for which it was created.</li>
<p>
Here is the (recursive) structure of the facet result:
<ul>
<li><b>Facet Result</b>
<ul>
<li><b>Facet Request</b> - the request for which this result was obtained.</li>
<li><b>Valid Descendants</b> - how many valid descendants were encountered
over the set of matching documents (some of which might have been
filtered out because e.g. only top 10 results were requested).</li>
<li><b>Root Result Node</b> - root facet result for the request
<ul>
<li><b>Ordinal</b> - unique internal ID of the facet</li>
<li><b>Label</b> - full label of the facet (possibly null)</li>
<li><b>Value</b> - facet value, e.g. count</li>
<li><b>Sub-results-nodes</b> - child result nodes (possibly null)</li>
</ul></li>
</ul></li>
</ul>
<p>
Note that not always there would be sub result nodes - this depends on the
requested result mode:
<ul>
<li><b>PER_NODE_IN_TREE</b> - a tree, and so there may be sub results.</li>
<li><b>GLOBAL_FLAT</b> - here the results tree would be rather flat, with only (at
most) leaves below the root result node.</li>
</ul>
<h1 class="section"><a name="indexed_facet_info">Indexed Facet Information</a></h1>
<p>
When indexing a document to which categories were added, information on these
categories is added to the search index, in two locations:
<ul>
<li><i>Category Tokens</i> are added to the document for each category attached to
that document. These categories can be used at search time for drill-down.</li>
<li>A special <i>Category List Token</i> is added to each document containing
information on all the categories that were added to this document. This can
be used at search time for facet accumulation, e.g. facet counting.</li>
</ul>
<p>
When a category is added to the index (that is, when a document containing a
category is indexed), all its parent categories are added as well. For example, indexing
a document with the category <code>&lt;<span style="color: blue">"author"</span>,
<span style="color: blue">"American-</span>, <span style="color: blue">"Mark Twain"</span>&gt;</code> results in
creating three tokens: <code>"/author"</code>, <code>"/author/American"</code>, and
<code>"/author/American/Mark Twain"</code> (the character <code>'/'</code> here is just a human
readable separator - there's no such element in the actual index). This allows drilling down
and counting any category in the taxonomy, and not just leaf nodes, enabling a
UI application to show either how many books have authors, or how many books
have American authors, or how many books have Mark Twain as their (American)
author.
<p>
Similarly, Drill-down capabilities are this way possible also for node categories.
<p>
In order to keep the counting list compact, it is built using category ordinal - an
ordinal is an integer number attached to a category when it is added for the first time
into the taxonomy.
<p>
For ways to further alter facet index see the section below on <a href="#indexing_params">Facet Indexing
Parameters</a>.
<h1 class="section"><a name="taxonomy_index">Taxonomy Index</a></h1>
<p>
The taxonomy is an auxiliary data-structure maintained side-by-side with the regular
index to support faceted search operations. It contains information about all the
categories that ever existed in any document in the index. Its API is open and allows
simple usage, or more advanced for the interested users.
<p>
When a category is added to a document, a corresponding node is added to the
taxonomy (unless already there). In fact, sometimes more than one node is added -
each parent category is added as well, so that the taxonomy is maintained as a Tree,
with a virtual root.
<p>
So, for the above example, adding the category the category <code>&lt;<span style="color: blue">"author"</span>,
<span style="color: blue">"American-</span>, <span style="color: blue">"Mark Twain"</span>&gt;</code>
actually added three nodes: one for <code>"/author"</code>, one for <code>"/author/American"</code> and one for
<code>"/author/American/Mark Twain"</code>.
<p>
An integer number - called ordinal is attached to each category the first time the
category is added to the taxonomy. This allows for a compact representation of
category list tokens in the index, for facets accumulation.
<p>
One interesting fact about the taxonomy index is worth knowing: once a category
is added to the taxonomy, it is never removed, even if all related documents are
removed. This differs from a regular index, where if all documents containing a
certain term are removed, and their segments are merged, the term will also be
removed. This might cause a performance issue: large taxonomy means large ordinal
numbers for categories, and hence large categories values arrays would be maintained
during accumulation. It is probably not a real problem for most applications, but be
aware of this. If, for example, an application at a certain point in time removes an
index entirely in order to recreate it, or, if it removed all the documents from the index
in order to re-populate it, it also makes sense in this opportunity to remove the
taxonomy index and create a new, fresh one, without the unused categories.
<h1 class="section"><a name="facet_params">Facet Parameters</a></h1>
<p>
Facet parameters control how categories and facets are indexed and searched. Apart
from specifying facet requests within facet search parameters, under default settings it
is not required to provide any parameters, as there are ready to use working defaults
for everything.
