联合会员
周边
新闻
博问
闪存
赞助商
Chat2DB
所有博客
当前博客
我的博客
我的园子
账号设置
会员中心
简洁模式
...
退出登录
注册
登录
xd_xumaomao
博客园
首页
新随笔
联系
管理
订阅
随笔 - 390 文章 - 120 评论 - 14 阅读 -
25万
05 2022 档案
《CAN: Revisiting Feature Co-Action for Click-Through Rate Prediction》论文阅读
摘要:背景 作者发现把预估商品信息(如item_id)和用户用户历史行为序列(如item_id sequence)做笛卡尔积,形成一个新的id suquence, 对其做embedding后pooling效果会在din/dien的基础上有明显的提升。DIN/DIEN等建模方式的特征交叉最终算出了一个wei
阅读全文
posted @
2022-05-01 14:37
xd_xumaomao
阅读(209)
评论(0)
推荐(0)
编辑
公告
昵称:
xd_xumaomao
园龄:
5年9个月
粉丝:
1
关注:
0
+加关注
<
2025年3月
>
日
一
二
三
四
五
六
23
24
25
26
27
28
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
1
2
3
4
5
搜索
我的标签
推荐系统
(110)
C++
(52)
linux
(37)
机器学习
(35)
c
(26)
经典算法题
(24)
hadoop
(23)
python
(22)
java
(19)
深度学习
(18)
更多
随笔档案
2025年3月(1)
2025年2月(10)
2025年1月(5)
2024年12月(6)
2024年11月(2)
2024年10月(1)
2024年9月(1)
2024年6月(1)
2024年5月(15)
2024年4月(2)
2024年3月(2)
2024年2月(6)
2024年1月(15)
2023年12月(5)
2023年6月(3)
2023年5月(14)
2023年3月(6)
2022年11月(4)
2022年9月(1)
2022年8月(1)
2022年6月(2)
2022年5月(1)
2022年4月(2)
2022年3月(1)
2022年2月(1)
2021年12月(1)
2021年10月(2)
2021年9月(14)
2021年8月(13)
2021年7月(12)
2021年3月(3)
2020年9月(1)
2020年6月(32)
2020年5月(32)
2020年4月(27)
2020年3月(5)
2020年2月(12)
2019年12月(19)
2019年11月(23)
2019年10月(5)
2019年9月(18)
2019年8月(21)
2019年7月(22)
2019年6月(20)
更多
阅读排行榜
1. 经典算法面试题(15786)
2. max{X,Y}和min{X,Y}的概率分布(15179)
3. 推荐系统中为什么要有召回、粗排、精排(9258)
4. python 浮点数保留几位小数(8137)
5. softmax+交叉熵损失函数代码实现(4078)
推荐排行榜
1. max{X,Y}和min{X,Y}的概率分布(2)
2. C++ std::move(1)
3. Java wait()、notify()、notifyAll()方法(1)
4. softmax+交叉熵损失函数代码实现(1)
最新评论
1. Re:Weighted LR (WCE Weighted cross entropy)
@gxxxxxxl 是的,第三行最右边是y'...
--xd_xumaomao
2. Re:Weighted LR (WCE Weighted cross entropy)
倒数第二张图的公式推导有问题吧,第三行的 y' 就开始少了,对吗?
--gxxxxxxl
3. Re:ListNet和ListMLE
@葡萄和咖啡 是的,不小心写错了,感谢指正...
--xd_xumaomao
4. Re:ListNet和ListMLE
博主你好,请问你的例子里面是不是应该是三个排列的概率相乘而非相加呢?
--葡萄和咖啡
5. Re:对比学习《Self-supervised Learning for Large-scale Item Recommendations》
@dddddddddsdf 按论文说的,应该是先按互信息把特征分成两个集合:集合1和集合2,样本1是mask集合1中的特征+dropout得到的,样本2时mask集合2中的特征+dropout得到的,...
--xd_xumaomao
点击右上角即可分享