Ebook: Knowledge Graphs for eXplainable Artificial Intelligence: Foundations, Applications and Challenges

Explanations have been the subject of study in a variety of fields for a long time, and are experiencing a new wave of popularity due to the latest advancements in Artificial Intelligence (AI). While machine and deep learning systems are now widely adopted for decision making, they also revealed a major drawback, namely the inability to explain their decisions in a way that humans can easily understand them. As a result, eXplainable AI (XAI) rapidly became an active area of research in response to the need of improving the understandability and trustworthiness of modern AI systems – a crucial aspect for their adoption at large scale, and particularly in life-critical contexts.

The field of Knowledge Representation and Reasoning (KRR), on the other hand, has a long standing tradition in managing structured knowledge, i.e. modeling, creating, standardising, publishing and sharing information in symbolic form. KRR methods and technologies developed over the years result by now in large amounts of structured knowledge (in the form of ontologies, knowledge graphs, and other structured representations), that are not only machine-readable and in standard formats, but also openly available, and covering a variety domains at large scale. These structured sources, designed to capture causation as opposed to correlation in Machine Learning methods, could therefore be exploited as sources of background knowledge by eXplainable AI methods in order to build more insightful, trustworthy explanations.

This book provides the very first comprehensive collection of research contributions on the role of knowledge graphs for eXplainable AI (KG4XAI). We gather studies using KRR as a framework to enable intelligent systems to explain their decisions in a more understandable way, presenting academic and industrial research focused on the theory, methods and implementations of AI systems that use structured knowledge to generate reliable explanations.

We include both introductory material on knowledge graphs for readers with only a minimal background in the field, and advanced specific chapters devoted to methods, applications and case-studies using knowledge graphs as a part of knowledge-based, explainable systems (KBX-systems). The final chapters convey current challenges and future directions of research in the area of knowledge graphs for eXplainable AI.

Our goal is not only to provide a scholarly, state-of-the-art overview of research in this field, but also to foster the hybrid combination of symbolic and and subsymbolic AI methods, motivated by the complementary strengths and limitations of both the field of KRR and Machine Learning.

The editors would like to thank all contributing authors for their efforts in making this book possible.

March 2020

Ilaria Tiddi

Freddy Lécu’e

Pascal Hitzler

Knowledge Graphs are an emerging form of knowledge representation. While Google coined the term Knowledge Graph first and promoted it as a means to improve their search results, they are used in many applications today. In a knowledge graph, entities in the real world and/or a business domain (e.g., people, places, or events) are represented as nodes, which are connected by edges representing the relations between those entities. While companies such as Google, Microsoft, and Facebook have their own, non-public knowledge graphs, there is also a larger body of publicly available knowledge graphs, such as DBpedia or Wikidata. In this chapter, we provide an overview and comparison of those publicly available knowledge graphs, and given insights into their contents, size, coverage, and overlap.

Explainability has been an important goal since the early days of Artificial Intelligence. Several approaches for producing explanations have been developed. However, many of these approaches were tightly coupled with the capabilities of the artificial intelligence systems at the time. With the proliferation of AI-enabled systems in sometimes critical settings, there is a need for them to be explainable to end-users and decision-makers. We present a historical overview of explainable artificial intelligence systems, with a focus on knowledge-enabled systems, spanning the expert systems, cognitive assistants, semantic applications, and machine learning domains. Additionally, borrowing from the strengths of past approaches and identifying gaps needed to make explanations user- and context-focused, we propose new definitions for explanations and explainable knowledge-enabled systems.

Knowledge graph embeddings are now a widely adopted approach to knowledge representation in which entities and relationships are embedded in vector spaces. In this chapter, we introduce the reader to the concept of knowledge graph embeddings by explaining what they are, how they can be generated and how they can be evaluated. We summarize the state-of-the-art in this field by describing the approaches that have been introduced to represent knowledge in the vector space. In relation to knowledge representation, we consider the problem of explainability, and discuss models and methods for explaining predictions obtained via knowledge graph embeddings.

The growing interest in making use of Knowledge Graphs for developing explainable artificial intelligence, there is an increasing need for a comparable and repeatable comparison of the performance of Knowledge Graph-based systems. History in computer science has shown that a main driver to scientific advances, and in fact a core element of the scientific method as a whole, is the provision of benchmarks to make progress measurable. Benchmarks have several purposes: (1) they highlight weak and strong points of systems, (2) they stimulate technical progress and (3) they make technology viable. Benchmarks are an essential part of the scientific method as they allow to track the advancements in an area over time and make competing systems comparable. This chapter gives an overview of benchmarks used to evaluate systems that process Knowledge Graphs.

