This is a contributed article by Szymon Klarman, Knowledge Architect at BlackSwan Technologies.
Artificial intelligence is often compared to human intelligence. But AI capabilities in isolation can only match some parts of human intelligence. For instance, a large proportion of AI products use machine learning to categorise, classify or make predictions based on data – often with huge success. And yet, using these insights to make a demonstrable difference to an organisation can be problematic due to a lack of contextual information, which is a fundamental component of human intelligence. This is where graph technology comes in.
As many as 50% of Gartner client inquiries on the topic of AI are about the use of graph technology. And looking to the future, Gartner predicts that by 2023, graph technologies will facilitate rapid contextualisation for decision making in 30% of organisations worldwide.
Knowledge graphs are one of the key technologies gaining traction from enterprises. First, they can represent all the concepts of a particular domain such as ‘corporate risk’ or ‘equipment repair’. Second, they allow for the exploration of relationships between entities of interest such as organisations, people, locations and transactions. As organisations are often unable to act upon a large proportion of the data they possess, they’re forced to make decisions based on contradicting, duplicated or incomplete information. Knowledge graphs can help by creating a single source of truth, contextualising huge amounts of data in a similar way to the human mind.
Humans make decisions based on what we consciously or unconsciously know or believe about the world. For example, a person may decide to apply for a job with a company because the company offers products that they have knowledge of and because they are familiar with many of the company’s customers. This context allows the person to better evaluate their chances of being hired. This is the same context that knowledge graphs emulate, which becomes a foundation for extracting insights using AI techniques. Ultimately, enterprises can use these insights to make decisions.
How does a knowledge graph work?
Without knowing, most people interact with a knowledge graph on a regular basis; for instance, searching on Google or even using popular food delivery apps. In the Google example, if we type the phrase ‘Alexander Fleming’ into Google’s search box, users would not just get a set of links containing this phrase, but also an information box with data about a specific entity; UK biologist Alexander Fleming, that matches the query.
Within the box, the information includes points such as ‘Fleming invented Penicillin’, his birthplace and birthdate. Under the surface, this data corresponds to a small fragment of a knowledge graph. The nodes of this graph represent real-world things or concepts, while the edges represent their properties or the relationship between them. The labels, like ‘date’, ‘person’, ‘born’, ‘invented’ and ‘object’ are what make the meaning of the whole structure – the semantics – explicit.
Google’s knowledge graph views these facts, found from different sources on the internet, as connected pieces of the same conceptual map. This networked representation of data is not only more intuitive, but it makes the existing relationships between entities in the given domain transparent. This enables users to put everything in context, in an easy to comprehend and accessible way.
For instance, when searching ‘When was the birthdate of the inventor of penicillin’, the response is the correct date of August 6, 1881. The response is the result of traversing the underlying graph the way a human with the same knowledge would: starting from the named object, via the link to a person, and finally to the date of birth. This vital feature of connecting data and uncovering non-obvious and indirect relationships is one of the main benefits of using a knowledge graph. A targeted enterprise knowledge graph would do a similar job for a specific domain, such as cyber security or risk management.
From a technology standpoint, a knowledge graph is a live information asset. It involves data management processes that rely on a number of AI techniques to thoroughly interpret data and solve a wide range of business problems. This use of multiple AI techniques is what Gartner calls ‘Composite AI’.
In response to a query, knowledge graphs can present all relevant pieces of information in one consistent conceptual map; helping decision-makers gain insights of unprecedented granularity, depth, and precision. A clear overview of all data and relationships is particularly important when a business decision relates to safety, due diligence, or compliance.
How enterprises use knowledge graphs
It's not just consumer-facing applications that a knowledge graph can be useful for. Enterprises can use the technology to process data about products, services, and clients, even when the data is spread over multiple databases, spreadsheets and documents in disconnected systems. The technology helps organisations to combine structured, unstructured, internal and external data; enable complex logic with conditional rules and machine learning inference; and adapt to changing user and market needs through constant updates.
One such use case is the ability to undertake research on any given company; potentially a prospective client, customer or partner.
Traditionally, to conduct company research, an analyst has to access and review tens of internal databases, knowledge platforms, documents and spreadsheets. They then have to crawl public APIs, websites and news sources, just to obtain reasonably comprehensive coverage of the legal entities in question, before any genuine analysis can even begin. This process is time-consuming, error-prone, and not scalable, thus significantly increasing the cost of making informed decisions.
In contrast, once the single source of truth in the form of a knowledge graph is set up, it is easy to answer questions like Who are the direct and indirect subsidiaries of a company? Advanced network analysis can provide answers to more complex questions such as Who is the most important shareholder in the network? Machine learning algorithms can be fed by a knowledge graph in a way that would be incredibly difficult to achieve otherwise, to create advanced classification results or predictions. To obtain the same accuracy, organisations would have to collect many common features related to each of the entities they want to classify.
The knowledge graph then becomes a base for a current, 360-degree view of any entity. The structures and connections between entities in the graph can be captured and analysed by artificial and autonomous agents. If combined with automated workflows and AI technologies for processing and evaluation, companies can benefit from greater flexibility, better insights, a reduction in computation power and storage resources, and perhaps most significantly, the knowledge graph can be used as the backbone for sophisticated AI systems that cater to a wide range of applications, such as thwarting fraud and money laundering.
Knowledge graphs are an important capability to add to an enterprise’s arsenal – whether used directly or through a third-party system. Building an in-house knowledge graph can be resource-intensive, meaning a white box AI design is in critical demand for the wider use of the technology. Knowledge graphs allow enterprises to contextualise tremendous amounts of data that would otherwise be impossible for humans to carry out on their own. The technology empowers analysts to efficiently analyse the knowledge that has been extracted, adding context and depth to other AI techniques and serving as a bridge between humans and systems.