Lucene/solr的评分公式

「爱情、让人受尽委屈。」 2022-06-02 10:48 180阅读 0赞

#     #

# 1. lucene 评分公式 #

（

评分公式中，对大多数因子的控制和实现都是通过Similarity抽象类的子类完成的。lucene默认使用DefaultSimilarity类。如果要详细了解的话可以直接看lucene源代码Similarity和DefaultSimilarity类。  
）

## Lucene Scoring 评分机制: ##

## [http://blog.chenlb.com/2009/08/lucene-scoring-architecture.html][http_blog.chenlb.com_2009_08_lucene-scoring-architecture.html] ##

# 2.[Lucene的评分(score)机制的简单解释][Lucene_score] #

# 3. dismax/edismax评分公式 #

## 3.1 dismax介绍： ##

What’s a “DisMax” ?：

[http://searchhub.org/dev/2010/05/23/whats-a-dismax/][http_searchhub.org_dev_2010_05_23_whats-a-dismax]

solr dismax/edismax ：

[http://wiki.apache.org/solr/DisMax][http_wiki.apache.org_solr_DisMax]

[http://wiki.apache.org/solr/ExtendedDisMax][http_wiki.apache.org_solr_ExtendedDisMax]

使用solr时可以在solrconfig.xml的<requestHandler name="search">结点中增加defType的设置：

<requestHandler name="search" class="solr.SearchHandler" default="true">
    	<lst name="defaults">
    		<str name="defType">edismax</str>	
    	</lst>
    </requesHandler>

## 3.2 solr使用edismax时的评分公式 ##

solr管理页面查询时增加参数debugQuery=true即可启用debug信息，可以清楚的看到在查询多个域时评分规则如下：

下面借用用python的对齐方式来展示计算规则：
    score = max plus 0.1 times others of:
    	value1 = field1 weight,product of:
    		queryWeight, product of:
    			boost
    			idf
    			queryNorm
    		fieldWeight, product of:
    			tf(termFreq)
    			idf
    			fieldNorm
    	
    	...
    	valuen = fieldn weight(计算方法同上)

假设有field1,field2,且field1 weight > field2 weight(即value1 > value2), 那么score = value1 plus 0.1 times value2（即score = value1 + 0.1 \* value2）

其中field weight时lucene的评分公式计算出来的，可以看出跟lucene用explain query显示的格式一样。

# 4.评分相关的Solr FAQ #

1.  [How can I increase the score for specific documents][]
    
    1.  [Query Elevation Component][]
    2.  [index-time boosts][]
    3.  [Field Based Boosting][]
2.  [How can I change the score of a document based on the \*value\* of a field (say, "popularity")][How can I change the score of a document based on the _value_ of a field _say_ _popularity]
3.  [How are documents scored][]
4.  [How can I boost the score of newer documents][]

以上内容是基于当前solr4.4版本的~

[http_blog.chenlb.com_2009_08_lucene-scoring-architecture.html]: http://blog.chenlb.com/2009/08/lucene-scoring-architecture.html
[Lucene_score]: http://lucene-group.group.iteye.com/group/wiki/1526-lucene-39-s-score-score-a-simple-mechanism-to-explain
[http_searchhub.org_dev_2010_05_23_whats-a-dismax]: http://searchhub.org/dev/2010/05/23/whats-a-dismax/
[http_wiki.apache.org_solr_DisMax]: http://wiki.apache.org/solr/DisMax
[http_wiki.apache.org_solr_ExtendedDisMax]: http://wiki.apache.org/solr/ExtendedDisMax
[How can I increase the score for specific documents]: http://wiki.apache.org/solr/SolrRelevancyFAQ#How_can_I_increase_the_score_for_specific_documents
[Query Elevation Component]: http://wiki.apache.org/solr/SolrRelevancyFAQ#Query_Elevation_Component
[index-time boosts]: http://wiki.apache.org/solr/SolrRelevancyFAQ#index-time_boosts
[Field Based Boosting]: http://wiki.apache.org/solr/SolrRelevancyFAQ#Field_Based_Boosting
[How can I change the score of a document based on the _value_ of a field _say_ _popularity]: http://wiki.apache.org/solr/SolrRelevancyFAQ#How_can_I_change_the_score_of_a_document_based_on_the_.2Avalue.2A_of_a_field_.28say.2C_.22popularity.22.29
[How are documents scored]: http://wiki.apache.org/solr/SolrRelevancyFAQ#How_are_documents_scored
[How can I boost the score of newer documents]: http://wiki.apache.org/solr/SolrRelevancyFAQ#How_can_I_boost_the_score_of_newer_documents