다중 모달 추천을 위한 인과 관계 기반 공정한 표현 학습
다중 모달 콘텐츠의 복잡한 인과 관계로 인해 사용자의 민감한 정보가 누출될 수 있으며, 이는 공정한 표현 학습에 위협이 된다. 본 연구는 인과 관계 기반의 모달 분리와 관계 인식 공정성 학습을 통해 다중 모달 추천의 공정성을 향상시킨다.
Mitigating Sensitive Information Leakage in Multimodal Recommendation through Causality-Inspired Fair Representation Learning
Multimodal recommendations can lead to increased leakage of users' sensitive information due to entangled causal relationships. This work proposes a novel fair multimodal recommendation approach (FMMRec) that addresses this issue through causality-inspired fairness-oriented modal disentanglement and relation-aware fairness learning.
Harnessing Multimodal Large Language Models to Enhance Personalized Sequential Recommendations
Leveraging the capabilities of Multimodal Large Language Models (MLLMs) to effectively integrate multimodal data and capture the dynamic evolution of user preferences, thereby enhancing the accuracy and interpretability of sequential recommendations.
A Multimodal Single-Branch Embedding Network for Effective Recommendations in Cold-Start and Missing Modality Scenarios
A novel multimodal recommender system, SiBraR, leverages a single-branch embedding network to effectively combine collaborative and content information, leading to improved recommendations in cold-start and missing modality scenarios.
Efficient Multimodal Representation Learning for Scalable Recommendation
A lightweight framework called full-scale Matryoshka Representation Learning for Recommendation (fMRLRec) that captures item features at different granularities, enabling efficient multimodal recommendation across multiple model sizes.
Comprehensive Benchmarking of Multimodal Recommendation Systems with Advanced Feature Extractors
This work presents a comprehensive benchmarking study on multimodal recommendation systems, focusing on the impact of advanced multimodal feature extractors on recommendation performance.
End-to-end Training of Multimodal and Ranking Models for Improved Recommendation Performance
An industrial multimodal recommendation framework named EM3 that sufficiently utilizes multimodal information and allows personalized ranking tasks to directly train the core modules in the multimodal model, obtaining more task-oriented content representations without overburdening resource consumption.
Addressing Missing Modalities in Multimodal Recommendation: A Feature Propagation-based Approach
The core message of this paper is to propose a feature propagation-based approach, called FeatProp, to address the problem of missing modalities in multimodal recommendation.
An Aligning and Training Framework for Multimodal Recommendations
AlignRec proposes a solution to the misalignment issue in multimodal recommendations, integrating three alignments within its framework to improve performance.
© 2024 by Linnk AI