[论文解读] Reinforced Anytime Bottom Up Rule Learning for Knowledge Graph Completion

Christian Meilicke, Melisachew Wudage Chekol+2 more|arXiv (Cornell University)|Apr 9, 2020

Advanced Graph Neural Networks参考文献 26被引用 31

一句话总结

本文通过在 AnyBURL 上应用对象身份约束和强化学习来引导自下而上的规则采样，在与现有的最先进的子符号方法相比时表现具有竞争力。

ABSTRACT

Most of todays work on knowledge graph completion is concerned with sub-symbolic approaches that focus on the concept of embedding a given graph in a low dimensional vector space. Against this trend, we propose an approach called AnyBURL that is rooted in the symbolic space. Its core algorithm is based on sampling paths, which are generalized into Horn rules. Previously published results show that the prediction quality of AnyBURL is on the same level as current state of the art with the additional benefit of offering an explanation for the predicted fact. In this paper, we are concerned with two extensions of AnyBURL. Firstly, we change AnyBURLs interpretation of rules from $Θ$-subsumption into $Θ$-subsumption under Object Identity. Secondly, we introduce reinforcement learning to better guide the sampling process. We found out that reinforcement learning helps finding more valuable rules earlier in the search process. We measure the impact of both extensions and compare the resulting approach with current state of the art approaches. Our results show that AnyBURL outperforms most sub-symbolic methods.

研究动机与目标

• 在符号方法与子符号方法并行推进知识图谱补全的研究
• 研究对象身份对规则置信度与冗余的影响
• 引入强化学习以引导 AnyBURL 的自下而上采样过程
• 在标准 KG 基准数据集上评估扩展方法对当前最先进方法的表现

提出的方法

• 通过对知识图谱中路径的自下而上采样学习规则，以形成广义的 Horn 规则
• 应用对象身份以抑制冗余或误导性的规则并调整置信度
• 引入强化学习以使用奖励策略在路径概况之间分配计算资源
• 基于规则支持、置信度和规则长度定义三种奖励策略
• 实现两种决策策略（epsilon-greedy 和加权）以管理概况探索/开发

实验结果

研究问题

• RQ1 将对象身份纳入是否提升 AnyBURL 学到的规则的精确度和可靠性？
• RQ2 强化学习是否能有效引导路径采样过程以更早发现有价值的规则？
• RQ3 在标准数据集上提出的扩展方法相较于最先进的子符号 KG 完整性方法的表现如何？
• RQ4 奖励策略和策略对规则发现效率与质量的影响是什么？

主要发现

阈值	对象身份	规则	减少的规则	命中率@1	命中率@10
s ≥ 100, c ≥ 0.5	off	2004	10.7%	0.739	0.88
s ≥ 100, c ≥ 0.5	on	215	0.938	0.942
s ≥ 10, c ≥ 0.1	off	12832	58.3%	0.765	0.88
s ≥ 10, c ≥ 0.1	on	7475	0.944	0.957

• 对象身份约束显著减少高置信度冗余规则的数量，并在更严格阈值下提升 WN18 的 Hits@1 和 Hits@10
• 强化路径采样更早发现有价值的规则，相较于饱和搜索的做法，预测性能得到提升
• 在常见 KG 基准数据集如 WN18 和 FB15-237 上，所提出扩展的 AnyBURL 比多数子符号方法表现更好