ICCS 2021 Tutorials

Since the beginning of the 2000s, Knowledge Graphs have been widely used for modeling various domains ranging from linguistics to biomedicine. Recently, Knowledge Graphs have become even more crucial for improving diverse real-world applications at the intersection of Natural Language Processing (NLP) and Knowledge Management, such as question answering, named entity disambiguation, information extraction, etc. Raising awareness about Knowledge Graphs in other research communities will allow them to benefit from the versatile Knowledge Graph for- malisms, methods, and tools. To this end, this tutorial focuses on the foundations of Knowledge Graphs. Starting from basic notions and techniques of Knowledge Graphs, the tutorial will then move on to more advanced topics such as how logical reasoning over these Knowledge Graphs, where formally specified background knowledge is taken into account to enrich the explicitly stated information by facts that can be logically inferred. Furthermore, we will discuss how to express real-world aspects such as context, time, and uncertainty in the Knowledge Graph framework. As they are typically used in an open-world setting, Knowledge Graphs can almost never be assumed to be complete, i.e., some information will typically be missing. In order to address this problem, different Knowledge Graph embedding models have been pro- posed for automated Knowledge Graph completion. These models are mostly based on the tasks such as link prediction, triple classification, and entity classification/typing. This tutorial will also target the topic of Knowledge Graph embedding techniques. Finally, various applications of Knowledge Graphs and Knowledge Graph embeddings will be discussed.

In this tutorial, we will explore traditional and contemporary approaches to concept representation and reasoning. Our treatment will be pragmatic and focus on practical application. Most theories can be seen as a response to the Classical Theory [1], thus, we will begin with classical approaches that model concepts as (complete) definitional structures; we will show how this has developed into less strict approaches often referred to as ‘neo-classical’, where concepts are modelled as partial structures with conditions of necessity. Contemporary approaches will include geometric approaches [7], conceptual blending [10], reasoning by analogy and metaphor [15], argumentation structures [11], and prototypes and family resemblance, where concepts are represented by similarity to so-called exemplars [9]. In each case, using examples to support, key notions will be outlined, benefits and limitations summarised, and how each addresses shortcomings and criticisms of the Classical Theory will be highlighted. We shall close the tutorial by examining the complementary aspects of the approaches reviewed, with an intention of exploring how these may be used in combination.