SIGIR2020推荐系统论文聚焦

2020 年 6 月 9 日 机器学习与推荐算法

前言

第43届国际信息检索研究和发展大会（SIGIR）将于2020年7月25-30日在美丽的中国西安举行。此次大会共收到了555篇长文投稿，录用147篇，长文录取率26.4%；共收到了507篇短文投稿，录用153篇，短文录取率30%。

正因为推荐与搜索是解决信息过载的两种有效途径，因此虽然是关于检索的会议，但通过下图可以看出推荐（Recommendation）占据了很大比例，与搜索（Search+Retrieval）不相上下。另外，图与网络（Graph/Network）数据成为研究的主要对象，毕竟许多待研究的对象都可以表示为图。值得注意的是，神经网络（Neural）仍然排在前列；融合知识（Knowledge）的搜索/推荐系统也被许多研究者研究。除此之外，强化学习也出现在了排行榜中，可见利用强化学习的思想来迭代优化搜索/推荐逐渐成为流行。

另外，还注意到今年SIGIR开办了一场关于对话推荐/检索的Tutorial，感兴趣的小伙伴可以多多关注。想提前了解对话推荐（Conversational RS，CRS）的朋友，可以公众号后台回复【CRS】获取对话推荐系统最新综述。

推荐论文列表

本次只对大会的长文（Full Papers）进行梳理，因此共整理出63篇关于推荐系统的论文。为了方便查看与了解，我们主要将其分为了以下几类：Sequential RS，Graph-based RS，Cold-start in RS，Efficient RS，Knowledge-aware RS，Robust RS，Group RS，Conversational RS，RL for RS，Cross-domain RS，Explainable RS，POI RS。另外，对于有一些不包含在以上类别的文章，我们统一归为了Others。当然，以上分类仁者见仁，智者见智，目的是给大家一个相对清晰的结构。具体的各个类别所包含的论文数见下表。

分类	数量
Sequential RS	8
Graph-based RS	6
Robust RS	6
Efficient RS	5
Knowledge-aware RS	5
Cold-start in RS	4
Group RS	4
Conversational RS	4
RL for RS	3
Cross-domain RS	2
Explainable RS	2
POI RS	1
Others	13

可见，序列化推荐的文章占比较大；随后是基于图的推荐、鲁棒的推荐系统；其次是提升推荐效率的文章、基于知识的推荐以及解决冷启动问题的推荐文章、组推荐、对话推荐系统；最后是强化学习推荐、跨域推荐、可解释推荐以及兴趣点推荐。当然其他类别中也包含了许多有意思的研究，比如消除推荐偏置（Bias）的文章、分布式训练推荐系统的文章以及如何retrain推荐系统的文章等。

接下来是分类好的推荐论文列表，大家可以根据自己的研究子方向进行精读。

Sequential RS

Incorporating User Micro-behaviors and Item Knowledge into Multi-task Learning for Session-based Recommendation.
GAG: Global Attributed Graph Neural Network for Streaming Session-based Recommendation.
Sequential Recommendation with Self-attentive Multi-adversarial Network.
A General Network Compression Framework for Sequential Recommender Systems.
Next-item Recommendation with Sequential Hypergraphs.
KERL: A Knowledge-Guided Reinforcement Learning Model for Sequential Recommendation.
Time Matters: Sequential Recommendation with Complex Temporal Information.
Modeling Personalized Item Frequency Information for Next-basket Recommendation.

Graph-based RS

Learning to Transfer Graph Embeddings for Inductive Graph based Recommendation.
Joint Item Recommendation and Attribute Inference: An Adaptive Graph Convolutional Network Approach.
Multi-behavior Recommendation with Graph Convolution Networks.
Hierarchical Fashion Graph Network for Personalised Outfit Recommendation.
Neighbor Interaction Aware Graph Convolution Networks for Recommendation.
Disentangled Representations for Graph-based Collaborative Filtering.

