当前位置:网站首页>自适应非欧表征广告检索系统AMCAD
自适应非欧表征广告检索系统AMCAD
2022-07-06 20:35:00 【智源社区】
图表征(Graph Embedding)是近年来信息检索领域内最流行的方法之一,但它们往往建模在平坦的欧氏空间中。近年来的研究发现欧氏空间在建模特定图结构如层次、环状结构时天然存在误差,在本文中,我们将介绍如何利用非欧图表征提高复杂异构图建模精度同时在阿里妈妈搜索广告场景下取得线上收益。该项工作已被国际数据工程顶级会议ICDE 2022 (International Conference on Data Engineering) 收录,对应框架也已开源。
论文标题:
AMCAD: Adaptive Mixed-Curvature Representation based Advertisement Retrieval System
下载链接:
https://arxiv.org/abs/2203.14683
开源地址:
https://github.com/alibaba/Curvature-Learning-Framework
本文提出了自适应非欧表征广告检索系统AMCAD,首次将混合曲率空间应用到大规模工业数据上,使模型端到端的根据复杂数据结构自动学习出损失最低的表征空间。
基于CurvLearn(开源地址:https://github.com/alibaba/Curvature-Learning-Framework)提供的非欧深度学习算子,AMCAD旨在使用多种曲率空间表征不同异构节点,通过自动学习空间曲率及动态组合权重实现对复杂异构图的高精度表征。下图展示了AMCAD的具体设计,左侧为模型总体架构,从底至上分为以下三阶段:

AMCAD整体架构,分为点级别自适应混合曲率编码、边级别空间映射及子空间距离融合三部分,分别对应复杂结构的混合曲率表征、异构节点的自适应边空间及多几何空间的权重融合
边栏推荐
- 安装 torch 0.4.1
- VHDL implementation of arbitrary size matrix multiplication
- LAB1配置脚本
- Netperf and network performance measurement
- Flutter3.0了,小程序不止于移动应用跨端运行
- Free PHP online decryption tool source code v1.2
- Can the applet run in its own app and realize live broadcast and connection?
- 数学归纳与递归
- Shangsilicon Valley JVM Chapter 1 class loading subsystem
- 我的勇敢对线之路--详细阐述,浏览器输入URL发生了什么
猜你喜欢

函数重入、函数重载、函数重写自己理解

Not All Points Are Equal Learning Highly Efficient Point-based Detectors for 3D LiDAR Point
![[safe office and productivity application] Shanghai daoning provides you with onlyoffice download, trial and tutorial](/img/58/d869939157669891f369fb274d32af.jpg)
[safe office and productivity application] Shanghai daoning provides you with onlyoffice download, trial and tutorial

HMS Core 机器学习服务打造同传翻译新“声”态,AI让国际交流更顺畅

24.(arcgis api for js篇)arcgis api for js点修改点编辑(SketchViewModel)

About Confidence Intervals

Variables, process control and cursors (MySQL)

Calculation of time and space complexity (notes of runners)

1200.Minimum Absolute Difference

R数据分析:cox模型如何做预测,高分文章复现
随机推荐
Room rate system - login optimization
23. (ArcGIS API for JS) ArcGIS API for JS ellipse collection (sketchviewmodel)
20. (ArcGIS API for JS) ArcGIS API for JS surface collection (sketchviewmodel)
大白话高并发(二)
变量、流程控制与游标(MySQL)
21. (article ArcGIS API for JS) ArcGIS API for JS rectangular acquisition (sketchviewmodel)
Stored procedures and functions (MySQL)
Shell programming basics
22.(arcgis api for js篇)arcgis api for js圆采集(SketchViewModel)
R data analysis: how to predict Cox model and reproduce high score articles
树莓派设置wifi自动连接
qt-线程等01概念
如何自定义Latex停止运行的快捷键
sshd[12282]: fatal: matching cipher is not supported: aes256- [email protected] [preauth]
U.S. Air Force Research Laboratory, "exploring the vulnerability and robustness of deep learning systems", the latest 85 page technical report in 2022
When you go to the toilet, you can clearly explain the three Scheduling Strategies of scheduled tasks
MySQL的存储引擎
华为小米互“抄作业”
[leetcode] 700 and 701 (search and insert of binary search tree)
Restcloud ETL Community Edition June featured Q & A