Issues-translate-bot
commented
1 year ago Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑🤝🧑👫🧑🏿🤝🧑🏻👩🏾🤝👨🏿👬🏿
We are a start-up company and are preparing to build a simple recommendation system using Gorse, but we encountered a problem. I noticed that the similar algorithm above uses more of the characteristics of the item itself as the label, and does not use the popularity of the item (only positive feedback, but does not pay attention to those secondary indicators that are not the final optimization goal). For example, when we make video recommendations, we usually focus on the classification of the content itself, what topics it has, who the author is, etc. (These are relatively easy to apply as content tags). Many times we will also pay attention to the click-through rate, completion rate, per capita viewing time and other indicators of the content itself. How are these latter consumer characteristics reflected?
In addition, what if we want to do multi-objective weighting for the final optimization goal?
Thanks for the answer.
alicksnake22
kolychen
我们是一家初创公司，正准备用Gorse搭建一套简单的推荐系统，遇到了一个问题。我注意到上面的相似算法更多的是采用物品本身的特征作为标签，没有采用物品的受欢迎程度（只有正反馈，但是不关注不是最终优化目标的那些次级指标）。 举个例子，我们做视频推荐，一般会关注内容本身什么分类，有那些主题，作者是谁等等（这些作为内容的标签是相对容易套入的）。很多时候我们还会关注这个内容本身的点击率，完播率，人均观看时长等等这些指标。后面这些消费性的特征如何体现呢？
另外如果要给最终的优化目标做多目标加权怎么搞呢？
感谢解答。