Xavier Amatriain,Quora工程VP。一起来看看他分享的Quora推荐系统(recommender systems,下面也会简写为recsys)构建经验。
Quora的使命
Quora的数据情况
大量高质量文本信息
大量的数据关联
Quora的推荐系统
在Quora,很多地方都会用到推荐。
模型
从推荐系统构建过程中学到的经验
implicit signals beat explicit ones (almost always)
(注:显式信号指的是直接收集的反馈,比如让用户打分,或者点反对/支持。隐式信号指的是通过用户行为分析出的信息,比如根据用户日志来分析。)
2.be thoughtful about your training data
3. your model will learn what you teach it to learn
4.explanations might matter more than the prediction
5. if you have to pick one single approach, matrix factorization is your best bet
6. everything is an ensemble
7. building recommender systems is also about feature engineering
8. why you should care about answering questions (about your recsys)
9. Data and models are great. You know what's even better? The right evaluation approach!
10. You don’t need to distribute your recsys
(一定要分布式吗?不见得)
结论
幻灯片链接:http://www.slideshare.net/xamat/recsys-2016-tutorial-lessons-learned-from-building-reallife-recommender-systems
(完)
近期热文