Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

DIN model中为何没有拼接user vector #64

Open
zhiweishen opened this issue Nov 21, 2019 · 4 comments
Open

DIN model中为何没有拼接user vector #64

zhiweishen opened this issue Nov 21, 2019 · 4 comments

Comments

@zhiweishen
Copy link

No description provided.

@JAPTRA
Copy link

JAPTRA commented Apr 13, 2020

我也注意到了,也就是说训练的时候没有考虑user特征,用户行为数据代表了user;我觉得是避免训练的时候出现user_vec,[item_list1],next_item,y 中如果,[item_list1]不停的改变,user_vec 不变,那么最后决定y的只有next_item。拟合过程中就会出现针对一个user只有next_item起作用的搞笑情况?

@zhougr1993
Copy link
Owner

其他的模型里u_emb也是被hist替换掉了。我对小数据集没有特别强的感觉和经验,生产的时候也很难用上user id的,user id单独还好维度不是很大,但是一旦样本里user id和 目标item id都存在的时候,整个样本的特征表达和区分能力就非常强,比较容易过拟合。不过是听说过有的场景用user id成功的经历,我们的数据体量也非常大了,结论就是要么过拟合要么无效,没有做过多的研究。

@chouchou1988
Copy link

我也注意到了,也就是说训练的时候没有考虑user特征,用户行为数据代表了user;我觉得是避免训练的时候出现user_vec,[item_list1],next_item,y 中如果,[item_list1]不停的改变,user_vec 不变,那么最后决定y的只有next_item。拟合过程中就会出现针对一个user只有next_item起作用的搞笑情况?

不能这么理解吧,虽然不同样本user id不变,target id在变,但是不太表target id决定了label,只能说每个特征在学习和label的相关性

@oliverlwang
Copy link

din(hist, uid_embedding, item) 可以学到哪些hist对该用户更有价值,哪些是误点的噪声?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants