Elasticsearch地址字段模糊分析器搜索排序异常问题

我为address字段配置了模糊分析器，但搜索"13 Madison Ave"时，完全匹配的文档（_id:16）未排在首位，反而"6138 Madison Ave"（_id:5）位列第一，需解决该排序异常问题。

索引映射配置 #

PUT test_fuzzy
{
  "settings": {
    "index": {
      "max_ngram_diff": "40",
      "mapping": {
        "total_fields": {
          "limit": "2000"
        }
      },
      "number_of_shards": "3",
      "max_result_window": "15000",
      "analysis": {
        "filter": {
          "autocomplete": {
            "type": "ngram",
            "min_gram": "1",
            "max_gram": "40"
          }
        },
        "normalizer": {
          "lowercase_normalizer": {
            "filter": ["lowercase"]
          }
        },
        "analyzer": {
          "autocomplete": {
            "filter": ["lowercase", "autocomplete"],
            "tokenizer": "whitespace"
          },
          "autocomplete_search": {
            "filter": ["lowercase"],
            "tokenizer": "whitespace"
          }
        }
      },
      "number_of_replicas": "0"
    }
  },
  "mappings": {
    "dynamic": "true",
    "dynamic_templates": [
      {
        "named_analyzers": {
          "match": "address",
          "match_mapping_type": "string",
          "mapping": {
            "analyzer": "autocomplete",
            "fields": {
              "keyword": {
                "ignore_above": 256,
                "type": "keyword"
              }
            },
            "search_analyzer": "autocomplete_search",
            "type": "text"
          }
        }
      }
    ],
    "properties": {
      "address": {
        "type": "text",
        "analyzer": "autocomplete",
        "search_analyzer": "autocomplete_search",
        "fields": {
          "keyword": {
            "type": "keyword",
            "ignore_above": 256
          }
        }
      },
      "identifier": {
        "type": "text"
      }
    }
  }
}

测试数据 #

POST test_fuzzy/_bulk
{ "index": { "_id": "1" } }
{ "address": "1136 N Madison Ave", "identifier": 1 }
{ "index": { "_id": "2" } }
{ "address": "7135 Madison Ave W", "identifier": 2 }
{ "index": { "_id": "3" } }
{ "address": "1333 Madison Ave", "identifier": 3 }
{ "index": { "_id": "4" } }
{ "address": "1303 Madison Ave", "identifier": 4 }
{ "index": { "_id": "5" } }
{ "address": "6138 Madison Ave", "identifier": 5 }
{ "index": { "_id": "6" } }
{ "address": "1373 E Madison Ave", "identifier": 6 }
{ "index": { "_id": "7" } }
{ "address": "1333 E Madison Ave Ste 200", "identifier": 7 }
{ "index": { "_id": "8" } }
{ "address": "1311 Madison Ave", "identifier": 8 }
{ "index": { "_id": "9" } }
{ "address": "132 Madison Ave", "identifier": null }
{ "index": { "_id": "10" } }
{ "address": "1213 Madison Ave", "identifier": null }
{ "index": { "_id": "11" } }
{ "address": "413 Madison Ave", "identifier": null }
{ "index": { "_id": "12" } }
{ "address": "134 W Madison Ave", "identifier": null }
{ "index": { "_id": "13" } }
{ "address": "5138 Madison Ave", "identifier": null }
{ "index": { "_id": "14" } }
{ "address": "513 Madison Ave", "identifier": null }
{ "index": { "_id": "15" } }
{ "address": "1330 Madison Ave", "identifier": null }
{ "index": { "_id": "16" } }
{ "address": "13 Madison Ave", "identifier": null }
{ "index": { "_id": "17" } }
{ "address": "130 Madison Ave", "identifier": null }
{ "index": { "_id": "18" } }
{ "address": "131 W Madison Ave", "identifier": null }

当前使用的查询语句 #

GET test_fuzzy/_search
{
  "query": {
    "bool": {
      "must": [
        {
          "bool": {
            "must": [
              {
                "bool": {
                  "should": [
                    {
                      "match": {
                        "address": {
                          "query": "13 Madison Ave",
                          "operator": "OR",
                          "prefix_length": 0,
                          "max_expansions": 50,
                          "fuzzy_transpositions": true,
                          "lenient": false,
                          "zero_terms_query": "NONE",
                          "auto_generate_synonyms_phrase_query": true,
                          "boost": 1
                        }
                      }
                    }
                  ],
                  "adjust_pure_negative": true,
                  "boost": 1
                }
              }
            ],
            "adjust_pure_negative": true,
            "boost": 1
          }
        }
      ],
      "adjust_pure_negative": true,
      "boost": 1
    }
  }
}

问题原因 #

1. ngram分词导致词频虚高：当前autocomplete分析器的min_gram=1，会生成大量短片段token，比如"6138"会被拆出"13"这样的片段。这些额外的匹配片段让"6138 Madison Ave"的匹配词频超过了完全匹配的文档，导致评分更高。
2. 查询结构冗余且未加权精确匹配：嵌套多层bool结构无实际意义，也没有针对完全匹配的文档设置更高权重，无法引导排序。

解决方案 #

方案1：优化查询，为精确匹配加权 #

通过should子句同时匹配text字段和keyword字段，给精确匹配赋予更高权重，确保完全匹配的文档评分优先：

GET test_fuzzy/_search
{
  "query": {
    "bool": {
      "should": [
        // 原模糊匹配逻辑
        {
          "match": {
            "address": {
              "query": "13 Madison Ave",
              "operator": "OR",
              "prefix_length": 0,
              "max_expansions": 50,
              "boost": 1
            }
          }
        },
        // 精确匹配keyword字段，赋予10倍权重
        {
          "match": {
            "address.keyword": {
              "query": "13 Madison Ave",
              "boost": 10
            }
          }
        }
      ]
    }
  }
}

方案2：调整ngram配置，减少无效匹配 #

若无需单字符匹配，可将min_gram调整为2或3，减少短片段的误匹配：
修改索引settings中的autocomplete过滤器：

"filter": {
  "autocomplete": {
    "type": "ngram",
    "min_gram": "2",
    "max_gram": "40"
  }
}

注意：修改分析器配置后需要重建索引并重新导入数据。

方案3：使用function_score自定义评分逻辑 #

通过function_score对完全匹配的文档额外加分，强制提升排序优先级：

GET test_fuzzy/_search
{
  "query": {
    "function_score": {
      "query": {
        "match": {
          "address": "13 Madison Ave"
        }
      },
      "functions": [
        {
          "filter": {
            "term": {
              "address.keyword": "13 Madison Ave"
            }
          },
          "weight": 10
        }
      ],
      "boost_mode": "sum"
    }
  }
}

内容的提问来源于stack exchange，提问作者amrit