当前位置:网站首页>Development trend of Ali Taobao fine sorting model
Development trend of Ali Taobao fine sorting model
2022-07-06 01:06:00 【learner_ ctr】
One 、DIN The Internet 2018 year 7 month 19 Japan
, Put forward attention Structure is used for user sequence characteristics , such attention No self-attention, It's an estimate item Used as query, Each in the user sequence item As key value
Two 、DIEN The Internet 2018 year
be relative to DIN The network has such improvements ,DIN The network only makes user sequence attention This matter ,DIEN First, add a loss, use gru Calculate the click sequence ( The length is T) Of T Outputs , Then take the front. T-1 And output after T-1 Seek accumulation loss, In this way, this prediction will include the significance of predicting the next .
For this T Output using tape attention Of gru Get the final output final_state2, And other features (user Of embedding、item Of embedding、 Various sequences 、 Dot product of the first two )concat together , Send it to the back mlp layer
dien Thesis translation _1066196847 The blog of -CSDN Blog
3、 ... and 、 Long sequence modeling MIMN 2019 year 7 month 25 Japan
It's solved DIN DIEN It can't deal with the bottleneck of long sequence , This model can handle lengths up to 1000 Sequence modeling
Modeling the long-term historical user behavior of Alibaba mom ——MIMN The model, - You know
Four 、sim Model 2020 year 6 month
MIMN Ask for more , No consideration target item;din Be accurate , But it can't handle long sequences ,sim The model can handle lengths up to 10000 Sequence , This is in the e-commerce scene ( TaoBao ) Or content scene ( Tiktok ), It's very common , For example, I learned from the article ,180 Day behavior data 30% The click sequence of samples exceeds 10000, This proportion is quite a lot
sim First, search out a batch of suitable item, recycling din Thoughts are processed
边栏推荐
- Mobilenet series (5): use pytorch to build mobilenetv3 and learn and train based on migration
- [groovy] XML serialization (use markupbuilder to generate XML data | create sub tags under tag closures | use markupbuilderhelper to add XML comments)
- Exciting, 2022 open atom global open source summit registration is hot
- Xunrui CMS plug-in automatically collects fake original free plug-ins
- The detailed page returns to the list and retains the original position of the scroll bar
- The inconsistency between the versions of dynamic library and static library will lead to bugs
- Beginner redis
- 直播系统代码,自定义软键盘样式:字母、数字、标点三种切换
- 95后CV工程师晒出工资单,狠补了这个,真香...
- The population logic of the request to read product data on the sap Spartacus home page
猜你喜欢
Convert binary search tree into cumulative tree (reverse middle order traversal)
cf:H. Maximal AND【位运算练习 + k次操作 + 最大And】
After Luke zettlemoyer, head of meta AI Seattle research | trillion parameters, will the large model continue to grow?
JVM_ 15_ Concepts related to garbage collection
2020.2.13
[groovy] XML serialization (use markupbuilder to generate XML data | create sub tags under tag closures | use markupbuilderhelper to add XML comments)
猿桌派第三季开播在即,打开出海浪潮下的开发者新视野
[groovy] compile time meta programming (compile time method interception | method interception in myasttransformation visit method)
如何制作自己的機器人
Vulhub vulnerability recurrence 74_ Wordpress
随机推荐
Finding the nearest common ancestor of binary tree by recursion
cf:C. The Third Problem【关于排列这件事】
小程序容器可以发挥的价值
激动人心,2022开放原子全球开源峰会报名火热开启
[simple implementation of file IO]
282. Stone consolidation (interval DP)
在产业互联网时代,将会凭借大的产业范畴,实现足够多的发展
【第30天】给定一个整数 n ,求它的因数之和
从 1.5 开始搭建一个微服务框架——调用链追踪 traceId
Recursive method converts ordered array into binary search tree
Getting started with devkit
C language programming (Chapter 6 functions)
Spark SQL null value, Nan judgment and processing
[pat (basic level) practice] - [simple mathematics] 1062 simplest fraction
视频直播源码,实现本地存储搜索历史记录
关于#数据库#的问题:(5)查询库存表中每本书的条码、位置和借阅的读者编号
[groovy] JSON string deserialization (use jsonslurper to deserialize JSON strings | construct related classes according to the map set)
Differences between standard library functions and operators
Browser reflow and redraw
Cf:h. maximum and [bit operation practice + K operations + maximum and]