Elasticsearch单向同义词分析器不符合预期行为问题求助

期望实现的行为：

• 查询"cable"时，返回包含"cable"和"cordon"的结果
• 查询"cordon"时，仅返回包含"cordon"的结果

当前问题：执行以下两个查询时，均返回包含"cable"和"cordon"的所有结果。

当前查询语句 #

Query #1（查询"cable"） #

{
    "query": {
        "bool": {
            "must": [
                {
                    "bool": {
                        "should": [
                            {
                                "simple_query_string": {
                                    "query": "cable",
                                    "analyzer": "french",
                                    "fields": [
                                        "label"
                                    ]
                                }
                            }
                        ]
                    }
                }
            ]
        }
    },
    "sort": {
        "_score": {
            "order": "desc"
        }
    }
}

Query #2（查询"cordon"） #

{
    "query": {
        "bool": {
            "must": [
                {
                    "bool": {
                        "should": [
                            {
                                "simple_query_string": {
                                    "query": "cordon",
                                    "analyzer": "french",
                                    "fields": [
                                        "label"
                                    ]
                                }
                            }
                        ]
                    }
                }
            ]
        }
    },
    "sort": {
        "_score": {
            "order": "desc"
        }
    }
}

索引设置 #

{
    "lifecycle": {
        "name": "pdb-policy"
    },
    "routing": {
        "allocation": {
            "include": {
                "_tier_preference": "data_content"
            }
        }
    },
    "mapping": {
        "total_fields": {
            "limit": "10000"
        }
    },
    "number_of_shards": "1",
    "provided_name": "products2",
    "creation_date": "1719059028223",
    "analysis": {
        "filter": {
            "alphanumeric": {
                "pattern": "[^a-zA-Z0-9]",
                "type": "pattern_replace",
                "replacement": ""
            },
            "french_stop": {
                "type": "stop",
                "stopwords": "_french_"
            },
            "limit1term": {
                "type": "limit",
                "max_token_count": "1"
            },
            "synonym_monodirectional": {
                "type": "synonym",
                "synonyms_path": "synonyms\/words.txt",
                "expand": "false"
            },
            "french_elision": {
                "type": "elision",
                "articles": [
                    "l",
                    "m",
                    "t",
                    "qu",
                    "n",
                    "s",
                    "j",
                    "d",
                    "c",
                    "jusqu",
                    "quoiqu",
                    "lorsqu",
                    "puisqu"
                ]
            },
            "limit3term": {
                "type": "limit",
                "max_token_count": "3"
            },
            "french_stemmer": {
                "name": "french",
                "type": "stemmer"
            }
        },
        "normalizer": {
            "transliterator": {
                "filter": [
                    "asciifolding",
                    "lowercase"
                ]
            }
        },
        "analyzer": {
            "space_addition": {
                "filter": [
                    "asciifolding",
                    "french_elision",
                    "lowercase",
                    "synonym_monodirectional",
                    "french_stop"
                ],
                "tokenizer": "standard"
            },
            "prefix1term": {
                "filter": [
                    "asciifolding",
                    "french_elision",
                    "lowercase",
                    "synonym_monodirectional",
                    "french_stop",
                    "limit1term"
                ],
                "tokenizer": "standard"
            },
            "space_removal": {
                "filter": [
                    "asciifolding",
                    "french_elision",
                    "lowercase",
                    "synonym_monodirectional",
                    "french_stop"
                ],
                "tokenizer": "standard"
            },
            "french": {
                "filter": [
                    "synonym_monodirectional"
                ],
                "tokenizer": "standard"
            },
            "prefix3term": {
                "filter": [
                    "asciifolding",
                    "french_elision",
                    "lowercase",
                    "synonym_monodirectional",
                    "french_stop",
                    "limit3term"
                ],
                "tokenizer": "standard"
            },
            "transliterator": {
                "filter": [
                    "asciifolding",
                    "lowercase"
                ],
                "tokenizer": "keyword"
            }
        }
    },
    "priority": "100",
    "number_of_replicas": "0",
    "uuid": "ePDgepPcTm-NLcaePn6f3Q",
    "version": {
        "created": "7170999"
    }
}

同义词文件内容 #

cable => cordon

分析器测试结果 #

{
    "tokens": [
        {
            "token": "cordon",
            "start_offset": 0,
            "end_offset": 6,
            "type": "<ALPHANUM>",
            "position": 0
        }
    ]
}

{
    "tokens": [
        {
            "token": "cordon",
            "start_offset": 0,
            "end_offset": 5,
            "type": "SYNONYM",
            "position": 0
        }
    ]
}

解决思路 #

问题核心在于索引与查询阶段的分析器逻辑不匹配：当前的配置会在索引时将cable转换为cordon存储，导致查询cordon时会命中所有原cable的文档。要实现单向同义词匹配，需按以下步骤调整：

1. 修改同义词规则 #

将同义词文件内容改为：

cable => cable, cordon

此规则表示查询cable时，会同时生成cable和cordon两个检索词；而查询cordon时不会触发同义词转换，仅保留原词。

2. 分离索引与查询分析器 #

确保字段label在索引时使用不含同义词过滤器的分析器，仅在查询时应用同义词规则：

• 新增标准法语分析器（用于索引）：

  "standard_french": {
      "filter": [
          "french_elision",
          "lowercase",
          "french_stop",
          "french_stemmer"
      ],
      "tokenizer": "standard"
  }
  

• 调整label字段映射：

  "label": {
      "type": "text",
      "analyzer": "standard_french", // 索引时保留原词
      "search_analyzer": "french" // 查询时应用同义词
  }
  

3. 重新索引文档 #

修改索引设置和映射后，需要重新导入所有文档，确保新的分析规则生效。

验证逻辑 #

• 索引阶段：文档中的cable会被保留为cable（经过法语标准化处理），cordon保留为cordon；
• 查询cable：检索词会被转换为cable+cordon，匹配所有包含这两个词的文档；
• 查询cordon：检索词仅为cordon，仅匹配包含该词的文档。

内容的提问来源于stack exchange，提问作者JohnsM