如何找到最受欢迎的产品
这更像是一个一般性的、集思广益的查询，而不是一个问题。 所以就这样了。 假设我有 1000 件商品，可以随时在网站上出售。 在任何一天，我只能销售（比如说）100 件商品。 可以说，明显的目标是在任何一天销售最大数量的产品（收入最大化）。 解决这个问题的最好方法是什么。问题是.. 我如何选择 1000 种产品中的哪 100 种？我确信，有了这些信息，这是相当困难的.. 如果我必须使用机器学习来解决它，这个问题将属于哪一类。 这不是监督学习..？ 基本上，我只是在寻找周围的人给我一些想法..:) 谢谢
This is more like a general, brainstroming query rather than a question.
So here it goes.
Suppose, I have 1000 items which I can sell on a website at any given day.
On any given day, I can sell (lets say) only 100 items..
And lets say, the obvious goal is to sell the maximum number of products any given day (maximimzing revenue)..
What is the best way to solve this problem. The problem being.. how do i select which 100 products of 1000 to select??I am sure, with this information it is pretty hard.. If I have to solve it using machine learning, which category will this problem fall into.
This is not supervised learning.. ?
Basically, I am just looking for the people around here to throw me some ideas.. :)
Thanks
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（4）
这是多臂老虎机问题的变体。优化利润的最基本方法是选择一个数字 ε є [0,1]，然后对于您选择的每种产品，以概率 (1-ε) 或以概率 ε 随机选择最受欢迎的产品。这样，随着时间的推移，您将对每种产品的受欢迎程度进行越来越准确的估计，同时保持较高的总利润。更复杂的方法包括跟踪产品受欢迎程度的置信上限。
This is a variation of the multi-armed bandit problem. The most basic way of optimizing your profit is to pick a number ε є [0,1], then for each of the products you select choose the most popular one with probability (1-ε) or random with probability ε. This way, over time, you will have increasingly accurate popularity estimates for each product, while keeping your total profits high. More sophisticated approaches include tracking upper confidence bounds on product popularity.
天真的方法是存储每个产品被销售的概率（多少次显示多少次）。并根据这些概率从 1000 个中选择 100 个。因此任何产品都有一个可供选择的变化，并且变化等于其概率。顺便说一句，对于此概率，最好使用置信下限。因此，在 100 次中销售 70 次的产品比在 10 次中销售 7 次的产品更好。
Naive approach is store for each product the probability of being sold (how many times shown how many time solt). And based on those probability select 100 among 1000. So any product has a change to be selected and the changes is equal to its propability. BTW for this probability it is better to use lower confidence bound. So product which is sold 70 from 100 times is better then product which sold 7 from 10.
我受到这篇文章的启发，在 Hadoop 中实现了一堆多臂老虎机算法，作为我在 github 上基于 Hadoop 的开源机器学习项目 avenir 的一部分。我最近还发布了一篇博客，内容是关于使用一些已实现的算法找到最佳产品价格。
http://pkghosh.wordpress.com/ 2013/08/25/bandits-know-the-best-product-price/
https://github.com/pranab/avenir
I was inspired by this post to implement bunch of multi-arm bandit algorithms in Hadoop as part of my open source Hadoop based machine learning project avenir on github. I also posted a blog recently which is about finding optimum product price using some the implemented algorithms.
http://pkghosh.wordpress.com/2013/08/25/bandits-know-the-best-product-price/
https://github.com/pranab/avenir
那么您从哪里获得有关物品的数据？例如，如果您有每个商品的一些销售数据，您只需在数据库中为其创建一个列，然后 SELECT * FROM items ORDER BY 'number_sold' DESC LIMIT 100 并每次更新为 +1商品已售出
well where would you get ur data on the items from? If for example you had some sales figures for each item you would simply create a column for that in the database then
SELECT * FROM items ORDER BY 'number_sold' DESC LIMIT 100and update it with +1 every time an item is sold