Paper highlights from WSDM 2024 on Recommendation Systems
Notes on the latest work from WSDM 2024 in the area of recommender systems
➡️ Only include details of papers that have preprints available online.
Defense Against Model Extraction Attacks on Recommender Systems
Primer:
Motif-based Prompt Learning for Universal Cross-domain Recommendation
Primer:
To Copy, or not to Copy; That is a Critical Issue of the Output Softmax Layer in Neural Sequential Recommenders
Primer:
Linear Recurrent Units for Sequential Recommendation
Primer:
User Behavior Enriched Temporal Knowledge Graph for Sequential Recommendation
Primer:
CausalMMM: Learning Causal Structure for Marketing Mix Modeling
Primer:
Intent Contrastive Learning with Cross Subsequences for Sequential Recommendation
Primer:
Budgeted Embedding Table For Recommender Systems
Primer:
Pre-trained Recommender Systems: A Causal Debiasing Perspective
Primer:
Dynamic Sparse Learning: A Novel Paradigm for Efficient Recommendation
Primer:
PEACE: Prototype lEarning Augmented transferable framework for Cross-domain rEcommendation
Primer:
Collaboration and Transition: Distilling Item Transitions into Multi-Query Self-Attention for Sequential Recommendation
Primer:
Unified Pretraining for Recommendation via Task Hypergraphs
Primer:
SSLRec: A Self-Supervised Learning Library for Recommendation
Primer:
Multi-Sequence Attentive User Representation Learning for Side-information Integrated Sequential Recommendation
Primer:
LabelCraft: Empowering Short Video Recommendations with Automated Label Crafting
Primer:
MONET: Modality-Embracing Graph Convolutional Network and Target-Aware Attention for Multimedia Recommendation
Primer:
Debiasing Sequential Recommenders through Distributionally Robust Optimization over System Exposure
Primer:
Knowledge Graph Diffusion Model for Recommendation
Primer:
Large Language Models for Data Aumgnetation in Recommendation - Current preprint title is LLMRec: Large Language Models with Graph Augmentation for Recommendation
Primer:
Leveraging Multimodal Features and Item-level User Feedback for Bundle Construction
Primer:
Interact with the Explanations: Causal Debiased Explainable Recommendation System
Primer:
Global Heterogeneous Graph and Target Interest Denoising for Multi-behavior Sequential Recommendation
Primer:
MultiFS: Automated Multi-Scenario Feature Selection in Deep Recommender Systems
Primer:
MADM: A Model-agnostic Denoising Module for Graph-based Social Recommendation
Primer:
CDRNP: Cross-Domain Recommendation to Cold-Start Users via Neural Process
Primer:
Inverse Learning with Extremely Sparse Feedback for Recommendation
Primer:
Contextual MAB Oriented Embedding Denoising for Sequential Recommendation
Primer:
Mixed Attention Network for Cross-domain Sequential Recommendation
Primer:
Knowledge Graph Context-Enhanced Diversified Recommendation
Primer:
Exploring Adapter-based Transfer Learning for Recommender Systems: Empirical Studies and Practical Insights
Primer:
Diff-MSR: A Diffusion Model Enhanced Paradigm for Cold-Start Multi-Scenario Recommendation
Primer:
AutoPooling: Automated Pooling Search for Multi-valued Features in Recommendations
Primer:
C^2DR: Robust Cross-Domain Recommendation based on Causal Disentanglement
Primer:
RecJPQ: Training Large-Catalogue Sequential Recommenders
Primer:
On the Effectiveness of Unlearning in Session-Based Recommendation
Primer:
Proxy-based Item Representation for Attribute and Context-aware Recommendation
Primer:
IncMSR: An Incremental Learning Approach for Multi-Scenario Recommendation
Primer:
Deep Evolutional Instant Interest Network for CTR Prediction in Trigger-Induced Recommendation
Primer:
User Consented Federated Recommender System Against Personalized Attribute Inference Attack
Primer:
Neural Kalman Filtering for Robust Temporal Recommendation
Primer:
ONCE: Boosting Content-based Recommendation with Both Open- and Closed-source Large Language Models
Primer: Addressing the limitations of existing content-based recommender systems, this paper presents the ONCE framework, which leverages both open- and closed-source large language models (LLMs) to significantly enhance recommendation performance. Their findings demonstrate that combining finetuning on open-source LLMs with prompting-based data augmentation on closed-source models yields substantial improvements, with relative gains reaching up to 19.32% compared to state-of-the-art models. These results highlight the immense potential of LLMs in content-based recommendation and hold significant implications for online content platforms. Notably, the ONCE framework extends beyond news and book recommendation, demonstrating its applicability to diverse domains.