Two papers on recommendation systems are retrieved, one is about improving negative sampling strategy on knowledge graph and the other is a review of adversarial learning on graph. \
Reinforced Negative Sampling over Knowledge Graph for Recommendation
Key words: reinforcement learning + knowledge graph + matrix decomposition
Summary: Handling missing data correctly is a fundamental challenge in recommendation scenarios. Most of the current work involves negative sampling from never-observed data to provide negative samples for the training of recommendation models. However, the existing static or adaptive negative sampling strategies are not enough to produce high-quality negative sampling, and are not conducive to model training and reflecting the actual needs of users.
In this work, he hypothesized that the project knowledge graph (KG) provides a rich relationship between the project and the KG entities and may help to infer information-rich and true negative samples. Therefore, a new negative sampling model, knowledge Graph Strategy network (KGPolicy), is developed, which can be used as a reinforcement learning agent to explore high-quality negative samples. Specifically, by performing the designed exploration operation, it can navigate from the target positive interaction, adaptively receive negative signals with knowledge awareness, and finally generate potential negative samples to train the recommendation system.
Open source code:
Github.com/xiangwang12…
A Survey of Adversarial Learning on Graph
Key words: adversarial learning + graph representation + summary \
Overview: Deep learning models on graphs have achieved excellent performance in various graph analysis tasks, such as node classification, link prediction and graph clustering. However, they expose the uncertainty and unreliability of well-designed inputs (i.e., adversarial samples). As a result, various studies of attack and defense in different graph analysis tasks have emerged, leading to an arms race in graph adversarial learning. For example, if the attacker has poisoning and evading attacks, the defense team has corresponding preprocessing and countervailing methods.
Despite these efforts, there is still a lack of a unified definition of the problem and a comprehensive review. In order to bridge this gap, this paper systematically studies and summarizes the existing work on graph antagonistic learning tasks. Specifically, it investigates and consolidates the attacks and defenses of existing work in graph analysis tasks while providing the correct definitions and taxonomies. In addition, the importance of relevant evaluation indicators is emphasized, and a comprehensive investigation and summary are made.
Open source code:
Github.com/gitgiter/Gr…
Machine Learning Online Manual Deep Learning online Manual AI Basics download (PDF updated to25Set) site QQ group1003271085To join the wechat group, please reply "Add group" to obtain a discount website knowledge planet coupon, copy the link directly open: HTTPS://t.zsxq.com/yFQV7am like the article, click on it
Copy the code