写在前面

在这篇文章 我们看了使用dis_max查询来进行单值多字段场景下的查询，如下：

POST /blogs/_search
{"query": {"dis_max": {"queries": [{"match": {"title": "Brown fox"}},{"match": {"body": "Brown fox"}}]}}
}

这里不知道你注意到没有，Brown fox我们重复写了N遍，即要查询的字段越多则重复写的次数也越多，想要解决这个问题，就可以使用本文要学习的multi_match了。

1：multi_match的三种方式

1.1：best_field

这种方式使用每个文档中字段的最高得分作为最终得分进行匹配，这和dis max query 是一样的效果，如下的查询：

DELETE blogs
PUT /blogs/_doc/1
{"title": "Quick brown rabbits","body": "Brown rabbits are commonly seen."
}PUT /blogs/_doc/2
{"title": "Keeping pets healthy","body": "My quick brown fox eats rabbits on a regular basis"
}POST /blogs/_search
{"query": {"dis_max": {"tie_breaker": 0, "queries": [{"match": {"title": "Brown fox"}},{"match": {"body": "Brown fox"}}]}}
}

同样也可以使用multi_match的best_field来实现：

POST /blogs/_search
{"query": {"multi_match": {"query": "Brown fox","type": "best_fields","fields": ["title","body"],"tie_breaker": 0}}
}

1.2：most_field

这种方式是某个文档匹配的字段越多，则得分越高，如下：

POST news/_bulk
{"index": {"_id": 1}}
{"f1":"aa", "f2": "bb"}
{"index": {"_id": 2}}
{"f1":"aa", "f2": "bb", "f3": "cc"}
{"index": {"_id": 3}}
{"f1":"aa", "f2": "bb", "f3": "cc", "f4": "dd"}POST /news/_search
{"query": {"multi_match": {"query": "aa bb cc dd","type": "most_fields","fields": ["f1","f2","f3","f4"]}}
}

可以看到匹配的fields越多则越靠前。

其他特殊的情况分析。

• 如果没有任何匹配的field则不会匹配返回，如下：
  
• 如果是field匹配的field数完全相同，且匹配的term数相同，则得分完全相同，如下：

DELETE news/POST news/_bulk
{"index": {"_id": 1}}
{"f1":"aa", "f2": "bb"}
{"index": {"_id": 2}}
{"f1":"aa", "f2": "bb"}POST /news/_search
{"query": {"multi_match": {"query": "aa bb","type": "most_fields","fields": ["f1","f2","f3","f4"]}}
}

• 如果field匹配的field数完全相同，匹配的term越多，则得分越高，如下：

DELETE news/POST news/_bulk
{"index": {"_id": 1}}
{"f1":"aa", "f2": "bb"}
{"index": {"_id": 2}}
{"f1":"aa", "f2": "aa bb"}POST /news/_search
{"query": {"multi_match": {"query": "aa bb","type": "most_fields","fields": ["f1","f2","f3","f4"]}}
}

即匹配的field数优先，field数相同则按照总得分排优先级。

1.3：cross_field

这种匹配方式是将查询的字段作为一个整体来进行查询，每个要查询的词项都需要在文档中存在，才会匹配成功,如果是不使用multi_match的cross field的话我们也可以使用copy_to的方式，来将要查询的字段全部copy_to到同一个字段中，然后对该段进行普通的查询，如下：

DELETE news/
PUT news
{"mappings": {"properties": {"f1": {"type": "text","copy_to": "f_full"},"f2": {"type": "text","copy_to": "f_full"},"f3": {"type": "text","copy_to": "f_full"}}}
}POST news/_bulk
{"index": {"_id": 1}}
{"f1":"aa", "f2": "bb"}
{"index": {"_id": 2}}
{"f1":"aa", "f2": "bb"}
{"index": {"_id": 3}}
{"f1":"aa", "f2": "bb", "f3": "cc"}POST /news/_search
{"query": {"match": {"f_full": {"query": "aa bb cc","operator": "and"}}}
}

查询的是在f_full中包含aa，bb，cc的文档，但是copy_to的方式有一个缺点就是，会增加磁盘的负担，如果是使用cross_field可以等效的解决这个问题，如下：

DELETE news/
PUT news
{"mappings": {"properties": {"f1": {"type": "text","copy_to": "f_full"},"f2": {"type": "text","copy_to": "f_full"},"f3": {"type": "text","copy_to": "f_full"}}}
}POST news/_bulk
{"index": {"_id": 1}}
{"f1":"aa", "f2": "bb"}
{"index": {"_id": 2}}
{"f1":"aa", "f2": "bb"}
{"index": {"_id": 3}}
{"f1":"aa", "f2": "bb", "f3": "cc"}POST /news/_search
{"query": {"multi_match": {"query": "aa bb cc","type": "cross_fields","fields": ["f1","f2","f3"],"operator": "and"}},"profile": "true"
}

通过profile可以看到查询的方式是+(f2:aa | f3:aa | f1:aa) +(f2:bb | f3:bb | f1:bb) +(f2:cc | f3:cc | f1:cc),因为operator是and，所以同sql(f1 like '%aa%' or f2 like '%aa%' or f3 like '%aa%') and (f1 like '%bb%' or f2 like '%bb%' or f3 like '%bb%') and (f1 like '%cc%' or f2 like '%cc%' or f3 like '%cc%'),同样的如果是将operator改为or，则同sql(f1 like '%aa%' or f2 like '%aa%' or f3 like '%aa%') or (f1 like '%bb%' or f2 like '%bb%' or f3 like '%bb%') or (f1 like '%cc%' or f2 like '%cc%' or f3 like '%cc%')。

写在后面

参考文章列表

ES中的Multi_match深入解读：best_fields、most_fields、cross_fields用法一览 。