As 2022 dawns, knowledge graphs bear the dubious distinction of being at the epicenter of AI and machine learning for two reasons. One is that, unassisted, they are one of the myriad manifestations of AI due in part to their highly contextualized understanding of the relationships between data, enterprise knowledge, and the terms that populate both. Second, they’re perhaps the most creditable means of assembling all data at the enterprise’s disposal—regardless of structure variation, type, or format— for building the machine learning models whose predictive and prescriptive power makes AI so desired by end users. In this regard, they’re either the foundation of the popular data fabric tenet or a crowning layer for integrating the data connected by this framework.
Here are some of the many use cases illustrating how knowledge graphs are employed and why they’re central to AI, machine learning, and knowledge management as a discipline:
♦ Enhancing knowledge: By inputting the outcome of machine learning predictions into knowledge graphs, organizations can increase their enterprise knowledge for more time-saving automation opportunities, such as search.
♦ Enabling smart inferences: Knowledge graphs can intelligently reason about enterprise knowledge to infer new facts from existing ones, adding a critical logic component to AI’ s more hyped learning prowess. When used in tandem, these capabilities reduce the costs and time AI deployments otherwise require.
♦ Contributing to data fabrics and machine learning: Knowledge graphs can harmonize data to facilitate the proper selection of training data and features that inform building machine learning models. Their integration capabilities are optimal for linking together all data in a single data fabric.
♦ Optimizing outputs: There are numerous forms of AI (from fundamentals of unsupervised learning to contemporary word embeddings) that either can only be done, or deliver the best results, with graphics.
Consequently, knowledge graphs provide three key functions for AI and machine learning, according to Kendall Clark, CEO of Stardog. “There’s an analytics capability on one side, and a data integration or data prep capability on the other side. And you can think about the interaction of them as almost a third capability,” he said.
Mastering the data preparation side to influence analytics results by engaging these two capabilities (producing the third mentioned by Clark) ultimately engenders the summit of enterprise AI. If you can put data together in certain ways and do the right kind of analytics, you’ve achieved insights that let you drive a new business process, save money, or deal with whatever underlying business problem you are trying to handle, said Sean Martin, CEO of Cambridge Semantics. “That is data being turned into gold: those insights.”
All graphs are not equal
The ability to transmute enterprise knowledge into insights, or what Martin termed “gold,” varies depending on what type of knowledge graph is employed. Traditionally, there were two types: semantic graphs (typified by RDF [Resource Description Framework], although there are proprietary ones) and labeled property graphs. The latter allows users to add properties or attributes about data for vital things such as metadata or provenance. Semantic graphs, however, have greater faculties for both data preparation and analytics The knowledge graph moniker originated with semantic graphs that make logical inferences about their data.
More importantly, knowledge graphs are designed for interoperability and linking together all data via uniform standards. Property graphs simply function as silos. This distinction is why the former are sought for data fabrics. “In a single department or single source environment, property graphs can really help to make more out of the existing data,” said Andreas Blumauer, CEO of the Semantic Web Company. “However, they don’t resolve the systemic data problem you have in large enterprises.” RDF*, a semantic graph that supports labeled properties, is a third option.
The cardinal benefit knowledge graphs offer knowledge management is a scalable means of amassing enterprise knowledge that focuses on its relationships. When machine learning and AI are added, organizations can incorporate greater amounts of knowledge that, as Martin noted, are more profound. One way to do so is by inserting the outputs of machine learning model predictions back into the graph. “If your models are good, they’re predicting information that’s equally valuable as part of analytics,” Martin said. This approach broadens the amount of enterprise knowledge and is accessible via search, another AI capability.
According to Alex Smith, product management lead for iManage AI, “With search, knowledge graphs, AI, and machine learning, we’re able to solve the problem of how to deliver content at the right point in time. By content, I mean ‘knowledge.’” This can go from anything, such as templates, best practices, and information that has been curated, to information about a project people are working on, said Smith. Organizations can improve the usability of knowledge graph search with natural language processing (NLP). The connections between terms, documents, projects, and people are readily discernible with search to “allow you to build a network of connected assets that you can tie together so when you find one of them, it’s like pulling a string, you find the others,” Martin said.
KMWorld is the leading publisher, conference organizer, and information provider serving the knowledge management, content management, and document management markets.
All Content Copyright © 1998 – 2022
Information Today Inc.
22 Bayview Street, 3rd Floor
PO Box 404
Camden, ME 04843