KDD 2021 | 推荐系统论文集锦[持续更新]

作者 | ‍GuoXun  

整理 | NewBeeNLP

KDD2021 放榜，其中research track共收到了1541篇投稿，接收了238篇长文，Applied Data Science Track共收到了705篇投稿，接收了138篇长文。现在我们来学习一下今年推荐方向论文。KDD最近几年的热门主题之一就是商业智能方向，即推荐系统和计算广告。本文整理了KDD2021上推荐系统和计算广告方向的论文。

KDD官网上列出了今年的完整List：

• https:kdd.org/kdd2021/accepted-papers/index

1.  推理

因果推断是推荐系统近期的热点，可以为推荐效果提升、AB实验等带来可靠性分析。这三篇分别是新闻推荐推理的增强锚点知识图生成、社会意识自监督的立体推荐系统、不可知反事实推理模型消除推荐系统的流行偏差。

• Reinforced Anchor Knowledge Graph Generation for News Recommendation Reasoning

• Socially-Aware Self-Supervised Tri-Training for Recommendation

• Model-Agnostic Counterfactual Reasoning for Eliminating Popularity Bias in Recommender System

2.  多任务、多目标、跨领域推荐场景

多目标优化一直是推荐业务追求的目标。主要有以下四篇文章：序列依赖多任务学习、混合场景多任务学习、对抗特征迁移多任务学习、迁移学习去偏。

• Modeling the Sequential Dependence among Audience Multi-step Conversions with Multi-task Learning in Targeted Display Advertising

• Learning to Expand Audience via Meta Hybrid Experts and Critics for Recommendation and Advertising

• Adversarial Feature Translation for Multi-domain Recommendation

• Debiasing Learning based Cross-domain Recommendation

3.  纠偏

构建一个稳定运行的推荐生态系统，纠偏的措施必不可少。涉及有以下五篇：反事实模型推断纠偏、动态推荐系统的热度纠偏、大规模推荐系统纠偏、跨域推荐纠偏等。

• Deconfounded Recommendation for Alleviating Bias Amplification

• Model-Agnostic Counterfactual Reasoning for Eliminating Popularity Bias in Recommender System

• Popularity Bias in Dynamic Recommendation

• Contrastive Learning for Debiased Candidate Generation in Large-Scale Recommender Systems

• Debiasing Learning based Cross-domain Recommendation

4. 基于图的推荐系统

图神经网络落地推荐系统是近期的热点，是建模类图关系的有效工具。今年涉及文章如下：高效图神经网络训练、面向冷启动推荐的异构信息网络多视图去噪图自动编码器、新闻推荐推理的增强锚点知识图生成。

• MixGCF: An Improved Training Method for Graph Neural Network-based Recommender Systems

• Multi-view Denoising Graph Auto-Encoders on Heterogeneous Information Networks for Cold-start Recommendation

• Reinforced Anchor Knowledge Graph Generation for News Recommendation Reasoning

5. 冷启动

冷启动是推荐系统建立初期必然面对的问题，今年文章如下：异构信息网络多视图去噪图自动编码器实现冷启动、半个性化的音乐流媒体应用冷启动推荐系统、在线推荐系统的架构及其自适应网络的操作。

• Multi-view Denoising Graph Auto-Encoders on Heterogeneous Information Networks for Cold-start Recommendation

• A Semi-Personalized System for User Cold Start Recommendation on Music Streaming Apps

• Architecture and Operation Adaptive Network for Online Recommendations

6. 序列推荐

序列推荐今年只有一篇文章，基于序列多模态信息传输网络的电商微视频推荐系统。

• SEMI: A Sequential Multi-Modal Information Transfer Network for E-Commerce Micro-Video Recommendations

7. 兴趣推荐

兴趣点推荐：基于元学习的下一代兴趣点推荐系统。

• Curriculum Meta-Learning for Next POI Recommendation

8. Embedding

Embedding可以认为是推荐算法的核心基石之一，无表embedding是一个不错的尝试。文章如下：定制设备上的弹性embedding、无embedding表的推荐系统特征建模、推荐系统中的偏好放大、推荐系统中网络嵌入方法的综合分析。

• Learning Elastic Embeddings for Customizing On-Device Recommenders

• Learning to Embed Categorical Features without Embedding Tables for Recommendation

• Preference Amplification in Recommender Systems

• Where are we in embedding spaces? A Comprehensive Analysis on Network Embedding Approaches for Recommender System

9. 蒸馏

蒸馏是为了解决小型化的问题，今年有一篇文章。基于拓扑蒸馏的推荐系统。

• Topology Distillation for Recommender System

10. 对抗攻击

对抗攻击是机器学习场景当中，广泛存在的问题，同样也是推荐场景所要面对的问题之一。今年文章如下：不完整及扰动数据攻击推荐系统、基于正则化信息的流形神经网络推荐系统、三元对抗学习在推荐系统中毒攻击中的应用。

• Data Poisoning Attack against Recommender System Using Incomplete and Perturbed Data

• Initialization Matters: Regularizing Manifold-informed Initialization for Neural Recommendation Systems

• Triple Adversarial Learning for Influence based Poisoning Attack in Recommender Systems

11. 计算广告

计算广告与推荐系统场景非常相似，本届KDD计算广告论文方向为策略、广告模型、对抗学习等。

• We Know What You Want: An Advertising Strategy Recommender System for Online Advertising

• Modeling the Sequential Dependence among Audience Multi-step Conversions with Multi-task Learning in Targeted Display Advertising

• Learning to Expand Audience via Meta Hybrid Experts and Critics for Recommendation and Advertising

• A Unified Solution to Constrained Bidding in Online Display Advertising

• Clustering for Private Interest-based Advertising

• Diversity driven Query Rewriting in Search Advertising

• Exploration in Online Advertising Systems with Deep Uncertainty-Aware Learning

• Neural Auction: End-to-End Learning of Auction Mechanisms for E-Commerce Advertising

• Reinforcing Pretrained Models for Generating Attractive Text Advertisements

一起交流

想和你一起学习进步！『NewBeeNLP』目前已经建立了多个不同方向交流群（机器学习 / 深度学习 / 自然语言处理 / 搜索推荐 / 图网络 / 面试交流 / 等），名额有限，赶紧添加下方微信加入一起讨论交流吧！（注意一定o要备注信息才能通过）

- END -

NLP中的范式迁移

2021-10-18

深度学习基础 | RNN家族全面解析

2021-10-19

聊一聊算法工程师复现算法的踩坑总结

2021-10-17

深度学习基础 | 从Language Model到RNN

2021-10-16