One Model to Serve All: Star Topology Adaptive Recommender for Multi-Domain CTR Prediction【applied paper,用于多领域CTR预估的自适应推荐】
Click-Through Rate Prediction with Multi-Modal Hypergraphs(金星)
CBML: A Cluster-based Meta-learning Model for Session-based Recommendation
1.操作数 有限, 冷启动, 难以捕获动态偏好.
2.元学习借助其他用户解决冷启动. 但是, 当前用户行为方面任然很弱.
Self-Supervised Graph Co-Training for Session-based Recommendation
1.在会话推荐的场景下,通过从Item和Session两个视角分别建立两个同质图,代表两种Level。
2.Item Encoder可生成Item Embedding和Session Embedding,Session Encoder只生成Session Embedding.
3.两个Encoder进行Co-Training[参考46,6,54],具体是其中一Encoder训练完后,依据其训练结果选出正负样本(这个称之为伪标签)传递给另一Encoder,迭代地更新两个Encoder同时生成更富含信息的样本,这也是本文自监督学习的过程
4.最大化的是一个session里last-item和该session预测的next-item之间的互信息
为了避免在多轮迭代之后,两个Encoder生成的Embedding过于相似,文中通过设置Divergence Constraint(差异约束),具体来说是增加了对抗性扰动样本,本文认为如果其中一Encoder能够抵抗(可以认为是识别出)来自另一Encoder的对抗扰动信号,那么他们就能做到不相似。
Social Recommendation with Self-Supervised Metagraph Informax Network
Expanding Relationship for Cross Domain Recommendation
在增强的图上,ReCDR采用了分层注意机制
Low-dimensional Alignment for Cross-Domain Recommendation
提醒我们可以用低维这个点
Double-Scale Self-Supervised Hypergraph Learning for Group Recommendation:
开发了基于用户和组级超图的分层超图卷积网络,以模拟组内和组外用户之间的复杂元组相关性。
Xpl-CF: Explainable Embeddings for Feature-based Collaborative Filtering:
CMML: Contextual Modulation Meta Learning for Cold-Start Recommendation
学习模型的初始参数来解决这个问题, 从而允许从有限的数据示例中快速适应特定的任务
Contrastive Curriculum Learning for Sequential User Behavior Modeling via Data Augmentation:
对比课程学习框架,用于生成有效的表示,用于建模连续用户行为
设计了基于模型的数据生成器: 给定一个目标用户,它可以利用融合的属性语义生成更接近真实的序列
Graph Structure Aware Contrastive Knowledge Distillation for Incremental Learning in Recommender Systems:
We combine the contrastive distillation formulation with intermediate layer distillation to inject layer-level supervision. (对比学习+蒸馏学习)
Distilling Knowledge from BERT into Simple Fully Connected Neural Networks for Efficient Vertical Retrieval:
老师模型为Bert,学生模型为bi-lstm
Adversarial Domain Adaptation for Cross-lingual Information Retrieval with Multilingual BERT:
通过对抗网络将不同语言的标记嵌入对齐,以帮助语言模型学习跨语言句子表示。
Supervised Contrastive Learning for Multimodal Unreliable News Detection in COVID-19 Pandemic:
1.但这种表示不能推广到只有少量镜头标记样本的不可见新类
https://www.bilibili.com/video/BV1wv411E7Jz/
1.一种是将最后一次点击视为表示用户当前偏好的查询向量,另一种是考虑会话内的所有项目都有利于最终结果,包括无关项目的影响(即,虚假用户行为)
本文提出了一种基于分解对抗学习和正交正则化的公平新闻推荐方法,可以缓解敏感用户属性带来的新闻推荐不公平
