ES搜索结果的排序算法选择和优化,涵盖从BM25算法到Function Score的全链路知识

# TF-IDF计算公式（示意逻辑）
score = tf * idf * fieldLengthNorm
# tf：词项频率，即搜索词在文档中出现的次数
# idf：逆文档频率，即搜索词在所有文档中的稀有程度
# fieldLengthNorm：字段长度归一化因子

// ES中的BM25参数配置示例
{
  "settings": {
    "index": {
      "similarity": {
        "custom_bm25": {
          "type": "BM25",
          "k1": 1.2,    // 控制词频饱和度
          "b": 0.75     // 控制字段长度影响
        }
      }
    }
  }
}

GET /products/_search
{
  "explain": true,
  "query": {
    "match": {
      "description": "无线蓝牙耳机"
    }
  }
}

// 按价格升序+评分降序的复合排序
{
  "sort": [
    { "price": "asc" },
    { "_score": "desc" }
  ]
}

// 按距离排序（5km内优先）
{
  "sort": [
    {
      "_geo_distance": {
        "location": [116.4074, 39.9042], // 北京天安门坐标
        "order": "asc",
        "unit": "km",
        "mode": "min"
      }
    }
  ]
}

// 结合库存和评分的自定义公式
{
  "sort": {
    "_script": {
      "type": "number",
      "script": {
        "source": """
          double stockWeight = doc['stock'].value * 0.3;
          double scoreWeight = _score * 0.7;
          return stockWeight + scoreWeight;
        """
      },
      "order": "desc"
    }
  }
}

// 提升新品权重
{
  "query": {
    "function_score": {
      "query": {"match_all": {}},
      "functions": [
        {
          "filter": {"range": {"publish_date": {"gte": "now-7d/d"}}},
          "weight": 3
        }
      ],
      "boost_mode": "multiply"
    }
  }
}

// 时间衰减+地理位置衰减
{
  "functions": [
    {
      "gauss": {
        "publish_date": {
          "origin": "now",
          "scale": "30d",
          "offset": "7d",
          "decay": 0.5
        }
      }
    },
    {
      "gauss": {
        "location": {
          "origin": "116.4074, 39.9042",
          "scale": "100km"
        }
      }
    }
  ]
}

// 预计算字段映射
{
  "mappings": {
    "properties": {
      "hot_score": {
        "type": "scaled_float",
        "scaling_factor": 1000
      }
    }
  }
}

// 搜索模板缓存
PUT /_scripts/hot_ranking
{
  "script": {
    "lang": "mustache",
    "source": {
      "sort": [
        {"hot_score": "desc"},
        {"_score": "desc"}
      ]
    }
  }
}

{
  "functions": [
    {
      "field_value_factor": {
        "field": "sales_volume",
        "factor": 1.2,
        "modifier": "log1p"
      }
    },
    {
      "filter": {"term": {"is_promotion": true}},
      "weight": 2
    }
  ]
}

// 时间衰减+用户偏好加权
{
  "functions": [
    {
      "exp": {
        "create_time": {
          "origin": "now",
          "scale": "7d"
        }
      }
    },
    {
      "filter": {"terms": {"category": ["科技","数码"]}},
      "weight": 1.5
    }
  ]
}