elasticsearch - 使完整单词比Edge NGram子集得分更高

Question

我试图在匹配全名的文档上获得更高的分数，而不是具有相同值的Edge NGram子集。

结果是:

Pos Name              _score       _id

1   Baritone horn     7.56878     1786
2   Baritone ukulele  7.56878     2313
3   Bari              7.56878     2360
4   Baritone voice    7.56878     1787

我本想使第三个(“Bari”)具有较高的分数，因为它是全名，但是，由于边缘ngram分解将使所有其他单词都具有完全被“bari”单词索引的索引。这样您就可以在结果表上看到所有分数都相等了，我什至都不知道 flex 搜索如何排序，因为_id甚至不是顺序的，也不是名称的顺序。

我该如何实现？

谢谢

示例“代码”

设定值

{
  "analysis": {
    "filter": {
      "edgeNGram_filter": {
        "type": "edgeNGram",
        "min_gram": 3,
        "max_gram": 20,
        "token_chars": [
          "letter",
          "digit",
          "punctuation",
          "symbol"
        ]
      }
    },
    "analyzer": {
      "edgeNGram_analyzer": {
        "type": "custom",
        "tokenizer": "whitespace",
        "filter": [
          "lowercase",
          "asciifolding",
          "edgeNGram_filter"
        ]
      },
      "whitespace_analyzer": {
        "type": "custom",
        "tokenizer": "whitespace",
        "filter": [
          "lowercase",
          "asciifolding"
        ]
      }
    }
  }
}

source

对应:

{
  "name": {
    "type": "string",
    "index": "not_analyzed"
  },
  "suggest": {
    "type": "completion",
    "index_analyzer": "nGram_analyzer",
    "search_analyzer": "whitespace_analyzer",
    "payloads": true
  }
}

查询:

POST /attribute-tree/attribute/_search
{
  "query": {
    "match": {
      "suggest": "Bari"
    }
  }
}

结果:

(仅留下相关数据)

{
  "took": 3,
  "timed_out": false,
  "_shards": {
    "total": 5,
    "successful": 5,
    "failed": 0
  },
  "hits": {
    "total": 4,
    "max_score": 7.56878,
    "hits": [
      {
        "_index": "attribute-tree",
        "_type": "attribute",
        "_id": "1786",
        "_score": 7.56878,
        "_source": {
          "name": "Baritone horn",
          "suggest": {
            "input": [
              "Baritone",
              "horn"
            ],
            "output": "Baritone horn"
          }
        }
      },
      {
        "_index": "attribute-tree",
        "_type": "attribute",
        "_id": "2313",
        "_score": 7.56878,
        "_source": {
          "name": "Baritone ukulele",
          "suggest": {
            "input": [
              "Baritone",
              "ukulele"
            ],
            "output": "Baritone ukulele"
          }
        }
      },
      {
        "_index": "attribute-tree",
        "_type": "attribute",
        "_id": "2360",
        "_score": 7.56878,
        "_source": {
          "name": "Bari",
          "suggest": {
            "input": [
              "Bari"
            ],
            "output": "Bari"
          }
        }
      },
      {
        "_index": "attribute-tree",
        "_type": "attribute",
        "_id": "1787",
        "_score": 7.568078,
        "_source": {
          "name": "Baritone voice",
          "suggest": {
            "input": [
              "Baritone",
              "voice"
            ],
            "output": "Baritone voice"
          }
        }
      }
    ]
  }
}

Answer 1

您可以使用bool查询运算符及其should子句将分数添加到完全匹配项中，如下所示:

POST /attribute-tree/attribute/_search
{
  "query": {
    "bool": {
      "must": [
        {
          "match": {
            "suggest": "Bari"
          }
        }
      ],
      "should": [
        {
          "match": {
            "name": "Bari"
          }
        }
      ]
    }
  }
}

should子句中的查询在 ElasticSearch definitive guide中称为signal子句，这是您可以区分完全匹配和ngram匹配的方式。您将拥有与must子句匹配的所有文档，但是由于 should查询的得分公式，与 bool查询匹配的文档的得分更高:

score = ("must" queries total score + matching "should" queries total score) / (total number of "must" queries and "should" queries)

结果就是您所期望的，Bari是第一个结果(在得分上遥遥领先:)):

"hits": {
      "total": 3,
      "max_score": 0.4339554,
      "hits": [
         {
            "_index": "attribute-tree",
            "_type": "attribute",
            "_id": "2360",
            "_score": 0.4339554,
            "_source": {
               "name": "Bari",
               "suggest": {
                  "input": [
                     "Bari"
                  ],
                  "output": "Bari"
               }
            }
         },
         {
            "_index": "attribute-tree",
            "_type": "attribute",
            "_id": "1786",
            "_score": 0.04500804,
            "_source": {
               "name": "Baritone horn",
               "suggest": {
                  "input": [
                     "Baritone",
                     "horn"
                  ],
                  "output": "Baritone horn"
               }
            }
         },
         {
            "_index": "attribute-tree",
            "_type": "attribute",
            "_id": "2313",
            "_score": 0.04500804,
            "_source": {
               "name": "Baritone ukulele",
               "suggest": {
                  "input": [
                     "Baritone",
                     "ukulele"
                  ],
                  "output": "Baritone ukulele"
               }
            }
         }
      ]