product sense 面试经典例题分析答案 分享求大米
69914
上次给大家分享了一个 怎么提高 facebook engagement rate 的product题 求到了一些大米 谢谢大家 这次在分析第二个经典题目 父老乡亲 请加大米啊 小的 穷的什么也干不了
FB - Should we add a love button? 这个题目很常见 基本每个公司都可以考 比如 what feature do you think to add to whatup and why？
Answer:
This is another super common question type, where the candidate is asked whether it would be a good idea to implement a new feature. Although it is not obvious from the way the question is phrased, even questions like this one should be answered following a very quantitative framework.
This is a particularly important question since a big part of the job of a DS is to look at the data and, based on findings, suggest new feature/product ideas.
Break down the answer into two steps. Firstly,
· If the feature were highly successful, would it be a good thing for the site?
To answer this, pick a crucial metric for the site that that feature is supposed to move. If you can't find any, then there is no point in even discussing further.
In this case, we pick engagement as the key metric, where engagement is as usual defined as number of actions (likes/posts/comments/love/pics uploads/etc) per user within a certain time frame.
We can realistically expect that if the feature becomes highly popular, that will benefit engagement for two reasons: clicking on the love button itself will improve engagement + it might incentivize posting new statuses. After all, receiving a lot of "love", or positive attention in general, is a nice feeling and will make people more likely to post.
The real key part of the answer is the second part though:
· You need to find a proxy for demand of that feature in your current data
The safest way for a data scientist to drive new feature decisions is to look at what users are doing today on the site. See if you want to incentivize that behavior and, if so, try to simplify it, so that users can complete it more easily. The fact that users are already doing it today shows demand. Since there is demand, simplifying it will for sure increase the number of users doing it.
For instance, if many users are performing some sort of activity on your site and that requires several clicks in a row, find a way so that they can just do it with one click (think about buying with one click for instance). Please, note that the converse is also true. If you want to disincentivize some actions, add complexity (think about how many clicks it takes logging out or contacting customer service on many sites).
So, back to the Facebook example, we need some data that can be used as a proxy for demand for that feature. The easiest thing to do is to look at current comments and use NLP to extract sentiments. Are there many comments that can be broadly assigned to a "love category"? If so, that's a pretty strong indicator of demand for a love button. Indeed, that's telling us that today users are going through a longer path to express love (i.e. actually typing a comment instead of just clicking on a button) and it is realistic to assume that, as you simplify the path, you will have even more users doing it, i.e. all those people who share the same feeling, but don't feel like going through the multi-click path.
Please, note that even if your conclusions are that (a) the feature, if successful, would benefit the main metric, (b) you found a proxy for large demand, and (c) you are simplifying the path, this doesn't mean you should implement the feature. It means you should test it. Then, if and only if the test is successful, you should implement it. There is no way to know if that feature will actually move an important metric without running an experiment.
大家如果还有别的建议想法 欢迎留言哦 求各位父老乡亲不要吝啬 给小的一些大米 跪求大米 谢谢
大家不要只收藏 不加大米啊 给别人加大米 你不用用自己的大米 每人每天有十个大米可以加给别人 不加也浪费了 不要浪费粮食 我的衣食父母们 跪求大米 😂
FB - Should we add a love button? 这个题目很常见 基本每个公司都可以考 比如 what feature do you think to add to whatup and why？
Answer:
This is another super common question type, where the candidate is asked whether it would be a good idea to implement a new feature. Although it is not obvious from the way the question is phrased, even questions like this one should be answered following a very quantitative framework.
This is a particularly important question since a big part of the job of a DS is to look at the data and, based on findings, suggest new feature/product ideas.
Break down the answer into two steps. Firstly,
· If the feature were highly successful, would it be a good thing for the site?
To answer this, pick a crucial metric for the site that that feature is supposed to move. If you can't find any, then there is no point in even discussing further.
In this case, we pick engagement as the key metric, where engagement is as usual defined as number of actions (likes/posts/comments/love/pics uploads/etc) per user within a certain time frame.
We can realistically expect that if the feature becomes highly popular, that will benefit engagement for two reasons: clicking on the love button itself will improve engagement + it might incentivize posting new statuses. After all, receiving a lot of "love", or positive attention in general, is a nice feeling and will make people more likely to post.
The real key part of the answer is the second part though:
· You need to find a proxy for demand of that feature in your current data
The safest way for a data scientist to drive new feature decisions is to look at what users are doing today on the site. See if you want to incentivize that behavior and, if so, try to simplify it, so that users can complete it more easily. The fact that users are already doing it today shows demand. Since there is demand, simplifying it will for sure increase the number of users doing it.
For instance, if many users are performing some sort of activity on your site and that requires several clicks in a row, find a way so that they can just do it with one click (think about buying with one click for instance). Please, note that the converse is also true. If you want to disincentivize some actions, add complexity (think about how many clicks it takes logging out or contacting customer service on many sites).
So, back to the Facebook example, we need some data that can be used as a proxy for demand for that feature. The easiest thing to do is to look at current comments and use NLP to extract sentiments. Are there many comments that can be broadly assigned to a "love category"? If so, that's a pretty strong indicator of demand for a love button. Indeed, that's telling us that today users are going through a longer path to express love (i.e. actually typing a comment instead of just clicking on a button) and it is realistic to assume that, as you simplify the path, you will have even more users doing it, i.e. all those people who share the same feeling, but don't feel like going through the multi-click path.
Please, note that even if your conclusions are that (a) the feature, if successful, would benefit the main metric, (b) you found a proxy for large demand, and (c) you are simplifying the path, this doesn't mean you should implement the feature. It means you should test it. Then, if and only if the test is successful, you should implement it. There is no way to know if that feature will actually move an important metric without running an experiment.
大家如果还有别的建议想法 欢迎留言哦 求各位父老乡亲不要吝啬 给小的一些大米 跪求大米 谢谢
大家不要只收藏 不加大米啊 给别人加大米 你不用用自己的大米 每人每天有十个大米可以加给别人 不加也浪费了 不要浪费粮食 我的衣食父母们 跪求大米 😂
- 47
4条回复