27、深度探秘搜索技术_实战用function_score自定义相关度分数算法

我们可以做到自定义一个function_score函数，自己将某个field的值，跟es内置算出来的分数进行运算，然后由自己指定的field来进行分数的增强
给所有的帖子数据增加follower数量

POST /forum/_bulk
{ "update": { "_id": "1"} }
{ "doc" : {"follower_num" : 5} }
{ "update": { "_id": "2"} }
{ "doc" : {"follower_num" : 10} }
{ "update": { "_id": "3"} }
{ "doc" : {"follower_num" : 25} }
{ "update": { "_id": "4"} }
{ "doc" : {"follower_num" : 3} }
{ "update": { "_id": "5"} }
{ "doc" : {"follower_num" : 60} }
{ "update": { "_id": "6"} }
{ "doc" : {"follower_num" : 50} }
{ "update": { "_id": "7"} }
{ "doc" : {"follower_num" : 20} }
{ "update": { "_id": "8"} }
{ "doc" : {"follower_num" : 32} }
{ "update": { "_id": "9"} }
{ "doc" : {"follower_num" : 30} }
{ "update": { "_id": "10"} }
{ "doc" : {"follower_num" : 70} }
{ "update": { "_id": "11"} }
{ "doc" : {"follower_num" : 80} }

将对帖子搜索得到的分数，跟follower_num进行运算，由follower_num在一定程度上增强帖子的分数
看帖子的人越多，那么帖子的分数就越高

GET /forum/_search
{
  "query": {
    "function_score": {
      "query": {
        "multi_match": {
          "query": "java spark",
          "fields": ["tile", "content"]
        }
      },
      "field_value_factor": {
        "field": "follower_num",
        "modifier": "log1p",
        "factor": 0.5
      },
      "boost_mode": "sum",
      "max_boost": 2
    }
  }
}

如果只有field，那么会将每个doc的分数都乘以follower_num，如果有的doc follower是0，那么分数就会变为0，效果很不好。因此一般会加个log1p函数，公式会变为，new_score = old_score * log(1 + number_of_votes)，这样出来的分数会比较合理
再加个factor，可以进一步影响分数，new_score = old_score * log(1 + factor * number_of_votes)
boost_mode，可以决定分数与指定字段的值如何计算，multiply，sum，min，max，replace
max_boost，限制计算出来的分数不要超过max_boost指定的值