当前位置:网站首页>论文《Deep Multifaceted Transformers for Multi-objective Ranking in Large-Scale E-commerce Recommender》
论文《Deep Multifaceted Transformers for Multi-objective Ranking in Large-Scale E-commerce Recommender》
2022-08-02 06:14:00 【巴拉巴拉朵】
京东DMT
论文地址:https://dl.acm.org/doi/pdf/10.1145/3340531.3412697
论文提出用多个Transformer对用户多种类型的行为序列进行建模,在此基础上叠加MMOE建模多目标,最后使用一个消偏塔对数据进行消偏。
拿点击/未点击作为反馈通常会有位置偏差(position bias)和近邻偏差(neighboring
bias),不过论文对于消偏的处理比较简单。
DMT的结构如下
输入分为两种特征 Categorical features 和 Dense features
Categorical features:
1.用户不同的行为序列。 S = < s 1 , s 2 , . . . , s T > S = <s_1, s_2, ..., s_T> S=<s1,s2,...,sT>,其中 s t = ( t i , p i ) s_t=(t_i,p_i) st=(ti,pi)表示用户在时刻 t i t_i ti交互的物料 p i p_i pi,论文用到了点击click序列 S c S_c Sc,加入购物车cart序列 S a S_a Sa,购买order序列 S o S_o So
2. Embedding Layer,对每个物料,使用物料id p i p_i pi,类目id c i c_i ci,品牌id b i b_i bi,商铺id s i s_i si,分别映射成低维向量 e p i , e c i , e b i , e s i e_{p_i}, e_{c_i}, e_{b_i}, e_{s_i} epi,eci,ebi,esi,然后concat起来,形成向量 e i e_i ei
Dense features: 归一化处理
1.item profile features (e.g.,number of clicks, CTR, CVR, rating) ,
2. profile features (e.g., purchase power, preferred categories and brands),
3. user-item matching features (e.g., whether the item matches the user’s gender or age)
4. user-item interaction features (e.g., number of clicks on the category of the item within a time window)
Deep Multifaceted Transformers Layer
分别用3个Transformer来对点击、加入购物车、购买行为序列进行建模。encoder中,用序列的item-Embedding作为self-attention的输入,decoder中,使用target item的Embedding作为query,encoder输出的结果作为key和value。
Multi-gate Mixture-of-Experts Layers
专家网络输出 e 1 ( x ) , e 2 ( x ) , . . . , e N ( x ) e_1(x),e_2(x),...,e_N(x) e1(x),e2(x),...,eN(x),每个任务的门控网络 N N G k ( x ) NNG^k(x) NNGk(x)学习各个专家的权重 w k w^k wk,根据权重得到专家结果的加权和,然后送入到一个功能网络中得到任务在MMOE层的输出。
w k = s o f t m a x ( N N G k ( x ) ) w^k = softmax(NNG^k(x)) wk=softmax(NNGk(x))
f k ( x ) = ∑ i = 1 N w i k e i ( x ) f^k(x) = \sum_{i=1}^N w_i^k e_i(x) fk(x)=i=1∑Nwikei(x)
u k = N N U k ( f k ( x ) ) u_k = NN_U^k(f^k(x)) uk=NNUk(fk(x))
Bias Deep Neural Network
专门搭了一个bais塔,输入都是偏差相关的特征,对于位置偏差就是展示位置索引编号或者网页索引编号;对于近邻偏差,输入就是目标物料的类目和邻近K个物料的类目。
biase塔的输出
y b = N N B ( x b ) y_b = NN_B(x_b) yb=NNB(xb)
Model Training and Prediction
模型输出 y k y_k yk
都是分类任务,使用交叉熵损失函数
y k = σ ( u k + y b ) y_k = \sigma (u_k+y_b) yk=σ(uk+yb)
L k = − 1 N ∑ i = 1 N y i log ( y k ) + ( 1 − y i ) log ( 1 − y k ) L_k = - \frac 1 N \sum_{i=1}^N y_i \log(y_k) + (1-y_i) \log (1-y_k) Lk=−N1i=1∑Nyilog(yk)+(1−yi)log(1−yk)
L = ∑ i = 1 K λ k L k L = \sum_{i=1}^K \lambda_k L_k L=i=1∑KλkLk
上面是训练阶段,预测阶段,任务k输 y ^ k \hat y_k y^k,score由不同任务预估分加权得到,权重离线搜参得到
y ^ k = σ ( u k ) \hat y_k = \sigma (u_k) y^k=σ(uk)
y ^ = ∑ k = 1 K w k y ^ k ∑ k = 1 K w k \hat y = \frac {\sum_{k=1}^Kw_k \hat y_k} {\sum_{k=1}^Kw_k} y^=∑k=1Kwk∑k=1Kwky^k
EXPERIMENTAL

其实这个论文是对Transformer和MMOE以及消偏做了组合,不同的Transformer对不同种类的序列分别处理,能拿到一个比较好的这个序列的Embedding结果,这种组合竟然可以很好地work,说明几个基础组件还是非常有效的。
边栏推荐
- zabbix auto-discovery and auto-registration
- typescript ‘props‘ is declared but its value is never read 解决办法
- Two good php debug tutorials
- APP special test: traffic test
- rhce homework
- 实验7 MPLS实验
- abaqus如何快速导入其他cae文件的assembly?
- [Dataset][VOC] Male and female dataset voc format 6188 sheets
- Unity Shader学习(七)纹理图像的简单使用
- 数据库概论之MySQL表的增删改查1
猜你喜欢

Clapper that can interact with the audience in real time

Toolbox App 1.25 New Features at a Glance | Version Update

Leading the demand and justifying the HR value - the successful launch of the "Human Resource Leading Model HRLM"

文件上传漏洞(二)

MySQL 5.7 installation tutorial (full-step, nanny-level tutorial)

武汉高性能计算大会2022举办,高性能计算生态发展再添新动力
![[Dataset][VOC] Eyewear dataset 6000 in VOC format](/img/66/37f76d9ce5d5f68d6ea0e18710fa04.png)
[Dataset][VOC] Eyewear dataset 6000 in VOC format

How the Internet of Things is changing the efficiency of city operations

MySQL Advanced SQL Statements

chrome 插件开发指南
随机推荐
HCIP 第二天
有人开源全凭“为爱发电”,有人却用开源“搞到了钱”
optional
MySQL Advanced Statements (1)
MySQL 5.7 installation tutorial (full-step, nanny-level tutorial)
odoo field 设置匿名函数domain
Technology empowers Lhasa's "lungs", Huawei helps Lalu Wetland Smart Management to protect lucid waters and lush mountains
Ant three sides: MQ message loss, duplication, backlog problem, what are the solutions?
.NET Static Code Weaving - Rougamo Release 1.1.0
【21天学习挑战赛】顺序查找
MySQL Advanced Study Notes
MySQL high-level statements (1)
数据库概论之MySQL表的增删改查1
(笔记整理未完成)【图论】图的遍历
笔记本开机黑屏提示:ERROR 0199:System Security-Security password retry count exceeded
PHP Warning: putenv() has been disabled for security reasons in phar
HCIP day 3 experiment
Mining game (C language)
(Notes are not completed) [Graph Theory] Traversal of graphs
Reverse resolve dns server