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/article/_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} }
将对帖子搜索得到的分数,跟follower_num进行运算,由follower_num在一定成都上增强帖子的分数
看帖子的人越多,那么帖子的分数就越高
GET /forum/article/_search
{
"query": {
"function_score": {
"query": {
"multi_match": {
"query": "java spark",
"fields": ["title","content"]
}
},
"field_value_factor": {
"field": "follower_num",
"modifier": "log1p",
"factor": 0.1
},
"boost_mode": "sum",
"max_boost": 1.5
}
}
}
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的follower是0,那么分数就变成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+num_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
