.png){kind=icon}
返回
Dynamic negative sampling for recommendation with feature matching
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-023-17521-0.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Negative sampling is a strategy used to select nodes in unexposed items. Recently, most recommendation algorithms based on knowledge graph have adopted learning-effective sampling strategies. However, these methods regard negative sampling as a static process, assuming that the user's potential preferences remain unchanged, and ignoring the dynamic interaction between the user and the recommendation system. To solve the above limitations, we propose a Dynamic Negative Sampling for Recommendation with feature matching (DNSR) to explore high-quality negative examples. This method consists of two main components: the sampler and the recommender. Firstly, the sampler obtains the embedded representation of nodes through a simplified graph convolutional network, then perform node sampling in the collaborative knowledge graph based on the positive examples that users have interacted with. To improve the sampling process, we developed a multiple feature matching technique that calculates the probability distribution of candidate node sets. This technique allows the sampler to filter out representative negative examples. Secondly, the recommender with reinforcement learning aims to optimize the ranking of positive and negative examples by providing feedback in the form of reward signals. This incentivizes the sampler to generate negative examples with information value. On three real-world datasets, our proposed method performs remarkably better than the state-of-the-art baselines. We further verify DNSR model performance of feature matching through ablation experiments.
Keyword:
Recommendation
Knowledge graph
Negative sampling
Reinforcement learning
期刊
IF:
3
论文数:
2.0W
被引数:
3.2W
机构
引用论文
Symbolic dynamics of almost collision orbits and skew products of symplectic maps.png)
Nonlinearity
NonlinearityIF0
PREAI
分享收藏Evaluation of ICAM-1 rs5498 and rs3093030 Polymorphisms in Chinese Patients with Colorectal Cancer评估ICAM-1 rs5498和rs3093030多态性在中国结直肠癌患者中的研究
PREAI
分享收藏分享收藏
查看更多内容