Technology to connect people and knowledge

Classification and extraction

The metadata layer that’s foundational to a data mesh implementation serves to “decouple the physical location of the content from the business meaning and business context,” Nivala pointed out. As such, there’s a premium for the technologies that are responsible for actually populating the metadata for frameworks like data mesh, knowledge graphs, and smart document platforms.

In some cases, language models and other manifestations of machine learning are instrumental for doing the “extraction and classification so people can tap into PDFs and pull out relevant information,” Kato indicated. There are also technologies pertaining to metadata harvesting or scanning of entire repositories that replicate this task at scale, on demand.

Nivala described this phenomenon as a way to “index or scan those repositories to create a metadata index, the meta-data layer that is the connecting tissue.” Consequently, organizations can classify different documents and even glean the pertinent parts of them, according to the metadata, to understand, for example, which factory and piece of equipment are involved in content related to manufacturing.

When language models are involved in the metadata harvesting process, the degree of specificity for the extraction and classification process is astounding. According to Nivala, this methodology allows modern solutions to “automatically analyze the content of the documents and extract valuable business information like contracting parties or specific contractors. Does it have an assignment clause, a termination penalty, or something similar? We can surface that into content more like structured data to the metadata layer, which is then easier to analyze with business intelligence tools.”

The defining characteristics

The ability to centralize the knowledge found in the distributed tools, repositories, and databases for KM practitioners is the defining characteristic of today’s technologies. Generative machine learning models not only play pivotal roles in managing that knowledge to make it meaningful to users, but also in enabling them to interface with it on-demand, for ad-hoc use cases and in a manner Gu characterized as “very lively, very interactive.”