大数据中稳健且可扩展的实体对齐
Robust and Scalable Entity Alignment in Big Data
摘要：
实体对齐在众多不同的科学领域中一直有着重要的用途。特别地，随着诸如社交媒体的通信网络在规模和流行度上的扩展，跨网络匹配实体的概念在社会科学领域中的重要性已经增长。随着大数据的出现，提供大规模图形分析的需求越来越大。然而，由于有数百万个节点和数十亿条边，使用从潜在稀疏或不完整的数据集中提取的特征在无数类似规模的图之间进行对齐的想法变得令人生畏。在本文中，我们将提出一个解决方案，以多步管道的形式大规模对齐的问题。在这个管道中，我们引入了可扩展的特征提取强大的时间属性，伴随着新的和有效的聚类算法，以找到跨图形的相似节点的分组。这些特征及其聚类被输入到一个通用的对齐阶段，该阶段可以在数百万个可能的匹配中准确识别合作伙伴节点。我们的研究结果表明，流水线可以处理大的数据集，实现高效的运行时间内的内存限制。
英文摘要：
Entity alignment has always had significant uses within a multitude of diverse scientific fields. In particular, the concept of matching entities across networks has grown in significance in the world of social science as communicative networks such as social media have expanded in scale and popularity. With the advent of big data, there is a growing need to provide analysis on graphs of massive scale. However, with millions of nodes and billions of edges, the idea of alignment between a myriad of graphs of similar scale using features extracted from potentially sparse or incomplete datasets becomes daunting. In this paper we will propose a solution to the issue of large-scale alignments in the form of a multi-step pipeline. Within this pipeline we introduce scalable feature extraction for robust temporal attributes, accompanied by novel and efficient clustering algorithms in order to find groupings of similar nodes across graphs. The features and their clusters are fed into a versatile alignment stage that accurately identifies partner nodes among millions of possible matches. Our results show that the pipeline can process large data sets, achieving efficient runtimes within the memory constraints.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
