脸书onsite面经

上个礼拜去参加了onsite。。一个礼拜后收到了拒信。。。哭了。。但为了回馈大家还是奉上面经如下 （有些细节不太记得了）


1. product： 现在脸书想要有个新产品给用户做restaurant recommdation，类似friend recommendation。 如果你是产品负责

您好！
本帖隐藏的内容需要积分高于 188 才可浏览
您当前积分为 0。
使用VIP即刻解锁阅读权限或查看其他获取积分的方式

游客，您好！
本帖隐藏的内容需要积分高于 188 才可浏览
您当前积分为 0。
VIP即刻解锁阅读权限 或 查看其他获取积分的方式
Unlock interview details and practice with AI
Curated Interview Questions from Top Companies

ealth of group。。这也在地里提到好多次。。

我觉得大家把地里的题目都做一遍，都做明白了应该就没问题，毕竟我的面题里四道题里三道原题。。然后也没过，被自己蠢哭。

Suppose, from the above 3 points, we find this is a huge market and wan to launch a product. There are following product feature we need to design. The biggest challenge here is that we may answer something that are way too board(for example. Good UI, Good Algorithm). 此时，我们必须要清楚作为DS，我们可以怎样test，有哪些变量可以通过test掌控。
First of all, fix down the KPI. Through the new product, we want to:
1.        Get Facebook user to visit these restaurant or at least click on the restaurant page
2.        Increase engagement on FB
3.        Estimate potential effect on FB ads

To conduct test, we need to design and carefully consider the experiment:
1.        Who are we going to launch the test at.  I would highly recommend using the potential user analysis we have done before. We should test the region that have a lot of restaurant check-ins, posts and comments. Since restaurant is location sensitive, the test should be limited to just one or a few locations. We also need to estimate how is likely the test result happen in other regions that don’t have busy traffic and lots of restaurants.
2.         The user group we can test on. Since this is not a simple A/B test, we just cannot estimate a KPI that we can compare be two similar group together. However, we do want to see how does this new feature affect the engagement and ad revenue on facebook. (The interviewer probably will have a lot of questions about if the engagement/ad revenue is changed , what are you going to do). Besides that, we need to estimate: How many users is going to use the new features? What’s the click through rate? How much revenue we can get from recommend restaurant?  Any potential loss.
3.        About algorithm. All previous considerations are assuming the algorithm is perfect(nothing to improve). However, we also want to estimate how good is the recommendation algorithm and separate the effect of a good algorithm from a good UI. We can also design some tests comparing a algorithm  restaurant recommendation with purely previous visiting restaurant. In this way, we can estimate, how many user can be attracted by the recommendation feature vs how many users can be attracted by a really good recommendation system.

还有，我觉得time spent per active user 应当服从log normal distribution。这题并不考分布，也不是考概率知识，主要考察对一些KPI的常识，time spent per active 必然right skewed.因为大部分的人在FB上花的时间都不会太长，但是有一小部分人会不停刷，花的时间超级长，所以造成long tail。我觉得这种问分布的题和很多考distribution的概率题反而相辅相成，实际生活中，这种metrics都很符合二八法则，统计上来说，就是lognormal 分布。

kellynum3 发表于 2018-8-23 01:55
請問樓主 time spent per active user distribution 怎麼回答？

如果是每个user的time spent。我觉得类似poisson分布吧。如果是 literally average time spent per active user...中心极限感觉正太没毛病...

renewu427 发表于 2018-8-23 01:29
lz你有几年工作经验啊

差不多三年吧

lz你有几年工作经验啊

請問樓主 time spent per active user distribution 怎麼回答？

kellynum3 发表于 2018-8-23 01:55
請問樓主 time spent per active user distribution 怎麼回答？

我不是很确定欸。。我说正态分布。。

楼主recruiter有没有跟你说更具体一些的feedback？我看过了的帖子都有说positive， fair还是mixed feedback的

Sarah0813 发表于 2018-8-23 02:37
我不是很确定欸。。我说正态分布。。

和楼主背景差不多啊....可是电面product就挂了....楼主也加油！

hyper8866 发表于 2018-8-23 11:02
和楼主背景差不多啊....可是电面product就挂了....楼主也加油！

谢啦！同加油！