<p>
However many aspects are configurable and can be modified by providing altered
facet parameters for either search or indexing.
<h2 class="subsection"><a name="indexing_params">Facet Indexing Parameters</a></h2>
<p>
Facet Indexing Parameters are consulted with during indexing. Among several
parameters it defines, the following two are likely to interest many applications:
<ul>
<li><b>Category list definitions</b> - in the index, facets are maintained in two
forms: category-tokens (for drill-down) and category-list-tokens (for
accumulation). This parameter allows to specify, for each category, the
Lucene term used for maintaining the category-list-tokens for that category.
The default implementation in <code>DefaultFacetIndexingParams</code> maintains
this information for all categories under the same special dedicated term.
One case where it is needed to maintain two categories in separate category
lists, is when it is known that at search time it would be required to use
different types of accumulation logic for each, but at the same accumulation
call.</li>
<li><b>Partition size</b> - category lists can be maintained in a partitioned way. If,
for example, the partition size is set to 1000, a distinct sub-term is used for
maintaining each 1000 categories, e.g. term1 for categories 0 to 999, term2
for categories 1000 to 1999, etc. The default implementation in
<code>DefaultFacetIndexingParams</code> maintains category lists in a single
partition, hence it defines the partition size as <code>Integer.MAX_VALUE</code>. The
importance of this parameter is on allowing to handle very large
taxonomies without exhausting RAM resources. This is because at facet
accumulation time, facet values arrays are maintained in the size of the
partition. With a single partition, the size of these arrays is as the size of the
taxonomy, which might be OK for most applications. Limited partition
sizes allow to perform the accumulation with less RAM, but with some
runtime overhead, as the matching documents are processed for each of the
partitions.</li>
</ul>
<p>
See the API Javadocs of <code>FacetIndexingParams</code> for additional configuration
capabilities which were not discussed here.
<h2 class="subsection"><a name="search_params">Facet Search Parameters</a></h2>
<p>
Facet Search Parameters, consulted at search time (during facets accumulation) are
rather plain, providing the following:
<ul>
<li><b>Facet indexing parameters</b> - which were in effect at indexing time -
allowing facets accumulation to understand how facets are maintained in
the index.</li>
<li><b>Container of facet requests</b> - the requests which should be accumulated.</li>
</ul>
<h2 class="subsection"><a name="category_lists_multiple_dimensions">Category Lists, Multiple Dimensions</a></h2>
<p>
Category list parameters which are accessible through the facet indexing parameters
provide the information about:
<ul>
<li>Lucene Term under which category information is maintained in the index.</li>
<li>Encoding (and decoding) used for writing and reading the categories
information in the index.</li>
</ul>
<p>
For cases when certain categories should be maintained in different location than
others, use <code>PerDimensionIndexingParams</code>, which returns a different
<code>CategoryListParams</code> object for each <i>dimension</i>. This is a good opportunity to
explain about dimensions. This is just a notion: the top element - or first element - in
a category path is denoted as the dimension of that category. Indeed, the dimension
stands out as a top important part of the category path, such as <code>"Location"</code> for the
category <code>"Location/Europe/France/Paris"</code>.
<h1 class="section"><a name="advanced">Advanced Faceted Examples</a></h1>
<p>
We now provide examples for more advanced facet indexing and search, such as
drilling-down on facet values and multiple category lists.
<h2 class="subsection"><a name="drill_down">Drill-Down with Regular Facets</a></h2>
<p>
Drill-down allows users to focus on part of the results. Assume a commercial sport
equipment site where a user is searching for a tennis racquet. The user issues the
query <i>tennis racquet</i> and as result is shown a page with 10 tennis racquets, by
various providers, of various types and prices. In addition, the site UI shows to the
user a break down of all available racquets by price and make. The user now decides
to focus on racquets made by <i>Head</i>, and will now be shown a new page, with 10
Head racquets, and new break down of the results into racquet types and prices.
Additionally, the application can choose to display a new breakdown, by racquet
weights. This step of moving from results (and facet statistics) of the entire (or larger)
data set into a portion of it by specifying a certain category, is what we call <i>Drilldown</i>.
We now show the required code lines for implementing such a drill-down.
<pre class="prettyprint lang-java linenums">
Query baseQuery = queryParser.parse("tennis racquet");
Query q2 = DrillDown.query(baseQuery, new CategoryPath("make", "head"), 10));
</pre>
<p>
In line 1 the original user query is created and then used to obtain information on
all tennis racquets.
<p>
In line 2, a specific category from within the facet results was selected by the user,
and is hence used for creating the drill-down query.
