在elasticsearch的较早位置为包含搜索查询的匹配分配更高的分数

Question

这个问题类似于我的另一个问题，在这里输入Val回答的链接描述。

我有一个包含3个文档的索引。

    {
            "firstname": "Anne",
            "lastname": "Borg",
        }

    {
            "firstname": "Leanne",
            "lastname": "Ray"
        },

    {
            "firstname": "Anne",
            "middlename": "M",
            "lastname": "Stone"
        }

当我搜索“Ann”时，我希望elastic返回所有这3个文档(因为它们在一定程度上都与术语“Ann”匹配)。但是，我希望符晓薇·雷具有较低的分数(相关性排名),因为搜索词“Ann”在本文档中出现的位置比该词在其他两个文档中出现的位置要晚。

这是我的索引设置…

{
    "settings": {
        "analysis": {
            "analyzer": {
                "my_analyzer": {
                    "filter": [
                        "lowercase"
                    ],
                    "type": "custom",
                    "tokenizer": "my_tokenizer"
                }
            },
            "tokenizer": {
                "my_tokenizer": {
                    "token_chars": [
                        "letter",
                        "digit",
                        "custom"
                    ],
                    "custom_token_chars": "'-",
                    "min_gram": "1",
                    "type": "ngram",
                    "max_gram": "2"
                }
            }
        }
    },
    "mappings": {
        "properties": {
            "firstname": {
                "type": "text",
                "fields": {
                    "keyword": {
                        "type": "keyword"
                    }
                },
                "copy_to": [
                    "full_name"
                ]
            },
            "lastname": {
                "type": "text",
                "fields": {
                    "keyword": {
                        "type": "keyword"
                    }
                },
                "copy_to": [
                    "full_name"
                ]
            },
            "middlename": {
                "type": "text",
                "fields": {
                    "keyword": {
                        "type": "keyword",
                        "ignore_above": 256
                    }
                },
                "copy_to": [
                    "full_name"
                ]
            },
            "full_name": {
                "type": "text",
                "analyzer": "my_analyzer",
                "fields": {
                    "keyword": {
                        "type": "keyword"
                    }
                }
            }
        }
    }
}

以下查询返回了预期的文档，但Leanne Ray的得分高于Anne Borg。

{
    "query": {
        "bool": {
            "must": {
                "query_string": {
                    "query": "Ann",
                    "fields": ["full_name"]
                }
            },
            "should": {
                "match": {
                    "full_name": "Ann"}
            }
        }
    }
}

以下是结果。。。

"hits": [
        {
            "_index": "contacts_4",
            "_type": "_doc",
            "_id": "2",
            "_score": 6.6333585,
            "_source": {
                "firstname": "Anne",
                "middlename": "M",
                "lastname": "Stone"
            }
        },
        {
            "_index": "contacts_4",
            "_type": "_doc",
            "_id": "1",
            "_score": 6.142234,
            "_source": {
                "firstname": "Leanne",
                "lastname": "Ray"
            }
        },
        {
            "_index": "contacts_4",
            "_type": "_doc",
            "_id": "3",
            "_score": 6.079495,
            "_source": {
                "firstname": "Anne",
                "lastname": "Borg"
            }
        }

同时使用 ngram 令牌筛选器和 ngram 令牌化器似乎可以解决此问题...

{
    "settings": {
        "analysis": {
            "analyzer": {
                "my_analyzer": {
                    "filter": [
                        "ngram"
                    ],
                    "tokenizer": "ngram"
                }
            }
        }
    },
    "mappings": {
        "properties": {
            "firstname": {
                "type": "text",
                "fields": {
                    "keyword": {
                        "type": "keyword"
                    }
                },
                "copy_to": [
                    "full_name"
                ]
            },
            "lastname": {
                "type": "text",
                "fields": {
                    "keyword": {
                        "type": "keyword"
                    }
                },
                "copy_to": [
                    "full_name"
                ]
            },
            "middlename": {
                "type": "text",
                "fields": {
                    "keyword": {
                        "type": "keyword"
                    }
                },
                "copy_to": [
                    "full_name"
                ]
            },
            "full_name": {
                "type": "text",
                "analyzer": "my_analyzer",
                "search_analyzer": "my_analyzer"
            }
        }
    }
}

相同的查询会返回带有所需相对评分的预期结果。为什么会这样？请注意，上面，我使用的是带有小写过滤器的ngram标记器，这里唯一的区别是我使用的是ngram过滤器而不是小写过滤器。

以下是结果。请注意，根据需要，莉安·雷的得分低于安妮·博格和安妮·斯通。

"hits": [
    {
        "_index": "contacts_4",
        "_type": "_doc",
        "_id": "3",
        "_score": 4.953257,
        "_source": {
            "firstname": "Anne",
            "lastname": "Borg"
        }
    },
    {
        "_index": "contacts_4",
        "_type": "_doc",
        "_id": "2",
        "_score": 4.87168,
        "_source": {
            "firstname": "Anne",
            "middlename": "M",
            "lastname": "Stone"
        }
    },
    {
        "_index": "contacts_4",
        "_type": "_doc",
        "_id": "1",
        "_score": 1.0364896,
        "_source": {
            "firstname": "Leanne",
            "lastname": "Ray"
        }
    }

顺便说一句，当索引还包含其他文档时，此查询还会返回大量误报结果。这并不是一个问题，因为这些假阳性相对于理想点击的得分来说得分很低。但仍然不理想。例如，如果我将{firstname:Gideon，lastname:Grossma}添加到文档中，上面的查询也会将该文档返回到结果集中，尽管其分数比包含字符串“Ann”的文档低得多

Answer 1

答案与链接线程中的答案相同。由于您正在对所有索引数据进行编码，因此它与Ann的工作方式与Anne的工作方式相同，但是，您将获得完全相同的响应（见下文），但分数不同：

"hits" : [
  {
    "_index" : "test",
    "_type" : "_doc",
    "_id" : "5Jr-DHIBhYuDqANwSeiw",
    "_score" : 4.8442974,
    "_source" : {
      "firstname" : "Anne",
      "lastname" : "Borg"
    }
  },
  {
    "_index" : "test",
    "_type" : "_doc",
    "_id" : "5pr-DHIBhYuDqANwSeiw",
    "_score" : 4.828779,
    "_source" : {
      "firstname" : "Anne",
      "middlename" : "M",
      "lastname" : "Stone"
    }
  },
  {
    "_index" : "test",
    "_type" : "_doc",
    "_id" : "5Zr-DHIBhYuDqANwSeiw",
    "_score" : 0.12874341,
    "_source" : {
      "firstname" : "Leanne",
      "lastname" : "Ray"
    }
  }
]

更新

下面是一个修改后的查询，您可以使用它来检查零件(即< code > ann vs < code > Anne )。同样，这里的大小写没有区别，因为分析器在索引之前会将所有内容都小写。

{
  "query": {
    "bool": {
      "must": {
        "query_string": {
          "query": "ann",
          "fields": [
            "full_name"
          ]
        }
      },
      "should": [
        {
          "match_phrase_prefix": {
            "firstname": {
              "query": "ann",
              "boost": "10"
            }
          }
        },
        {
          "match_phrase_prefix": {
            "lastname": {
              "query": "ann",
              "boost": "10"
            }
          }
        }
      ]
    }
  }
}