Recommender systems are everywhere, from e-commerce to streaming platforms. They help users lost in the maze of available information, items and services to find their way. Among them, over the years, approaches based on machine learning techniques have shown particularly good performance for top-N recommendations engines. Unfortunately, they mostly behave as black-boxes and, even when they embed some form of description about the items to recommend, after the training phase they move such descriptions in a latent space thus loosing the actual explicit semantics of recommended items. As a consequence, the system designers struggle at providing satisfying explanations to the recommendation list provided to the end user. In this chapter, we describe two approaches to recommendation which make use of the semantics encoded in a knowledge graph to train interpretable models which keep the original semantics of the items description thus providing a powerful tool to automatically compute explainable results. The two methods relies on two completely different machine learning algorithms, namely, factorization machines and autoencoder neural networks. We also show how to measure the interpretability of the model through the introduction of two metrics: semantic accuracy and robustness.

Reasoning with knowledge expressed in natural language and Knowledge Bases (KBs) is a major challenge for Artificial Intelligence, with applications in machine reading, dialogue, and question answering. General neural architectures that jointly learn representations and transformations of text are very data-inefficient, and it is hard to analyze their reasoning process. These issues are addressed by end- to-end differentiable reasoning systems such as Neural Theorem Provers (NTPs), although they can only be used with small-scale symbolic KBs. In this paper we first propose Greedy NTPs (GNTPs), an extension to NTPs addressing their complexity and scalability limitations, thus making them applicable to real-world datasets. This result is achieved by dynamically constructing the computation graph of NTPs and including only the most promising proof paths during inference, thus obtaining orders of magnitude more efficient models.(Source code and datasets are available online at https://github.com/uclnlp/gntp. This is an extended version of [45], selected for an oral presentation at AAAI 2020.) Then, we propose a novel approach for jointly reasoning over KBs and textual mentions, by embedding logic facts and natural language sentences in a shared embedding space. We show that GNTPs perform on par with NTPs at a fraction of their cost while achieving competitive link prediction results on large datasets, providing explanations for predictions, and inducing interpretable models.

Computational context understanding refers to an agent’s ability to fuse disparate sources of information for decision-making and is, therefore, generally regarded as a prerequisite for sophisticated machine reasoning capabilities, such as in artificial intelligence (AI). Data-driven and knowledge-driven methods are two classical techniques in the pursuit of such machine sense-making capability. However, while data-driven methods seek to model the statistical regularities of events by making observations in the real-world, they remain difficult to interpret and they lack mechanisms for naturally incorporating external knowledge. Conversely, knowledge-driven methods, combine structured knowledge bases, perform symbolic reasoning based on axiomatic principles, and are more interpretable in their inferential processing; however, they often lack the ability to estimate the statistical salience of an inference. To combat these issues, we propose the use of hybrid AI methodology as a general framework for combining the strengths of both approaches. Specifically, we inherit the concept of neuro-symbolism as a way of using knowledge-bases to guide the learning progress of deep neural networks. We further ground our discussion in two applications of neuro-symbolism and, in both cases, show that our systems maintain interpretability while achieving comparable performance, relative to the state-of-the-art.

In this chapter we focus the use of knowledge representation and reasoning (KRR) methods as a guide to machine learning algorithms whereby relevant contextual knowledge can be leveraged upon. In this way, the learning methods improve performance by taking into account causal relationships behind errors. Performance improvement can be obtained by focusing the learning task on aspects that are particularly challenging (or prone to error), and then using added knowledge inferred by the reasoner as a means to provide further input to learning algorithms. Said differently, the KRR algorithms guide the learning algorithms, feeding it labels and data in order to iteratively reduce the errors calculated by a given cost function. This closed loop system comes with the added benefit that errors are also made more understandable to the human, as it is the task of the KRR system to contextualize the errors from the ML algorithm in accordance with its knowledge model. This represents a type of explainable AI that is focused on interpretability. This chapter will discuss the benefits of using KRR methods with ML methods in this way, and demonstrate an approach applied to satellite data for the purpose of improving classification and segmentation task.

Transfer learning aims at building robust prediction models by transferring knowledge gained from one problem to another. In the semantic Web, learning tasks are enhanced with semantic representations. We exploit their semantics to augment transfer learning by dealing with when to transfer with semantic measurements and what to transfer with semantic embeddings. We further present a general framework that integrates the above measurements and embeddings with existing transfer learning algorithms for higher performance. It has demonstrated to be robust in two real-world applications: bus delay forecasting and air quality forecasting.