<p>
Please refer to <code>SimpleSearcher.searchWithDrillDown()</code> for a more detailed
code example performing drill-down.
<h2 class="subsection"><a name="multi-category_list">Multiple Category Lists</a></h2>
<p>
The default is to maintain all categories information in a single list. While this will
suit most applications, in some situations an application may wish to use multiple
category lists, for example, when the distribution of some category values is different
than that of other categories and calls for using a different encoding, more efficient
for the specific distribution. Another example is when most facets are rarely used
while some facets are used very heavily, so an application may opt to maintain the
latter in memory - and in order to keep memory footprint lower it is useful to
maintain only those heavily used facets in a separate category list.
<p>
First we define indexing parameters with multiple category lists:
<pre class="prettyprint lang-java linenums">
PerDimensionIndexingParams iParams = new PerDimensionIndexingParams();
iParams.addCategoryListParams(new CategoryPath("Author"),
new CategoryListParams(new Term("$RarelyUsed", "Facets")));
iParams.addCategoryListParams(new CategoryPath("Language"),
new CategoryListParams(new Term("$HeavilyUsed", "Ones")));
</pre>
<p>
This will cause the Language categories to be maintained in one category list, and
Author facets to be maintained in a another category list. Note that any other category,
if encountered, will still be maintained in the default category list.
<p>
These non-default indexing parameters should now be used both at indexing and
search time. As depicted below, at indexing time this is done when creating the
category document builder, while at search time this is done when creating the search
parameters. Other than that the faceted search code is unmodified.
<pre class="prettyprint lang-java linenums">
DocumentBuilder categoryDocBuilder = new CategoryDocumentBuilder(taxo, iParams);
...
FacetSearchParams facetSearchParams = new FacetSearchParams(iParams);
</pre>
<p>
A complete simple example can be found in package <code>org.apache.lucene.facet.example.multiCL</code>
under the example code.
<h1 class="section"><a name="optimizations">Optimizations</a></h1>
<p>
Faceted search through a large collection of documents with large numbers of facets
altogether and/or large numbers of facets per document is challenging performance
wise, either in CPU, RAM, or both. A few ready to use optimizations exist to tackle
these challenges.
<h2 class="subsection"><a name="sampling">Sampling</a></h2>
<p>
Facet sampling allows to accumulate facets over a sample of the matching
documents set. In many cases, once top facets are found over the sample set, exact
accumulations are computed for those facets only, this time over the entire matching
document set.
<p>
Two kinds of sampling exist: complete support and wrapping support. The
complete support is through <code>SamplingAccumulator</code> and is tied to an extension of the
<code>StandardFacetsAccumulator</code> and has the benefit of automatically applying other
optimizations, such as <a href="#complements">Complements</a>. The wrapping support is through
<code>SamplingWrapper</code> and can wrap any accumulator, and as such, provides more
freedom for applications.
<h2 class="subsection"><a name="complements">Complements</a></h2>
<p>
When accumulating facets over a very large matching documents set, possibly
almost as large as the entire collection, it is possible to speed up accumulation by
looking at the complement set of documents, and then obtaining the actual results by
subtracting from the total results. It should be noted that this is available only for
count requests, and that the first invocation that involves this optimization might take
longer because the total counts have to be computed.
<p>
This optimization is applied automatically by <code>StandardFacetsAccumulator</code>.
<h2 class="subsection"><a name="partitions">Partitions</a></h2>
<p>
Partitions are also discussed in the section about <a href="#indexing_params">Facet Indexing parameters.</a>
<p>
Facets are internally accumulated by first accumulating all facets and later on
extracting the results for the requested facets. During this process, accumulation
arrays are maintained in the size of the taxonomy. For a very large taxonomy, with
multiple simultaneous faceted search operations, this might lead to excessive memory
footprint. Partitioning the faceted information allows to relax the memory usage, by
maintaining the category lists in several partitions, and by processing one partition at
a time. This is automatically done by <code>StandardFacetsAccumulator</code>. However the
default partition size is <code>Integer.MAX_VALUE</code>, practically setting to a single partition,
i.e. no partitions at all.
<p>
Decision to override this behavior and use multiple partitions must be taken at
indexing time. Once the index is created and already contains category lists it is too
late to modify this.
<p>
See <code>FacetIndexingParams.getPartitionSize()</code> for API to alter this default
behavior.
<h1 class="section"><a name="concurrent_indexing_search">Concurrent Indexing and Search</a></h1>
<p>
Sometimes, indexing is done once, and when the index is fully prepared, searching
starts. However, in most real applications indexing is <i>incremental</i> (new data comes in
once in a while, and needs to be indexed), and indexing often needs to happen while
searching is continuing at full steam.
