elasticsearch - 优先考虑某些字段的ES搜索结果

Question

我正在使用elasticsearch-6.4.3。我创建了一个索引flight-location_methods

      settings index: {
          analysis: {
              "filter": {
                  "autocomplete_filter": {
                      "type": "edge_ngram",
                      "min_gram": 1,
                      "max_gram": 20
                  }
              },
              "analyzer": {
                  "autocomplete": {
                      "type": "custom",
                      "tokenizer": "standard",
                      "filter": ["lowercase",  "autocomplete_filter"]
                  }
              }
          }
      }

      mapping do
        indexes :airport_code, type: "text", analyzer: "autocomplete", search_analyzer: "standard"
        indexes :airport_name, type: "text", analyzer: "autocomplete", search_analyzer: "standard"
        indexes :city_name, type: "text", analyzer: "autocomplete", search_analyzer: "standard"
        indexes :country_name, type: "text", analyzer: "autocomplete", search_analyzer: "standard"
      end

上面的代码片段来自我为索引创建的 represents the mapping的Ruby代码。

当我执行此查询时:

GET /flight-location_methods/_search
{
  "from": 0,
  "size": 1000,
  "query": {
    "function_score": {
      "functions": [
        {
          "filter": {
            "match": {
              "city_name": "new yo"
            }
          },
          "weight": 50
        },
        {
          "filter": {
            "match": {
              "country_name": "new yo"
            }
          },
          "weight": 50
        }
      ],
      "max_boost": 200,
      "score_mode": "max",
      "boost_mode": "multiply",
      "min_score": 10
    }
  }
}

我得到这个结果:

  {
    "_index": "flight-location_methods",
    "_type": "_doc",
    "_id": "tcoj1G0Bdo5Q9AduxCKi",
    "_score": 50,
    "_source": {
      "airport_name": "Ouvea",
      "airport_code": "UVE",
      "city_name": "Ouvea",
      "country_name": "New Caledonia"
    }
  },
  {
    "_index": "flight-location_methods",
    "_type": "_doc",
    "_id": "zMoj1G0Bdo5Q9AduxCKi",
    "_score": 50,
    "_source": {
      "airport_name": "Palmerston North",
      "airport_code": "PMR",
      "city_name": "Palmerston North",
      "country_name": "New Zealand"
    }
  },
  {
    "_index": "flight-location_methods",
    "_type": "_doc",
    "_id": "1Moj1G0Bdo5Q9AduxCKi",
    "_score": 50,
    "_source": {
      "airport_name": "Westport",
      "airport_code": "WSZ",
      "city_name": "Westport",
      "country_name": "New Zealand"
    }
  },
  {
    "_index": "flight-location_methods",
    "_type": "_doc",
    "_id": "1coj1G0Bdo5Q9AduxCKi",
    "_score": 50,
    "_source": {
      "airport_name": "Whangarei",
      "airport_code": "WRE",
      "city_name": "Whangarei",
      "country_name": "New Zealand"
    }
  },
  {
    "_index": "flight-location_methods",
    "_type": "_doc",
    "_id": "Rsoj1G0Bdo5Q9AduxCOi",
    "_score": 50,
    "_source": {
      "airport_name": "Municipal",
      "airport_code": "RNH",
      "city_name": "New Richmond",
      "country_name": "United States"
    }
  },
  {
    "_index": "flight-location_methods",
    "_type": "_doc",
    "_id": "fsoj1G0Bdo5Q9AduxCOi",
    "_score": 50,
    "_source": {
      "airport_name": "New London",
      "airport_code": "GON",
      "city_name": "New London",
      "country_name": "United States"
    }
  },
  {
    "_index": "flight-location_methods",
    "_type": "_doc",
    "_id": "gMoj1G0Bdo5Q9AduxCOi",
    "_score": 50,
    "_source": {
      "airport_name": "New Ulm",
      "airport_code": "ULM",
      "city_name": "New Ulm",
      "country_name": "United States"
    }
  },
  {
    "_index": "flight-location_methods",
    "_type": "_doc",
    "_id": "5coj1G0Bdo5Q9AduxCSi",
    "_score": 50,
    "_source": {
      "airport_name": "Cape Newenham",
      "airport_code": "EHM",
      "city_name": "Cape Newenham",
      "country_name": "United States"
    }
  },
  {
    "_index": "flight-location_methods",
    "_type": "_doc",
    "_id": "Ycoj1G0Bdo5Q9AduxCWi",
    "_score": 50,
    "_source": {
      "airport_name": "East 60th Street H/P",
      "airport_code": "JRE",
      "city_name": "New York",
      "country_name": "United States"
    }
  }

