Elasticsearch 6.8 match_phrase 搜索 N-gram 分词器效果不佳

Question

我使用 Elasticsearch N-gram tokenizer 并使用 match_phrase 进行模糊匹配我的索引和测试数据如下：

DELETE /m8
PUT m8
{
  "settings": {
    "analysis": {
      "analyzer": {
        "my_analyzer": {
          "tokenizer": "my_tokenizer"
        }
      },
      "tokenizer": {
        "my_tokenizer": {
          "type": "ngram",
          "min_gram": 1,
          "max_gram": 3,
          "custom_token_chars":"_."
        }
      }
    },
    "max_ngram_diff": 10
  },
  "mappings": {
    "table": {
      "properties": {
        "dataSourceId": {
          "type": "long"
        },
        "dataSourceType": {
          "type": "integer"
        },
        "dbName": {
          "type": "text",
          "analyzer": "my_analyzer",
          "fields": {
            "keyword": {
              "type": "keyword",
              "ignore_above": 256
            }
          }
        }
      }
    }
  }
}


PUT /m8/table/1
{
  "dataSourceId":1,
  "dataSourceType":2,
  "dbName":"rm.rf"
}

PUT /m8/table/2
{
  "dataSourceId":1,
  "dataSourceType":2,
  "dbName":"rm_rf"
}
PUT /m8/table/3
{
  "dataSourceId":1,
  "dataSourceType":2,
  "dbName":"rmrf"
}

检查_analyze:

POST m8/_analyze
{
  "tokenizer": "my_tokenizer",
  "text": "rm.rf"
}

_分析结果：

{
  "tokens" : [
    {
      "token" : "r",
      "start_offset" : 0,
      "end_offset" : 1,
      "type" : "word",
      "position" : 0
    },
    {
      "token" : "rm",
      "start_offset" : 0,
      "end_offset" : 2,
      "type" : "word",
      "position" : 1
    },
    {
      "token" : "rm.",
      "start_offset" : 0,
      "end_offset" : 3,
      "type" : "word",
      "position" : 2
    },
    {
      "token" : "m",
      "start_offset" : 1,
      "end_offset" : 2,
      "type" : "word",
      "position" : 3
    },
    {
      "token" : "m.",
      "start_offset" : 1,
      "end_offset" : 3,
      "type" : "word",
      "position" : 4
    },
    {
      "token" : "m.r",
      "start_offset" : 1,
      "end_offset" : 4,
      "type" : "word",
      "position" : 5
    },
    {
      "token" : ".",
      "start_offset" : 2,
      "end_offset" : 3,
      "type" : "word",
      "position" : 6
    },
    {
      "token" : ".r",
      "start_offset" : 2,
      "end_offset" : 4,
      "type" : "word",
      "position" : 7
    },
    {
      "token" : ".rf",
      "start_offset" : 2,
      "end_offset" : 5,
      "type" : "word",
      "position" : 8
    },
    {
      "token" : "r",
      "start_offset" : 3,
      "end_offset" : 4,
      "type" : "word",
      "position" : 9
    },
    {
      "token" : "rf",
      "start_offset" : 3,
      "end_offset" : 5,
      "type" : "word",
      "position" : 10
    },
    {
      "token" : "f",
      "start_offset" : 4,
      "end_offset" : 5,
      "type" : "word",
      "position" : 11
    }
  ]
}

当我搜索 'rm' 时，没有找到：

GET /m8/table/_search
{
  "query": {
    "bool": {
      "must": [
        {
          "match_phrase": {
            "dbName": "rm"
          }
        }
      ]
    }
  }
}

{
  "took" : 2,
  "timed_out" : false,
  "_shards" : {
    "total" : 5,
    "successful" : 5,
    "skipped" : 0,
    "failed" : 0
  },
  "hits" : {
    "total" : 0,
    "max_score" : null,
    "hits" : [ ]
  }
}

但是可以找到“.rf”：

{
  "took" : 1,
  "timed_out" : false,
  "_shards" : {
    "total" : 5,
    "successful" : 5,
    "skipped" : 0,
    "failed" : 0
  },
  "hits" : {
    "total" : 1,
    "max_score" : 1.7260926,
    "hits" : [
      {
        "_index" : "m8",
        "_type" : "table",
        "_id" : "1",
        "_score" : 1.7260926,
        "_source" : {
          "dataSourceId" : 1,
          "dataSourceType" : 2,
          "dbName" : "rm.rf"
        }
      }
    ]
  }
}

我的问题: 为什么'rm'连_analyze拆分了这些词组也找不到？

Answer 1

my_analyzer 也将在搜索期间使用。

"mapping":{
 "dbName": {
  "type": "text",
  "analyzer": "my_analyzer" 
  "search_analyzer":"my_analyzer"  // <==== If you don't provide a search analyzer then what you defined in analyzer will be used during search time as well.

Match_phrase 查询用于根据分析文本的位置来匹配短语。例如，搜索“Kal ho”将匹配分析文本中位置 X 处有“Kal”和位置 X+1 处有“ho”的文档。
当您搜索 'rm' (#1) 时，使用 my_analyzer 分析文本，将其转换为 n-gram 并在其顶部phrase_search 将被使用。因此结果不是预期的。

解法：

使用带有简单匹配查询的标准分析器

GET /m8/_search
{
 "query": {
 "bool": {
   "must": [
     {
       "match": {
         "dbName": {
           "query": "rm",
           "analyzer": "standard" // <=========
         }
       }
     }
   ]
 }
 }
 }

OR 在映射期间定义并使用匹配查询（不是 match_phrase）

"mapping":{
      "dbName": {
       "type": "text",
       "analyzer": "my_analyzer" 
       "search_analyzer":"standard" //<==========

后续问题：为什么要将 match_phrase 查询与 n-gram 分词器一起使用？

Elasticsearch 6.8 match_phrase 搜索 N-gram 分词器效果不佳

Elasticsearch 6.8 match_phrase search N-gram tokenizer works not well

tokenize

n-gram

elasticsearch

match-phrase