Elasticsearch 通过function score优化算分排序

布满荆棘的人生 2023-01-01 08:50 94阅读 0赞

Elasticsearch 采用默认的相关性算法进行算分排序，我们可以通过function score函数来影响算分排序，比如可以采用一个字段的值与老的算分进行运算生成新的算分，或者采用一致性随机函数，生成随机算分，相同的seed值算分排序一致。

初始化数据
    DELETE blogs
    PUT /blogs/_doc/1
    {
      "title":   "About popularity",
      "content": "In this post we will talk about...",
      "votes":   0
    }
    
    PUT /blogs/_doc/2
    {
      "title":   "About popularity",
      "content": "In this post we will talk about...",
      "votes":   100
    }
    
    PUT /blogs/_doc/3
    {
      "title":   "About popularity",
      "content": "In this post we will talk about...",
      "votes":   1000000
    }
    
    采用function score query中 的field_value_factor字段的值 与 原来算法相乘，作为最终算分
    post blogs/_search
    {
      "query":{
        "function_score": {
          "query": {
            "multi_match": {
              "query": "popularity",
              "fields": ["title","content"]
            }
          },
          "field_value_factor": {
            "field": "votes"
          }
        }
      }
    }
    
    function score query 新的算法=老的算分*log(1+投票数)
    POST blogs/_search
    {
      "query":{
        "function_score": {
          "query": {
            "multi_match":{
              "query":"popularity",
              "fields":["title","content"]
            }
          },
          "field_value_factor": {
            "field": "votes",
            "modifier": "log1p"
          }
        }
      }
    }
    
    function score query 新的算法=老的算分*log(1+factor*投票数)
    post blogs/_search
    {
      "query":{
        "function_score": {
          "query": {
            "multi_match": {
              "query": "popularity",
              "fields": ["title","content"]
            }
          },
          "field_value_factor":{
            "field": "votes",
            "modifier": "log1p",
            "factor": 0.1
          }
        }
      }
    }
    
    function score query默认新分=老的算分乘以一个函数，修改为加， 新的分=老的算分+log(1+factor*投票数);max_boost则规定加上的函数的最大值是3
    post blogs/_search
    {
      "query":{
        "function_score": {
          "query": {
            "multi_match": {
              "query": "popularity",
              "fields": ["title","content"]
            }
          },
          "field_value_factor":{
            "field": "votes",
            "modifier": "log1p",
            "factor": 0.1
          },
          "boost_mode": "sum",
          "max_boost": 3
        }
      }
    }
    
    随机生成一个算分排序，seed值相同则排序相同
    post blogs/_search
    {
      "query":{
        "function_score": {
          "random_score": {
            "seed": 111111
          }
        }
      }
    }