How Not To Sort By Average Ratin

2018-05-19 本文已影响0人空空格格

Score = (Positive ratings) − (Negative ratings)
- item 1: 1000 ratings (600 positive ratings, 400 negative ratings)
- item 2: 10000 ratings(5500 positive ratings, 4500 negative ratings)
  item1不应该放在item2前面
Score = Average rating = (Positive ratings) / (Total ratings)
- item 1: 1 positive , 0 negative
- item 2: 100 positive , 1 negative
  item 1 不应该放在item2前面

Score = Lower bound of Wilson score confidence interval for a Bernoulli parameter
评分给定下，95%的可能性positive rating的真实比例至少是多少

equation.png
(pos：积极评分的数目 n: 总评分的数 confidence:置信率)

r语言实现.png

sql语言实现.png
应用场景(不限于sorting)

检测垃圾邮件: What percentage of people who see this item will mark it as spam?
创造best of list: What percentage of people who see this item will mark it as “best of”?
most emailed list: What percentage of people who see this page will click “Email”?

rank score.png

随着时间的增长，得分变低。同时gravity增加的时候得分会减少地更快

python.png

对于old stories，时间影响变小（曲线平滑，主要依赖于vote）
对于new stories，是时间和vote的同时作用

score.png