本文将重点介绍Elasticsearch Query DSL之Compound queries(复合查询)。
复合查询将其他复合查询或叶子查询进行包装,组合它们的结果和分数,以此改变它们的行为,或从查询字句切换到过滤上下文模式。
主要的复合查询包括如下:
- constant_score query
- bool query
- dis_max query
- function_score query
- boosting query
常量(score)评分查询,该复合查询将忽略文档本身的匹配相关性评分,而是统一返回请求参数的boost。实例如下(Java):
1public static void testConstantScoreQuery() { 2 RestHighLevelClient client = EsClient.getClient(); 3 try { 4 SearchRequest searchRequest = new SearchRequest(); 5 searchRequest.indices("twitter"); 6 SearchSourceBuilder sourceBuilder = new SearchSourceBuilder(); 7 sourceBuilder.query( 8 QueryBuilders.constantScoreQuery(QueryBuilders.wildcardQuery("user", "ding*")) 9 .boost(1.5f) 10 ); 11 searchRequest.source(sourceBuilder); 12 SearchResponse result = client.search(searchRequest, RequestOptions.DEFAULT); 13 System.out.println(result); 14 } catch (Throwable e) { 15 e.printStackTrace(); 16 } finally { 17 EsClient.close(client); 18 } 19 }
返回的结果为:为了对比,左边的结果是Qu-eryBuilders.wildcardQuery("user", "ding*")查询,而右边的是constant_score (复合查询)。
1{ { 2 "took":4, "took":2, 3 "timed_out":false, "timed_out":false, 4 "_shards":{ "_shards":{ 5 "total":5, "total":5, 6 "successful":5, "successful":5, 7 "skipped":0, "skipped":0, 8 "failed":0 "failed":0 9 }, }, 10 "hits":{ "hits":{ 11 "total":1, "total":1, 12 "max_score":0.9808292, "max_score":1.5, 13 "hits":[ "hits":[ 14 { { 15 "_index":"twitter", "_index":"twitter", 16 "_type":"_doc", "_type":"_doc", 17 "_id":"12", "_id":"12", 18 "_score":0.9808292, "_score":1.5, 19 "_source":{ "_source":{ 20 "post_date":"2009-11-18T14:12:12", "post_date":"2009-11-18T14:12:12", 21 "message":"test bulk", "message":"test bulk", 22 "user":"dingw" "user":"dingw" 23 } } 24 } } 25 ] ] 26 } } 27}
布尔查询。bool query里能包含的主要子句类型如下:
- must
该字句类型的查询语句,文档必须满足,并对评分产生影响(相关度) - filter
子句(查询)必须出现在匹配的文档中。然而与must不同的是查询的分数将被忽略。过滤器子句在过滤器上下文中执行,子句被考虑用于缓存。 - should
应该匹配;如果没有must和filter,多个s-hould只需要至少一个匹配即可,该数据可以通过参数minimum_should_match控制,如果包含了must或filter,则should不参与实际过滤,但会参与评分。 - must_not
查询条件取反,及匹配到的文档必须不符合must_not的条件。
filter context中查询对相关性的影响
在过滤上下文环境的查询字句并不会对相关性产生影响,也就是说过滤上下文中的查询子句返回的score为0。
例如如下查询示例(使用Java编写):
1public static void testBoolQuery_filterContext_score() { 2 RestHighLevelClient client = EsClient.getClient(); 3 try { 4 SearchRequest searchRequest = new SearchRequest(); 5 searchRequest.indices("twitter"); 6 SearchSourceBuilder sourceBuilder = new SearchSourceBuilder(); 7 sourceBuilder.query( 8 QueryBuilders.boolQuery() 9 .filter(QueryBuilders.termQuery("user", "dingw")) 10 ); 11 searchRequest.source(sourceBuilder); 12 SearchResponse result = client.search(searchRequest, RequestOptions.DEFAULT); 13 System.out.println(result); 14 } catch (Throwable e) { 15 e.printStackTrace(); 16 } finally { 17 EsClient.close(client); 18 } 19 } }
其返回的结果,其score都为0,结果如下:
1public static void testBoolQuery_filterContext_score() { 2 RestHighLevelClient client = EsClient.getClient(); 3 try { 4 SearchRequest searchRequest = new SearchRequest(); 5 searchRequest.indices("twitter"); 6 SearchSourceBuilder sourceBuilder = new SearchSourceBuilder(); 7 sourceBuilder.query( 8 QueryBuilders.boolQuery() 9 .filter(QueryBuilders.termQuery("user", "dingw")) 10 ); 11 searchRequest.source(sourceBuilder); 12 SearchResponse result = client.search(searchRequest, RequestOptions.DEFAULT); 13 System.out.println(result); 14 } catch (Throwable e) { 15 e.printStackTrace(); 16 } finally { 17 EsClient.close(client); 18 } 19 } }
该查询方式将所有查询字句进行联合查询(uni-oion)只需要其中一个条件匹配则返回文档,但在计算相关性时不是将所有条件的匹配度(
score)相加,而是使用评分最高的查询条件的score;如果有指定tie_breaker的话,则为最大score 加上 其他score * tie_breaker。dis max query是实现(match query multi fields best_-fields)的核心。每个查询可以指定其评分因子(权重、boost)、dis max query使用示例:
1/** 2 * dis max query 3 */ 4 public static void testDisMaxQuery() { 5 RestHighLevelClient client = EsClient.getClient(); 6 try { 7 SearchRequest searchRequest = new SearchRequest(); 8 searchRequest.indices("twitter"); 9 SearchSourceBuilder sourceBuilder = new SearchSourceBuilder(); 10 sourceBuilder.query( 11 QueryBuilders.disMaxQuery() 12// .tieBreaker(0.1f) 13 .add(QueryBuilders.termQuery("user", "dingw").boost(1.2f)) 14 .add(QueryBuilders.termQuery("message", "bulk")) 15 ); 16 searchRequest.source(sourceBuilder); 17 SearchResponse result = client.search(searchRequest, RequestOptions.DEFAULT); 18 System.out.println(result); 19 } catch (Throwable e) { 20 e.printStackTrace(); 21 } finally { 22 EsClient.close(client); 23 } 24 } }
函数分数查询,暂不深入学习。
boosting query可以用来提升或降低某些查询条件的权重。举例如下:
1GET /_search 2{ 3 "query": { 4 "boosting" : { 5 "positive" : { // @1 6 "term" : { 7 "field1" : "value1" 8 } 9 }, 10 "negative" : { // @2 11 "term" : { 12 "field2" : "value2" 13 } 14 }, 15 "negative_boost" : 0.2 16 } 17 } 18} }
代码@1:积极的作用,提升其权重。
代码@2:负面的影响,希望降低其权重,其权重值通过negative_boost指定。