Knowledge graphs

Introduction

Knowledge graphs (KGs) organise data from multiple sources, capture information about entities of interest in a given domain or task (like people, places or events), and forge connections between them. In data science and AI, knowledge graphs are commonly used to:

Facilitate access to and integration of data sources;
Add context and depth to other, more data-driven AI techniques such as machine learning; and
Serve as bridges between humans and systems, such as generating human-readable explanations, or, on a bigger scale, enabling intelligent systems for scientists and engineers.

This interest group will facilitate research and innovation in a critical area of data science and AI.

Explaining the science

The term ‘knowledge graph’ has been introduced by Google in 2012 to refer to its general-purpose knowledge base, though similar approaches have been around since the beginning of modern AI in areas such as knowledge representation, knowledge acquisition, natural language processing, ontology engineering and the semantic web. Today, KGs are used extensively in anything from search engines and chatbots to product recommenders and autonomous systems. In data science, common use cases are around adding identifiers and descriptions to data of various modalities to enable sense-making, integration, and explainable analysis. In AI, knowledge graphs complement machine learning techniques to:

reduce the need of large, labelled datasets;
facilitate transfer learning and explainability;
encode domain, task and application knowledge that would be costly to learn from data alone.

A knowledge graph organises and integrates data according to an ontology, which is called the schema of the knowledge graph, and applies a reasoner to derive new knowledge. Knowledge graphs can be created from scratch, e.g., by domain experts, learned from unstructured or semi-structured data sources, or assembled from existing knowledge graphs, typically aided by various semi-automatic or automated data validation and integration mechanisms.

While there are many definitions of knowledge graphs around, most of them agree that knowledge graphs are:

Graphs: unlike knowledge bases, the content of KGs is organised as a graph, where nodes (entities of interest and their types), relationships between and attributes of the nodes are equally important. This makes it easy to integrate new datasets and formats and supports exploration by navigating from one part of the graph to the other through links.
Semantic: the meaning of the data is encoded for programmatic use in an ontology, which describes the types of entities in the graph and their characteristics and can be represented as a schema sub-graph. This means that the graph is both a place to organise and store data, and to reason what it is about and derive new information.
Alive: knowledge graphs are flexible in terms of the types of data and schemas they can support. They, including their schemas, evolve to reflect changes in the domain and new data is added to the graph as it becomes available.

Some knowledge graphs are used primarily within the organisation that created them. The most common example is the Google knowledge graph, which is used in web search, or Amazon’s product graph. Other knowledge graphs are openly available. These include DBpedia, Wikidata, WordNet, Geonames, etc.

Aims

The UK has been at the forefront of several strands of work that have made KGs as successful as they are today, in areas like: knowledge representation formalisms; natural language processing; machine learning; methodologies to construct, learn and manage ontologies and knowledge bases; de facto standard ontologies and vocabularies; scalable reasoning, linked data etc.

The main objectives of the interest group are:

To strengthen this national community of scholars and innovators to continue to pursue world-leading research and explore novel technologies related to and applications of KGs,
To reinforce the systematic use of KGs in practical data science and AI applications.
To identify a portfolio of joint research projects for current and future PG students.

The group will encourage members to:

Present latest ideas and achievements;
Share ideas, knowledge and experiences;
Forge collaborations;
Align and expand existing education and training activities;
Reflect upon and raise awareness of specific challenges in equality, diversity and inclusion in the field; and
Pool resources and expertise to collectively unlock new funding and partnership opportunities that are not accessible to members in isolation.

Talking points

Constructing and maintaining large-scale, yet high-quality knowledge graphs

Modern knowledge graphs are the result of complex assemblies of manual and automatic modelling and data ingestion pipelines. Staying on top of these processes while ensuring that the information remains up to date, consistent and trustworthy requires specialised socio-technical methods ranging from knowledge acquisition to natural language processing to machine learning and human-computer interaction.

Knowledge-enhanced data-driven technologies

Knowledge graphs are never used in isolation. They co-exist and complement data-driven solutions, including natural language processing (text understanding, knowledge extraction, text generation) and data management (semantic labelling, record linking, data cleansing etc.). Recent research looks, for instance, at trade-offs between semantic representations and deep learning, and, more broadly at designing symbolic/connectionist AI architectures.

New forms of knowledge graphs

Knowledge graphs need to be as rich as the AIs they serve. We need new ideas and formalisms to go beyond the types of knowledge captured in current open and enterprise KGs, which tend to focus on entities and their properties and on factual knowledge. New areas include common-sense knowledge, multiple viewpoints, as well as events and other types of temporal information, and cause-effect chains.