Cold-start RS

Content-aware Neural Hashing for Cold-start Recommendation.
Recommending Podcasts for Cold-Start Users Based on Music Listening and Taste.
Recommendation for New Users and New Items via Randomized Training and Mixture-of-Experts Transformation.
AR-CF: Augmenting Virtual Users and Items in Collaborative Filtering for Addressing Cold-Start Problems.

Efficient RS

Lightening Graph Convolution Network for Recommendation.
A General Knowledge Distillation Framework for Counterfactual Recommendation via Uniform Data.
Beyond User Embedding Matrix: Learning to Hash for Modeling Large-Scale Users in Recommendation.
Parameter-Efficient Transfer from Sequential Behaviors for User Modeling and Recommendation.
Online Collective Matrix Factorization Hashing for Large-Scale Cross-Media Retrieval.

Knowledge-aware RS

Jointly Non-Sampling Learning for Knowledge Graph Enhanced Recommendation.
Fairness-Aware Explainable Recommendation over Knowledge Graphs.
Attentional Graph Convolutional Networks for Knowledge Concept Recommendation in MOOCs in a Heterogeneous View.
Make It a CHORUS: Context- and Knowledge-aware Item Modeling for Recommendation.
CKAN: Collaborative Knowledge-aware Attentive Network for Recommender Systems.

Robust RS

How Dataset Characteristics Affect the Robustness of Collaborative Recommendation Models.
GCN-Based User Representation Learning for Unifying Robust Recommendation and Fraudster Identification.
How Dataset Characteristics Affect the Robustness of Collaborative Recommendation Models.
Certifiable Robustness to Discrete Adversarial Perturbations for Factorization Machines.
DPLCF: Differentially Private Local Collaborative Filtering.
Data Poisoning Attacks against Differentially Private Recommender Systems.

Group RS

GAME: Learning Graphical and Attentive Multi-view Embeddings for Occasional Group Recommendation.
GroupIM: A Mutual Information Maximizing Framework for Neural Group Recommendation.
Group-Aware Long- and Short-Term Graph Representation Learning for Sequential Group Recommendation.
Global Context Enhanced Graph Nerual Networks for Session-based Recommendation.

Conversational RS

Deep Critiquing for VAE-based Recommender Systems.
Interactive Recommender System via Knowledge Graph-enhanced Reinforcement Learning.
Towards Question-based Recommender Systems.
Neural Interactive Collaborative Filtering.

RL for RS

Self-Supervised Reinforcement Learning for Recommender Systems.
MaHRL: Multi-goals Abstraction based Deep Hierarchical Reinforcement Learning for Recommendations.
Leveraging Demonstrations for Reinforcement Recommendation Reasoning over Knowledge Graphs.

Cross-domain RS

Transfer Learning via Contextual Invariants for One-to-Many Cross-Domain Recommendation.
CATN: Cross-Domain Recommendation for Cold-Start Users via Aspect Transfer Network.

Explainable RS

Measuring Recommendation Explanation Quality: The Conflicting Goals of Explanations.
Try This Instead: Personalized and Interpretable Substitute Recommendation.

POI RS

HME: A Hyperbolic Metric Embedding Approach for Next-POI Recommendation.

Others

Learning Personalized Risk Preferences for Recommendation.
Octopus: Comprehensive and Elastic User Representation for the Generation of Recommendation Candidates.
Spatial Object Recommendation with Hints: When Spatial Granularity Matters.
Agreement and Disagreement between True and False-Positive Metrics in Recommender Systems Evaluation.
Distributed Equivalent Substitution Training for Large-Scale Recommender Systems.
The Impact of More Transparent Interfaces on Behavior in Personalized Recommendation.
MVIN: Learning multiview items for recommendation.
How to Retrain a Recommender System?
Measuring and Mitigating Item Under-Recommendation Bias in Personalized Ranking Systems.
BiANE: Bipartite Attributed Network Embedding.
ASiNE: Adversarial Signed Network Embedding.
Learning Dynamic Node Representations with Graph Neural Networks.
Asymmetric Tri-training for Debiasing Missing-Not-At-Random Explicit Feedback.