1.本文提出了一种具有公平性意识的新闻推荐方法,该方法采用分解对抗学习和正交正则化的方法,可以缓解敏感用户属性偏差带来的新闻推荐的不公平性。
4.我们还提出了一种正交正则化方法,以鼓励无偏差用户嵌入与有偏差感知的嵌入正交,从而更好地区分无偏差用户嵌入和有偏差感知的嵌入。
1.我们提出了无创性自我注意机制(NOVA)来有效地利用BERT框架下的边信息。
2.NOVA利用辅助信息来产生更好的注意力分布,而不是直接改变项目嵌入,这可能会导致信息泛滥。
3.在我们的试点实验中,我们发现一些简单的方法,直接将各种类型的辅助信息融合到项目嵌入中,通常带来很少甚至负面的影响。
Real Negatives Matter: Continuous Training with Real Negatives for Delayed Feedback Modeling(Alibaba)
Learning Reliable User Representations from Volatile and Sparse Data to Accurately Predict Customer Lifetime Value(腾讯)
利用历史收入时间序列和用户属性分别学习时间和结构用户表示
3.这种融合方式可以被视为低通表示空间中时间和结构表示的关联,这也有助于防止数据噪声在不同视图之间传输
Debiasing Learning based Cross-domain Recommendation(Alibaba)
开发了一种基于偶然性的方法,以减轻跨域传输用户信息时的域偏见。
Adversarial Feature Translation for Multi-domain Recommendation(腾讯)
多域推荐(MDR)是为了同时改进所有推荐域而提出的,其关键是从所有域中获取信息性的特定于域的特征。为了解决这个问题,我们提出了一种新的MDR对抗性特征转换(AFT)模型,该模型学习不同领域之间的特征转换,在生成性对抗网络框架下。
Contrastive Learning for Debiased Candidate Generation in Large-Scale Recommender Systems(Alibaba)
1.在本文中,我们从理论上证明了一个流行的对比损失选择相当于通过反向倾向加权来减少暴露偏差,这为理解对比学习的有效性提供了一个新的视角。
2.一种对比学习方法,用于在具有超大候选容量的推荐系统中提高DCG的公平性、有效性和效率
Debiasing Learning based Cross-domain Recommendation(Alibaba)
Signed Graph Neural Network with Latent Groups(美团)
1.提出了超越平衡理论假设的群符号图神经网络(GS-GNN)符号图表示学习模型。
2.采用了一个更为普遍的假设,即节点可以被划分为多个潜在组,并且这些组可以具有任意关系,并基于该假设提出了一种新的基于原型的GNN来学习节点表示
DMBGN: Deep Multi-Behavior Graph Networks for Voucher Redemption Rate Prediction(Alibaba)
MugRep: A Multi-Task Hierarchical Graph Representation Learning Framework for Real Estate Appraisal(百度)
1.通过获取和整合多源城市数据,我们首先构建丰富的特征集,从多个角度(例如地理分布、人员流动性分布和居民人口分布)全面分析房地产
2.提出了一个演化的房地产交易图和相应的事件图卷积模块,将房地产交易之间的时空依赖性异步地结合起来
3.为了从居住社区的角度进一步整合有价值的知识,我们设计了一个分层异构社区图卷积模块来捕获居住社区之间的各种相关性
4.引入了一个城区分区多任务学习模块,生成不同分布的房地产价值观
Understanding and Improving Fairness-Accuracy Trade-offs in Multi-Task Learning(Google)
1.在多任务学习中,多个任务被联合学习,以利用任务相关性进行更有效的归纳迁移。这呈现了一个多维帕累托边界:(1)每个任务的组公平性和准确性之间的权衡,以及(2)多个任务之间的权衡。
2.更深入地理解群体公平性与多任务学习准确性之间的相互作用,并且我们表明,主要关注优化多任务准确性帕累托边界的传统方法可能无法很好地实现公平性目标
3.提出了一组新的度量标准,以更好地捕捉在多任务学习环境中唯一呈现的公平-准确性权衡的多维帕累托前沿
4.提出了一种多任务感知公平性(MTA-F)方法来提高多任务学习的公平性
SEMI: A Sequential Multi-Modal Information Transfer Network for E-Commerce Micro-Video Recommendations(Alibaba)
1.我们设计了一个顺序多模式信息传输网络(SEMI),它利用产品域用户行为来辅助微视频推荐。
2.跨域对比学习(CCL)算法来预训练序列编码器,以模拟用户在这两个域中的序列行为。CCL的目标是最大化不同领域之间的互信息下限。
Curriculum Meta-Learning for Next POI Recommendation
我们提出了一个新的课程硬度感知元学习(CHAML)框架,该框架将硬样本挖掘和课程学习整合到元学习范式中。