<p>
Luckily, Lucene supports multiprocessing - one process writing to an index while
another is reading from it. One of the key insights behind how Lucene allows multiprocessing
is <i>Point In Time</i> semantics. The idea is that when an <code>IndexReader</code> is opened,
it gets a view of the index at the <i>point in time</i> it was opened. If an <code>IndexWriter</code>
in a different process or thread modifies the index, the reader does not know about it until a new
<code>IndexReader</code> is opened (or the reopen() method of an existing <code>IndexReader</code> is called).
<p>
In faceted search, we complicate things somewhat by adding a second index - the
taxonomy index. The taxonomy API also follows point-in-time semantics, but this is
not quite enough. Some attention must be paid by the user to keep those two indexes
consistently in sync:
<p>
The main index refers to category numbers defined in the taxonomy index.
Therefore, it is important that we open the <code>TaxonomyReader</code> <i>after</i> opening the
IndexReader. Moreover, every time an IndexReader is reopen()ed, the
TaxonomyReader needs to be refresh()'ed as well.
<p>
But there is one extra caution: whenever the application deems it has written
enough information worthy a commit, it must <b>first</b> call commit() for the
<code>TaxonomyWriter</code> and only <b>after</b> that call commit() for the <code>IndexWriter</code>.
Closing the indices should also be done in this order - <b>first</b> close the taxonomy, and only <b>after</b>
that close the index.
<p>
To summarize, if you're writing a faceted search application where searching and
indexing happens concurrently, please follow these guidelines (in addition to the usual
guidelines on how to use Lucene correctly in the concurrent case):
<ul>
<li>In the indexing process:
<ol>
<li>Before a writer commit()s the IndexWriter, it must commit() the
TaxonomyWriter. Nothing should be added to the index between these
two commit()s.</li>
<li>Similarly, before a writer close()s the IndexWriter, it must close() the
TaxonomyWriter.</li>
</ol></li>
<li>In the searching process:
<ol>
<li>Open the IndexReader first, and then the TaxonomyReader.</li>
<li>After a reopen() on the IndexReader, refresh() the TaxonomyReader.
No search should be performed on the new IndexReader until refresh()
has finished.</li>
</ol></li>
</ul>
<p>
Note that the above discussion assumes that the underlying file-system on which
the index and the taxonomy are stored respects ordering: if index A is written before
index B, then any reader finding a modified index B will also see a modified index A.
<p>
<b>Note:</b> <code>TaxonomyReader</code>'s refresh() is simpler than <code>IndexReader</code>'s reopen().
While the latter keeps both the old and new reader open, the former keeps only the new reader. The reason
is that a new <code>IndexReader</code> might have modified old information (old documents deleted, for
example) so a thread which is in the middle of a search needs to continue using the old information. With
<code>TaxonomyReader</code>, however, we are guaranteed that existing categories are never deleted or modified -
the only thing that can happen is that new categories are added. Since search threads do not care if new categories
are added in the middle of a search, there is no reason to keep around the old object, and the new one suffices.
</body>
</html>

View File

@ -62,10 +62,10 @@ public class TaxonomyMergeUtils {
* Merges the given taxonomy and index directories and commits the changes to
* the given writers. This method uses {@link MemoryOrdinalMap} to store the
* mapped ordinals. If you cannot afford the memory, you can use
* {@link #merge(Directory, Directory, OrdinalMap, IndexWriter, LuceneTaxonomyWriter)}
* {@link #merge(Directory, Directory, LuceneTaxonomyWriter.OrdinalMap, IndexWriter, LuceneTaxonomyWriter)}
* by passing {@link DiskOrdinalMap}.
*
* @see #merge(Directory, Directory, OrdinalMap, IndexWriter, LuceneTaxonomyWriter)
* @see #merge(Directory, Directory, LuceneTaxonomyWriter.OrdinalMap, IndexWriter, LuceneTaxonomyWriter)
*/
public static void merge(Directory srcIndexDir, Directory srcTaxDir,
IndexWriter destIndexWriter,

View File

@ -3,6 +3,6 @@
<title>Faceted Indexing and Search</title>
</head>
<body>
Provides faceted indexing and search capabilities.
Provides faceted indexing and search capabilities. The <a href="../../../../userguide.html">userguide</a> is recommended for a start.
</body>
</html>

View File

@ -21,6 +21,6 @@
</title>
</head>
<body>
facet
Provides faceted indexing and search capabilities (checkout the <a href="userguide.html">userguide</a>).
</body>
</html>