Elasticsearch Suggest自动补全返回非预期结果的原因及解决方法

问题重现 #

1. 创建索引Mapping #

PUT /movies
{
  "settings": {
    "analysis": {
      "filter": {
        "true_false_filter": {
          "type": "keep",
          "keep_words": [
            "true",
            "false"
          ]
        },
        "french_elision": {
          "type": "elision",
          "articles_case": false,
          "articles": [
            "puisqu"
          ]
        },
        "french_stemmer": {
          "type": "stemmer",
          "language": "light_french"
        },
        "organic-dictionary": {
          "type": "synonym",
          "expand": true,
          "lenient": true,
          "synonyms": [
            "non bio"
          ]
        },
        "french_stop_filter": {
          "type": "stop",
          "ignore_case": true,
          "stopwords": "_french_"
        }
      },
      "analyzer": {
        "lowercase_stop_analyzer": {
          "tokenizer": "lowercase",
          "filter": [
            "french_stop_filter"
          ]
        },
        "lowercase_asciifolding": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": [
            "asciifolding",
            "lowercase"
          ]
        },
        "french_analyzer_custom": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": [
            "asciifolding",
            "lowercase",
            "french_elision",
            "french_stemmer"
          ]
        },
        "custom_organic_analyzer": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": [
            "asciifolding",
            "lowercase",
            "french_elision",
            "organic-dictionary",
            "true_false_filter",
            "unique"
          ]
        }
      }
    },
    "mappings": {
      "properties": {
        "attr": {
          "type": "text",
          "analyzer": "french_analyzer_custom"
        },
        "brand_name": {
          "type": "keyword"
        },
        "brand_name_suggest": {
          "type": "completion",
          "analyzer": "lowercase_stop_analyzer",
          "search_analyzer": "lowercase_asciifolding",
          "preserve_separators": false,
          "preserve_position_increments": false,
          "max_input_length": 50
        }
      }
    }
}

2. 插入测试文档 #

POST /movies/_doc/1001
{
    "brand_name": "A LE MOUTON HUILE D'OLIVE",
    "brand_name_suggest": [
      "A LE MOUTON HUILE D'OLIVE"
    ]
}

3. 执行搜索请求 #

GET movies/_search
{
  "explain": true, 
  "suggest": {
    "completer": {
      "text": "amo",
      "completion": {
        "field": "brand_name_suggest",
        "size": 20,
        "skip_duplicates": true
      }
    }
  }
}

问题原因 #

当前配置下，brand_name_suggest字段的索引分析器lowercase_stop_analyzer会执行以下操作：

1. 通过lowercase分词器将输入文本转小写并分割为token：["a", "le", "mouton", "huile", "d", "olive"]
2. 经french_stop_filter过滤掉法语停用词"a"和"le"，剩余token为["mouton", "huile", "d", "olive"]
3. 结合preserve_separators: false的设置，Elasticsearch会将剩余token拼接为连续字符串（如"moutonhuiledolive"）

搜索时，lowercase_asciifolding分析器将查询词"amo"转为"amo"，由于拼接后的长字符串包含"amo"子串，再加上preserve_position_increments: false导致位置信息丢失，最终触发了非预期的匹配（尽管Completion Suggester默认是前缀匹配，但该配置组合导致了子串匹配的效果）。

解决方案 #

方案1：调整分析器与Completion字段设置 #

修改brand_name_suggest的索引分析器和字段参数，确保仅进行前缀匹配：

PUT /movies_new
{
  "settings": {
    "analysis": {
      "filter": {
        "true_false_filter": {
          "type": "keep",
          "keep_words": [
            "true",
            "false"
          ]
        },
        "french_elision": {
          "type": "elision",
          "articles_case": false,
          "articles": [
            "puisqu"
          ]
        },
        "french_stemmer": {
          "type": "stemmer",
          "language": "light_french"
        },
        "organic-dictionary": {
          "type": "synonym",
          "expand": true,
          "lenient": true,
          "synonyms": [
            "non bio"
          ]
        },
        "french_stop_filter": {
          "type": "stop",
          "ignore_case": true,
          "stopwords": "_french_"
        }
      },
      "analyzer": {
        // 重写分析器，使用standard分词器保证正确分割，再执行小写+停用词过滤
        "lowercase_stop_analyzer": {
          "tokenizer": "standard",
          "filter": [
            "lowercase",
            "french_stop_filter"
          ]
        },
        "lowercase_asciifolding": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": [
            "asciifolding",
            "lowercase"
          ]
        },
        "french_analyzer_custom": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": [
            "asciifolding",
            "lowercase",
            "french_elision",
            "french_stemmer"
          ]
        },
        "custom_organic_analyzer": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": [
            "asciifolding",
            "lowercase",
            "french_elision",
            "organic-dictionary",
            "true_false_filter",
            "unique"
          ]
        }
      }
    },
    "mappings": {
      "properties": {
        "attr": {
          "type": "text",
          "analyzer": "french_analyzer_custom"
        },
        "brand_name": {
          "type": "keyword"
        },
        "brand_name_suggest": {
          "type": "completion",
          "analyzer": "lowercase_stop_analyzer",
          "search_analyzer": "lowercase_asciifolding",
          "preserve_separators": true, // 保留分隔符，避免token无意义拼接
          "preserve_position_increments": true, // 保留位置增量，保证序列匹配逻辑
          "max_input_length": 50
        }
      }
    }
}

方案2：限制前缀匹配长度 #

如果不想重建索引，可以在搜索时添加prefix_length参数，要求查询词达到指定长度才触发匹配（比如设置为3，避免短词误匹配）：

GET movies/_search
{
  "suggest": {
    "completer": {
      "text": "amo",
      "completion": {
        "field": "brand_name_suggest",
        "size": 20,
        "skip_duplicates": true,
        "prefix_length": 3 // 仅当查询词长度≥3时才执行匹配，可根据需求调整
      }
    }
  }
}

注意事项 #

• 修改Mapping后需要重建索引，并重新导入数据；
• 可通过_analyze接口验证分析器效果，确保token生成符合预期：

POST _analyze
{
  "analyzer": "lowercase_stop_analyzer",
  "text": "A LE MOUTON HUILE D'OLIVE"
}

内容的提问来源于stack exchange，提问作者user1361815