Topology Distillation for Recommender System
1.采用了知识提取(knowledge Destruction)技术,这是一种模型压缩技术,它利用预先训练过的大型教师模型传递的知识来训练紧凑的学生模型。
3.简单地让学生学习整个拓扑结构并不总是有效的,甚至会降低学生的表现。我们证明,由于学生的能力与教师相比非常有限,因此学习整个拓扑结构对学生来说是令人望而生畏的。为了解决这个问题,我们提出了一种新的分层拓扑提取(HTD)方法,该方法通过对拓扑进行分层提取来应对大的容量缺口。
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
Modeling the Sequential Dependence among Audience Multi-step Conversions with Multi-task Learning in Targeted Display Advertising
1.提出了一个自适应信息传输多任务(AITM)框架,该框架通过自适应信息传输(AIT)模块对受众多步转换之间的顺序依赖性进行建模
2.可以自适应地了解在不同转换阶段传输什么信息以及传输多少信息
Learning to Expand Audience via Meta Hybrid Experts and Critics for Recommendation and Advertising
Adversarial Feature Translation for Multi-domain Recommendation
1.多域推荐是为了同时改进所有推荐域而提出的,其关键是从所有推荐域中获取特定于领域的信息特征。
2.提出了一种新的MDR对抗性特征转换(AFT)模型,该模型在生成的对抗性网络框架下学习不同领域之间的特征转换。
3.在多域生成器中,我们提出了一种特定于域的屏蔽编码器来突出域间的特征交互,然后通过转换器和特定于域的注意来聚合这些特征。
4.在多领域鉴别器中,受知识表示学习的启发,我们通过两步特征转换明确地建模了项目、领域和用户的通用/领域特定表示之间的关系。
5.我们通过两步特征转换明确地建模了项目、领域和用户的通用/领域特定表示之间的关系
Socially-Aware Self-Supervised Tri-Training for Recommendation
1.以往对比学习: 在两个不同视图中的节点之间建立的,这意味着忽略了来自其他节点的自我监控信号。
Model-Agnostic Counterfactual Reasoning for Eliminating Popularity Bias in Recommender System
A General Method For Automatic Discovery of Powerful Interactions In Click-Through Rate Prediction
Learning Graph Meta Embeddings for Cold-Start Ads in Click-Through Rate Prediction
Explicit Semantic Cross Feature Learning via Pre-trained Graph Neural Networks for CTR Prediction
Enhanced Graph Learning for Collaborative Filtering via Mutual Information Maximization
Dual Attention Transfer in Session-based Recommendation with Multi Dimensional Integration
Package Recommendation with Intra- and Inter-Package Attention Networks
The World is Binary: Contrastive Learning for Denoising Next Basket Recommendation
Lighter and Better: Low-Rank Decomposed Self-Attention Networks for Next-Item Recommendation
Joint Knowledge Pruning and Recurrent Graph Convolution for News Recommendation
Empowering News Recommendation with Pre-trained Language Models
AMM: Attentive Multi-field Matching for News Recommendation
RMBERT: News Recommendation via Recurrent Reasoning Memory Network over BERT
Federated Collaborative Transfer for Cross-Domain Recommendation
Learning Domain Semantics and Cross-Domain Correlations for Paper Recommendation
1.研究跨学科语义相关性??
