ES 自定义评分实现结果集排序问题

需求

需求是根据短句查询后，不分词全包含（类似mysql的like）的语句要在最上边，其次是分词后的数据全包含的排在后边，然后是包含部分分词的数据，最后这三类数据要根据时间倒叙，自带的评分机制无法满足需求，所以使用function_score 自定义结果的评分：

创建索引

用到了ik分词器

PUT /robot1018
{
  "settings" : {
      
     "index": {
      "routing": {
        "allocation": {
          "include": {
            "_tier_preference": "data_content"
          }
        }
      },
      "number_of_shards": "3",    
      "max_result_window": "10000000",
      "analysis": {
        "filter": {
          "my_synonyms": {
            "type": "synonym",
            "synonyms_path": "synonyms/my_synonyms1.txt",
            "updateable": "true"
          },
          "my_stopwords": {
            "type": "stop",
            "updateable": "true",
            "stopwords_path": "stops/my_stopwords1.txt"
          },
          "my_pinyin": {
            "ignore_pinyin_offset": "true",
            "keep_joined_full_pinyin": "true",
            "lowercase": "true",
            "keep_original": "false",
            "keep_first_letter": "false",
            "keep_separate_first_letter": "false",
            "type": "pinyin",
            "keep_none_chinese": "false",
            "limit_first_letter_length": "50",
            "keep_full_pinyin": "true"
          }
        },
        "analyzer": {
          "pinyin_ik_max_word_analyzer": {
            "filter": [
              "my_pinyin"
            ],
            "type": "custom",
            "tokenizer": "ik_max_word"
          },
          "pinyin_ik_smart_analyzer": {
            "filter": [
              "my_pinyin"
            ],
            "type": "custom",
            "tokenizer": "ik_smart"
          },
          "synonym_stop_ik_smart_analyzer": {
            "filter": [
              "my_stopwords",
              "my_synonyms"
            ],
            "tokenizer": "ik_smart"
          }
        }
      },
      "number_of_replicas": "3" 
    }
   },
    
    "mappings": {
    
      "properties": {
        "create_time": {
          "type": "date",
          "format": "yyyy-MM-dd HH:mm:ss||yyyy-MM-dd||epoch_millis"
        },
        "guid": {
          "type": "keyword"
        },
        "id": {
          "type": "text",
          "fields": {
            "keyword": {
              "type": "keyword",
              "ignore_above": 256
            }
          }
        },
        "question": {
         "type": "text",
          "fields": {
            "pinyin": {
              "type": "text",
              "analyzer": "pinyin_ik_max_word_analyzer",
              "search_analyzer": "pinyin_ik_smart_analyzer"
            },
             "question_text": {
              "type": "keyword"            
            }
          },

          "analyzer": "ik_smart",
          "search_analyzer": "synonym_stop_ik_smart_analyzer"
        }
        }
      }    
  }
}

由于match_phrasa 查询会自动过滤标点符号，所以在question下加了个question_text字段 类型为keyword 不分词，用于通过 wildcard 过滤第一层的全匹配条件
例如：

  "match_phrase": {
              "question": {
                 "query": "供暖申请",
                 "slop": 0
              }
            }

查询的结果集中 会出现 “个人能否申请市政供暖？申请市政供暖条件及流程？“的结果，客户觉得这种不是全匹配的，所以我用wildcard 过滤全匹配的，也有说修改ik源码的，可自行尝试那种方法。

查询语句

GET robot1020/_search
{
  "from": 0,
  "size": 30,
  "query": {
    "function_score": {
      "query": {
        "bool": {
          "adjust_pure_negative": true,
          "must": [
            {
              "match": {
                "question": {
                  "query": "老小区后期申请市政供暖交换站选址安装有无相关要求？",
                  "minimum_should_match": "1",
                  "max_expansions": 50
                }
              }
            }
          ]

        }
      },
      "functions": [
        {
          "filter": {
            "wildcard": {
              "question.question_text": "*老小区后期申请市政供暖交换站选址安装有无相关要求？*"
            }
          },
          "weight": 2
        },
        {
          "filter": {
            "match_phrase": {
              "question": {
                "query": "老小区后期申请市政供暖交换站选址安装有无相关要求？",
                "slop": 0
              }
            }
          },
          "weight": 2
        },
        {
          "filter": {
            "match_phrase": {
              "question": {
                "query": "老小区后期申请市政供暖交换站选址安装有无相关要求？",
                "slop": 4
              }
            }
          },
          "weight": 2
        },
        {
          "filter": {
            "match_phrase": {
              "question": {
                "query": "老小区后期申请市政供暖交换站选址安装有无相关要求？",
                "slop": 10
              }
            }
          },
          "weight": 2
        },
        {
          "filter": {
            "match": {
              "question": {
                "query": "老小区后期申请市政供暖交换站选址安装有无相关要求？",
                "minimum_should_match": "2"
              }
            }
          },
          "weight": 2
        },
        {
          "filter": {
            "match": {
              "question": {
                "query": "老小区后期申请市政供暖交换站选址安装有无相关要求？"
              }
            }
          },
          "weight": 2
        }
      ],
      "boost_mode": "replace"
    }
  },
  "sort": [
    {
      "_score": {
        "order": "desc"
      }
    },
    {
      "create_time": {
        "order": "desc"
      }
    }
  ],
  "track_total_hits": 2147483647,
  "_source": [
    "question",
    "create_time"
  ]
}

每满足一个条件评分都会乘以2，所以全部满足filter条件的话评分是所有过滤方法里的weight 乘积，这样就实现了需求，也可以多些几个filter过滤，把结果集分的更细一些

结果满足预期，问题解决

另外 boost_mode : replace 指的是 用自己的函数的得分替换掉原先的评分，默认是 multiply （与原评分相乘）

其他相关的参数含义 都很容易搜到