如您所见， New York should be on top但实际上不是。

另外，我使用 can not use AND operator是因为如果搜索文本包含多个单词，我希望搜索文本中的任何单词出现在任何字段中。但是，如果所有搜索文本都在一个字段中，则优先级应该更高。

Answer 1

让我们首先讨论elasticsearch标记化程序和标记化过程:

A tokenizer receives a stream of characters, breaks it up into individual tokens (usually individual words). ES docs

现在，让我们描述 自动完成分析器的工作方式:

提供了标准 token 生成器 token 作为标准elasticsearch token 生成器(为简化起见，我们说这是单词)

小写过滤器使所有字符变小。

然后edge_ngram过滤器将每个单词分解为 token 。

从这里开始魔术:我认为您对1到20的 token 的定义太多了。可能存在包含10个以上字符的单词，但对于我们而言，这是不相关的。同样，仅包含一个对我们不可用的字符的 token 。我改变它:

   "filter": {
        "autocomplete_filter": {
          "type": "edge_ngram",
          "min_gram": 2,
          "max_gram": 5
        }
      }

然后在我们的索引中将有很多单词部分，长度从2到5个字符。现在，当我们知道要搜索的内容时，就可以创建映射并编写查询:

{
  "settings": {
    "number_of_shards": 3,
    "number_of_replicas": 0,
    "analysis": {
      "filter": {
        "autocomplete_filter": {
          "type": "edge_ngram",
          "min_gram": 2,
          "max_gram": 5
        }
      },
      "analyzer": {
        "autocomplete": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": [
            "lowercase",
            "autocomplete_filter"
          ]
        }
      }
    }
  },
  "mappings": {
    "_doc": {
      "properties": {
        "airport_name": {
          "type": "text",
          "fields": {
            "ngram": {
              "type": "text",
              "analyzer": "autocomplete"
            }
          }
        },
        "airport_code": {
          "type": "keyword",
          "fields": {
            "ngram": {
              "type": "text",
              "analyzer": "autocomplete"
            }
          }
        },
        "city_name": {
          "type": "keyword",
          "fields": {
            "ngram": {
              "type": "text",
              "analyzer": "autocomplete"
            }
          }
        },
        "country_name": {
          "type": "keyword",
          "fields": {
            "ngram": {
              "type": "text",
              "analyzer": "autocomplete"
            }
          }
        }
      }
    }
  }
}

我使用ngram字段和常规字段来制作字段，以保持进行聚合的能力。例如，通过多个机场查找城市是很好的。

现在我们可以运行一个简单的查询来获取纽约:

{
   "size": 20, 
   "query": {
     "query_string": {
       "default_field": "city_name.ngram",
       "query": "new yo",
       "default_operator": "AND"
     }
   }
}

Answer
{
  "took": 15,
  "timed_out": false,
  "_shards": {
    "total": 3,
    "successful": 3,
    "skipped": 0,
    "failed": 0
  },
  "hits": {
    "total": 1,
    "max_score": 13.896059,
    "hits": [
      {
        "_index": "test-index",
        "_type": "_doc",
        "_id": "BtBD2W0BCDulLSY6pKM8",
        "_score": 13.896059,
        "_source": {
          "airport_name": "Flushing",
          "airport_code": "FLU",
          "city_name": "New York",
          "country_name": "United States"
        }
      }
    ]
  }
}

或者使用boosting创建 boosting或 text查询。在大数据列表上进行查询时，这也将更加有效。

您的查询应如下所示:

{
   "query": {
     "function_score": {
       "query": {
         "query_string": {
           "query": "new yo",
           "analyzer": "autocomplete"
         }
       },
       "functions": [
         {
           "filter": {"terms": {
             "city_name.ngram": [
               "new",
               "yo"
             ]
           }},
           "weight": 2
         },
         {
           "filter": {"terms": {
             "country_name.ngram": [
               "new",
               "yo"
             ]
           }},
           "weight": 2
         }
       ],
       "max_boost": 30,
       "min_score": 5, 
       "score_mode": "max",
       "boost_mode": "multiply"
     }
   }
}

在此查询中，纽约将是第一个，因为我们通过查询部分过滤了所有不相关的文档。并乘以2 city_name.ngram字段分数，在此字段中，我们有2个 token ，那么此字段将获得最高分数。同样，查询的底线是min_score，它过滤而不是相关文档。您可以阅读有关当前的Elasticsearch相关算法 here的信息。
顺便说一句，我不想​​将过滤器放在权重相同的函数中。您应该决定是否是更重要的 Realm 。这使您的搜索更加清晰。