Predictive analytic tasks identify likely future outcomes based on historical and current data. Predictions alone, however, are usually insufficient for users to make sound decisions, due to reasons such as regulatory requirements, distrust towards black box technology. To this end, explanations have been used as one way to enable adoption of predictive analytics. In particular, semantics-rich explanations that leverage knowledge graphs are explored by both academics and practitioners. This chapter presents three case studies: predictive analytics for identifying abnormal expense claims, mitigating project risks, and predicting pronunciations and learning the language model of Chinese characters. In addition to predictions, explanations play an important part in these cases. They could impact decision making, e.g., by showing that a project is risky likely because of the incompetent delivery centers, or they can enhance users’ trust in the predictive models, e.g., by presenting the dependencies exist between the pronunciation of a Chinese character and that of its substructures. Regardless of the size or the form of the knowledge graphs, the three case studies show that explanations built on domain knowledge add invaluable insights to predictive analytics.

With the ever-growing adoption of Artificial Inteligence (AI) models comes an increasing demand for making their output actions understandable.With this aim, it is crucial to generate natural language explanations of their models. One way of achieving this goal is to translate the languages of the Semantic Web (SW) into natural language. In this chapter, we give an overview of how SW languages can be used to generate texts and consequently explanations. We begin by presenting LD2NL, a framework for verbalizing the three key languages of the Semantic Web, i.e., RDF, OWL, and SPARQL. Afterward, we talk about the generation of texts by relying on Neural Network (NN) models. We hence present NeuralREG, an approach for generating referring expression of Knowledge Graph (KG) entities while generating texts. Both frameworks are evaluated in open surveys with 150 persons. The results suggest that although generating explanations from KGs is in its infancy, both LD2NL and NeuralREG can generate verbalizations that are close to natural languages, and non-experts can easily understand that. In addition to that, it enables non-domain experts to interpret AI actions with more than 91% of the accuracy of domain experts.

Interest in the field of Explainable Artificial Intelligence has been growing for decades, and has accelerated recently. As Artificial Intelligence models have become more complex, and often more opaque, with the incorporation of complex machine learning techniques, explainability has become more critical. Recently, researchers have been investigating and tackling explainability with a user-centric focus, looking for explanations to consider trustworthiness, comprehensibility, explicit provenance, and context-awareness. In this chapter, we leverage our survey of explanation literature in Artificial Intelligence and closely related fields and use these past efforts to generate a set of explanation types that we feel reflect the expanded needs of explanation for today’s artificial intelligence applications. We define each type and provide an example question that would motivate the need for this style of explanation. We believe this set of explanation types will help future system designers in their generation and prioritization of requirements and further help generate explanations that are better aligned to users’ and situational needs.

Explainable AI has recently gained momentum as an approach to overcome some of the more obvious ethical implications of the increasingly widespread application of AI (mostly machine learning). It is however not always completely evident whether providing explanations actually achieves to overcome those ethical issues, or rather create a false sense of control and transparency. This and other possible misuses of Explainable AI leads to the need to consider the possibility that providing explanations might itself represent a risk with respect to ethical implications at several levels. In this chapter, we explore through a series of scenarios how explanations in certain circumstances might affect negatively specific ethical values, from human agency to fairness. Through those scenarios, we discuss the need to consider ethical implications in the design and deployment of Explainable AI systems, focusing on how knowledge-based approaches can offer elements of solutions to the issues raised. We conclude on the requirements for ethical explanations, and on how hybrid-systems, combining machine learning with background knowledge, offer a way towards achieving those requirements.

eXplainable AI focuses on generating explanations for the output of an AI algorithm to a user, usually a decision-maker. Such user needs to interpret the AI system in order to decide whether to trust the machine outcome. When addressing this challenge, therefore, proper attention should be given to produce explanations that are interpretable by the target community of users.

In this chapter, we claim for the need to better investigate what constitutes a human explanation, i.e. a justification of the machine behavior that is interpretable and actionable by the human decision makers. In particular, we focus on the contributions that Human Intelligence can bring to eXplainable AI, especially in conjunction with the exploitation of Knowledge Graphs.

Indeed, we call for a better interplay between Knowledge Representation and Reasoning, Social Sciences, Human Computation and Human-Machine Cooperation research – as already explored in other AI branches – in order to support the goal of eXplainable AI with the adoption of a Human-in-the-Loop approach.

In this chapter, we focus on the role of identity links in knowledge-based explainable systems, i.e. systems that rely on background knowledge from knowledge graphs to build explanations. With the rise of explainable transparent methods in the area of eXplainable AI, systems integrating multiple sources of aligned knowledge will become more and more common. We hypothesize that the interpretability and results of these systems could be affected by the discrepancy and misalignment between knowledge sources – a widely known problem in the Knowledge Representation community. We therefore study the role of identity in knowledge-based explainable systems, i.e. if and how explainable systems do rely on multiple knowledge graphs, then show examples of the impact of misusing identity on the interpretability of a system. Finally, we describe methods that can promote the correct alignment of knowledge sources. Our hope is to provide support to improving current knowledge-based explainable methods and, more in general, foster a better integration of knowledge representation and explainable AI.