2.跨学科的语义相关性由影响函数、相关性度量和排序机制表示: 感觉又不是很厉害, 又有很高的时间复杂度??
3.top-k related cross-domain??
4.(无用)
Graph Meta Network for Multi-Behavior Recommendation with Interaction Heterogeneity and Diversity
Transfer-Meta Framework for Cross-domain Recommendation to Cold-Start Users
ConsisRec: Enhancing GNN for Social Recommendation via Consistent Neighbor Aggregation
Structured Graph Convolutional Networks with Stochastic Masks for Recommender Systems
Self-supervised Graph Learning for Recommendation
Variational Autoencoders for Top-K Recommendation with Implicit Feedback
Info-flow Enhanced GANs for Recommender
Learning Recommender Systems with Implicit Feedback via Soft Target Enhancement
Learning Graph Meta Embeddings for Cold-Start Ads in Click-Through Rate Prediction
FORM: Follow the Online Regularized Meta-Leader for Cold-Start Recommendation
Decoupling Representation and Regressor for Long-Tailed Information Cascade Prediction
(没找到, 放弃治疗)
Evaluation measures based on preference graphs
Retrieving Complex Tables with Multi-Granular Graph Representation Learning
Meta-Inductive Node Classification across Graphs
WGCN: Graph Convolutional Networks with Weighted Structural Features
1.感觉就是GCN的过程加了 有向图 出度和入度
2.(对我们应该没用)
GilBERT: Generative Vision-Language Pre-Training for Modality-Incomplete Visual-Linguistic Tasks
1.任务是图像文本检索: 图像-文本检索中的自然模态不完全性
2.(感觉我们应该用不到)
Hierarchical Cross-Modal Graph Consistency Learning for Video-Text Retrieval
(没找到, 放弃治疗)
Self-Supervised Learning of Contextual Embeddings for Link Prediction in Heterogeneous Networks
SUGAR: Subgraph Neural Network with Reinforcement Pooling and Self-Supervised Mutual Information Mechanism
On the Equivalence of Decoupled Graph Convolution Network and Label Propagation
Rumor Detection with Field of Linear and Non-Linear Propagation
Multi-level Connection Enhanced Representation Learning for Script Event Prediction
Self-Supervised Multi-Channel Hypergraph Convolutional Network for Social Recommendation
Bidirectional Distillation for Top-K Recommender System
Learning Heterogeneous Temporal Patterns of User Preference for Timely Recommendation
High-dimensional Sparse Embeddings for Collaborative Filtering
Rabbit Holes and Taste Distortion: Distribution-Aware Recommendation with Evolving Interests
User Simulation via Supervised Generative Adversarial Network
TG-GAN: Continuous-time Temporal Graph Deep Generative Models with Time-Validity Constraints
Curriculum CycleGAN for Textual Sentiment Domain Adaptation with Multiple Sources
Completing Missing Prevalence Rates for Multiple Chronic Diseases by Jointly Leveraging Both Intra- and Inter-Disease Population Health Data Correlations
Self-Supervised Multi-Channel Hypergraph ConvolutionalNetwork for Social Recommendation
1.超图可以提供一种自然的方式来对高阶关系进行建模。
2.通过利用高阶用户关系提出了多频道超图卷积网络来加强社交推荐。
EX3: Explainable Attribute-aware Item-set Recommendations
Towards Source-Aligned Variational Models for Cross-Domain Recommendation
Semi-Supervised Visual Representation Learning for Fashion Compatibility
Tops, Bottoms, and Shoes: Building Capsule Wardrobes via Cross-Attention Tensor Network
Matrix Factorization for Collaborative Filtering Is Just Solving an Adjoint Latent Dirichlet Allocation Model After All
Together is Better: Hybrid Recommendations Combining Graph Embeddings and Contextualized Word Representations