Elasticsearch中能否自定义父文档与嵌套文档的_score？

这个需求完全可行，Elasticsearch的function_score查询结合nested类型特性，能精准控制父文档和嵌套文档的评分，完全匹配你的场景要求。

1. 准备测试数据结构 #

首先确保嵌套字段用nested类型定义mapping：

PUT /cars_index
{
  "mappings": {
    "properties": {
      "title": {"type": "text"},
      "colors": {
        "type": "nested",
        "properties": {
          "color": {"type": "text"}
        }
      }
    }
  }
}

插入测试文档：

POST /cars_index/_doc
{
  "title": "car",
  "colors": [{"color": "red"}, {"color": "blue"}]
}

2. 场景一：搜索"red car"的评分控制 #

通过function_score为父文档固定评分0.5，同时对嵌套文档设置差异化评分：

GET /cars_index/_search
{
  "query": {
    "function_score": {
      "query": {"match": {"title": "car"}},
      "functions": [{"filter": {"match": {"title": "car"}}, "weight": 0.5}],
      "score_mode": "sum",
      "nested": {
        "path": "colors",
        "query": {
          "function_score": {
            "functions": [
              {"filter": {"match": {"colors.color": "red"}}, "weight": 1},
              {"filter": {"match": {"colors.color": "blue"}}, "weight": 0}
            ],
            "score_mode": "sum"
          }
        },
        "inner_hits": {}
      }
    }
  }
}

查询结果：

• 父文档_score为0.5
• 嵌套文档"red"的_score为1，"blue"的_score为0

3. 场景二：搜索"black car"的评分控制 #

调整嵌套文档的评分规则，统一设为0或0.5：

GET /cars_index/_search
{
  "query": {
    "function_score": {
      "query": {"match": {"title": "car"}},
      "functions": [{"filter": {"match": {"title": "car"}}, "weight": 0.5}],
      "score_mode": "sum",
      "nested": {
        "path": "colors",
        "query": {
          "function_score": {
            "functions": [
              {"filter": {"match_all": {}}, "weight": 0}
            ],
            "score_mode": "sum"
          }
        },
        "inner_hits": {}
      }
    }
  }
}

查询结果：

• 父文档_score为0.5
• 所有嵌套文档_score为0（若需要0.5，修改weight值即可）

4. 评分结果的业务使用 #

查询返回结果中：

• 父文档评分在顶层_score字段
• 每个嵌套文档的评分在inner_hits.colors.hits.hits数组对应元素的_score字段

你可直接读取这些评分值，判断返回父文档页面链接，还是对应嵌套文档的链接（比如评分1返回red的链接，0返回blue的链接，父文档0.5返回car的链接）。

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