Twitter的用户推荐算法

关于Twitter的用户推荐算法，Quora上的文章有一个说明。算法基本分4步：

First and foremost, we looked at who your friends follow, who they talk to, who they RT as gauges of your interest.
Then we applied either positive/negative biases towards individual factors like location, popularity, tweet volume, ratio, conversational likelihood, etc
We then further personalized based on who you tended to follow or skip. For example, if you follow a lot of Mark Suster's friends, we will boost up your "affinity score" to him
There is also the concept of exploration, which meant that we will throw in random people just for the heck of it

1. 遍历你关注的用户，找出他们关注、回复、转发的所有用户，作为候选；

2. 根据位置、热度、微博数目、微博内容相似度等为每一个候选计算一个权值；

3. 再根据你倾向于关注或无视哪些人的好友，对权值进行调整。例如你关注了很多张三的朋友，那么张三的朋友的权值可以调高ÿ