It’s becoming increasingly difficult to overlook the presence and overall worth of enterprise knowledge graphs in contemporary IT, particularly for applications of Artificial Intelligence.
Earlier this month, Franz and the Semantic Web Company unveiled a partnership to foster the Noam Chomsky Knowledge Graph Project, which will attempt to link all of the major literary works from the noted linguist and AI influencer—in addition to his several interviews and movies about his life and work—in a single platform searchable and retrievable to the general public.
Knowledge graphs have generated so much interest partly because of their multifaceted utility for AI and data management in general. They link together data of any variety, structure or format in business terms via uniform data models. Thus, organizations can join and traverse all of their data, semantically tagged with unique machine-readable identifiers—which are ideal for intelligent systems, machine learning analytics, interoperability, and an array of other benefits influential for AI applications.
Moreover, they’re rooted in graph database technologies which, according to Cambridge Semantics VP of Marketing John Rueter, have myriad benefits to the enterprise:
“Graph is a much more appropriate, effective and simpler way to work with all the data that’s coming at us and, by the way, it works not only with the structured data but the unstructured data. The kinds of queries you can make, the kinds of answers you can get back, are much richer, more effective, and can really advance the decision making process in ways that weren’t possible before.”
Harmonizing Data in Business Terms
By definition, knowledge graphs are able to connect all of the data relevant for a specific business purpose—as broad or as narrow as that might be. Whether organizations have an internal knowledge graph of their operations and employee characteristics, or an external one for their customers, according to Franz CEO Jans Aasman “all these knowledge graphs are still connected in some way; they can link together.” Individual knowledge graphs can link together in the same way that they join respective data types: via the semantic (RDF) graph foundation that focuses on the edges or connections between data, and which harmonizes data with uniform standards.
In this environment, all data readily conforms to common data models so that “graph is a better way to harmonize data coming from unstructured sources…and structured sources liked databases,” Cambridge Semantics CTO Sean Martin stated. “It’s a better way of tying those together and harmonizing them with business meaning.”
This latter point is one of the primary distinctions of knowledge graphs; they represent data via semantic statements based on what the data means for business purposes, as opposed to arcane IT schema. Moreover, the celerity with which graphs align structured, unstructured and semi-structured data is ideal for accelerating data preparation for data science. “It’s a good way of structuring information that comes from unstructured data,” Martin remarked about knowledge graphs. “Because they’re flexible, you can easily add to them.”
Unique Machine Identifiers
The resulting linked data of knowledge graphs forms the basis of knowledge associated with any particular domain. For delivery companies in the transportation industry, for instance, that would include “a knowledge base about what customers want, what drivers want, what the trucking company wants, and the routes, in a knowledge graph,” Aasman said.
Another critical facet about the way that data’s stored and connected in knowledge graphs is they contain unique identifiers that are machine-readable. The former of these characteristics is well suited for integrating data from all sources and for system interoperability; the latter abets machine intelligence capabilities associated with AI and intelligent systems.
Partly because of the prevalence of AI applications and applications generating sensor data, TopQuadrant CMO and VP of Professional Services Robert Coyne commented, “You need the ability to find pieces of data from anywhere, and also use pieces of data from anywhere: but not create copies of them.” The Uniform Resource Identifiers (URIs) of knowledge graphs is helpful for integrating data and keeping track of them however they’re used.
“There needs to be a standards-based way of identifying sources that’s bigger than the local application subject, or data source subject identifier,” Polikoff mentioned. URIs provide that standardized identification. “The global unique identifiers that operate on the World Wide Web, that’s the biggest system there is,” Coyne said. “That’s what standards-based knowledge graphs use.”
A Reciprocal Relationship
The relationship between enterprise knowledge graphs and AI applications is mutually beneficial. Knowledge graphs can hasten preparation for advanced data science models; they imbue data with machine-readable, unique identifiers for integrating data and facilitating intelligent system interoperability. In turn, machine learning capabilities can also increase—as well as improve—the knowledge contained within these graphs.
Organizations can run machine learning algorithms to determine the likelihood of equipment failure on, for example, airplane parts in the Industrial Internet. Depending on what maintenance or repair action is taken, organizations can run machine learning on those results to input them as predictions into the knowledge graph for this equipment and its operation. “What you learned, the predictions from your machine learning system, can go into the knowledge base for the future,” Aasman explained.
By repeating this cycle organizations add to the information of their knowledge graph with machine learning predictions “and then slowly you learn what works and what doesn’t work,” Aasman said. The knowledge graph improves by incorporating the findings of machine learning. According to Aasman, “This is a pattern that would apply to many more use cases.”
Knowledge graphs may well become a foundational platform for AI applications in the future. They expedite lengthy data preparation measures for data scientists engaged in advanced, predictive analytics. By enriching data with machine-readable global identifiers, they’re suitable for interoperability and feeding systems of machine intelligence. They offer a foundation of knowledge that can inform machine learning inputs, and are in turn bettered by the outputs of machine learning applied to real world business problems—once they’re input into the graph.
Their biggest distinction, however, is well summarized by Cambridge Semantics VP of Engineering Barry Zane, who observed, “Graphs sound new and spooky and dangerous and so forth, unless you’re a total geek. The reality is it’s enormously simpler than what you’ve been doing in data warehousing. You can get the same results from the same questions, but it’s all just a lot easier with graphs.”
Jelani Harper is an editorial consultant servicing the information technology market, specializing in data-driven applications focused on semantic technologies, data governance and analytics.