Tutorial on Link Discovery – Algorithms, Approaches and Benchmarks, at ESWC2017, Portoroz, Slovenia
By: Axel-Cyrille Ngonga Ngomo, Irini Fundulaki and Mohamed Ahmed Sherif
Link Discovery is a task of central importance when creating Linked Datasets. Two challenges need to be addressed when carrying out link discovery: The quadratic a-priori runtime complexity of this task demands the development of time-efficient approaches for linking large datasets. Second, the need for accuracy demands the development of generic approaches that can detect correct and complete sets of links. Third, the development of benchmarks that test the ability of instance matching techniques and tools is crucial for identifying and addressing the technical difficulties and challenges in this domain. In this tutorial, we aim to help the audience when faced with all three challenges. First, we will provide an overview of existing solutions for link discovery. Then, we will look into some of the state-of-art algorithms for the rapid execution of link discovery tasks. In particular, we will focus on algorithms which guarantee result completeness. Then, we will present algorithms for the detection of complete and correct sets of links. Here, our focus will be on supervised, active and unsupervised machine learning algorithms. Last, we will discuss existing instance matching benchmarks for Linked Data and We will conclude the tutorial by providing hands-on experience with one of the most commonly used link discovery frameworks, Limes.