https://zhuanlan.zhihu.com/p/93059665
Sequential and Session-based
Dynamic Item Block and Prediction Enhancing Block for Sequential Recommendation
Item对不同的用户和时间有不同的表示,设计了动态Item模块学习Item的动态表示。并且学习了用户的多维度表示。
- item动态表征
Feature-level Deeper Self-Attention Network for Sequential Recommendation
使用self-attention在item-level和feature-level建模,分别学习item和feature的转移模式。将学习到的表示concat接入全链接进行next item预测。
- self-attention
Graph Contextualized Self-Attention Network for Session-based Recommendation
结合Self-Attention和GNN,弥补Self-Attetion对局部依赖关系捕捉不足的缺点。
- GNN
Sequential and Diverse Recommendation with Long Tail
序列推荐增加aggregate diversity。
- diversity
ISLF: Interest Shift and Latent Factors Combination Model for Session-based Recommendation
在RNN基础上加VAE结构和attention机制。取得state-of-the-art。
- VAE
Sequential Recommender Systems: Challenges, Progress and Prospects
survey
- survey
A Review-Driven Neural Model for Sequential Recommendation
Chenliang
评论驱动的序列推荐,考虑长短期兴趣。
- 评论
领域(视频/新闻)
DeepAPF: Deep Attentive Probabilistic Factorization for Multi-site Video Recommendation
利用用户在多个站点的视频浏览行为,进行视频推荐
- 跨平台
Disparity-preserved Deep Cross-platform Association for Cross-platform Video Recommendation
跨平台视频推荐
- 跨平台
Multi-View Active Learning for Video Recommendation
使用文本进行推荐,但文本确实需要标注,主动学习降低标注成本。最终在分类和推荐任务上取得效果。
- 主动学习
Neural News Recommendation with Attentive Multi-View Learning
user和news两个encoder,考虑muti-view信息
- encoder
强化学习
Reinforced Negative Sampling for Recommendation with Exposure Data Jingtao
学习了一个embedding-based的负样本生成器。为另一个推荐器提供有力的负样本进行pair-wise learning。
- 负样本生成
SLATEQ: A Tractable Decomposition for Reinforcement Learning with Recommendation Sets
推荐一个slate的物品,强化的Action定义在slate上,通过假设简化到item上。转化成q-function的学习和一个choice model,通过松弛和线性规划完成k-set的选择。
- 线性规划
Heterogeneous Information Network
Learning Shared Vertex Representation in Heterogeneous Graphs with Convolutional Networks for Recommendation
构建三个网络,item-item网络,user-item网络,user-subseq网络(n-item subsequences)。使用GCN建模。
- GCN
Unified Embedding Model over Heterogeneous Information Network for Personalized Recommendation
使用全部的meta-path信息去学习一个统一的user和item的embedding。
- meta-path
网络结构
Adaptive User Modeling with Long and Short-Term Preferences for Personalized Recommendatio
构造了新的RNN结构来建模用户序列,使用attention机制融合长短兴趣,在工业和公开数据上取得SOTA。
- RNN改进
BPAM: Recommendation Based on BP Neural Network with Attention Mechanism
使用DNN建模,增加attention机制,降低计算和存储成本,并降低过拟合风险。
- DNN+Attention
CFM: Convolutional Factorization Machines for Context-Aware Recommendation
通过卷积结构,弥补FM的能力的不足,FM的二阶交叉使用外积,构造出一个三维的交互tensor,在其上使用3D卷积
- FM外积+3D卷积
Collaborative Metric Learning with Memory Network for Multi-Relational Recommender Systems
利用多种用户行为,结合Memory Network,能建模细粒度的用户画像。
- Memory Network
Convolutional Gaussian Embeddings for Personalized Recommendation with Uncertainty
使用Gaussian embeddings来描述用户意图的不确定性,在建模时使用了Monte-Carlo采样和卷积神经网络
- Gaussian embeddings
RecoNet: An Interpretable Neural Architecture
for Recommender Systems
在特征级别上有更好的解释性。能够适应冷启动的情况。
- 解释性
PD-GAN: Adversarial Learning for Personalized Diversity-Promoting Recommendation
结合DPP核矩阵生成兼具相关性和多样性的推荐。学习用到了GAN和pair-wise learning。
- GAN
STAR-GCN: Stacked and Reconstructed Graph Convolutional Networks for
Recommender Systems
- GCN
Bundle Recommendation
Matching User with Item Set: Collaborative Bundle Recommendation with Deep Attention Network
Correlation-Sensitive Next-Basket Recommendation
考虑购物篮中商品pair-wise的Correlation。
Explainable Recommendation
Co-Attentive Multi-Task Learning for Explainable Recommendation
多任务,一个任务做推荐,一个任务做解释性,提供语义解释。
- Multi-Task
Explainable Fashion Recommendation: A Semantic Attribute Region Guided Approach
推荐服饰。引入细粒度解释空间,通过两个网络分别将用户和物品映射到空间,完成推荐和解释。
- 解释性空间
Socail Recommendation
Discrete Trust-aware Matrix Factorization for Fast Recommendation
增加可信度约束矩阵来约束用户之间的社会关系,进行推荐。其中用户和物品向量都是零一向量,使用海明距离。
- 约束矩阵
Recommending Links to Maximize the Influence in Social Networks
关注关系推荐。不使用链接数评估节点的影响,而使用对用户观点产生变化来评估影响力,可以通过更少的连接关系和更小的计算量,达到更大的影响力。
Feature Evolution Based Multi-Task Learning for Collaborative Filtering with Social Trust.
通过建模用户偏好向量和用户关系向量进行推荐。同时预测关系概率和点击率。
Deep Adversarial Social Recommendation
通过对抗学习,学习用户在item域和social域的双向关联映射。
- 对抗
Graph Convolutional Networks on User Mobility Heterogeneous Graphs
for Social Relationship Inference
- GCN
POI Recommendation
Geo-ALM: POI Recommendation by Fusing Geographical Information and
Adversarial Learning Mechanism
其他
Hybrid Item-Item Recommendation via Semi-Parametric Embedding
一个item-embedding架构,能够利用side-information,缓解冷启动。
XMind: ZEN - Trial Version