The Next Generation of Knowledge Management

It would be a gross oversimplification—and a great disservice to the profession—to consider the next generation of KM solely in terms of new breakthrough technologies. Autonomous agents, language models, vector databases, knowledge graphs, and other technologies certainly enable fundamental KM capabilities. But that isn’t the whole picture. Instead, the next generation of KM will be characterized by intelligent document classification and processing, universal question-answering, low-code application building, systems integrations, and realtime feedback mechanisms. Integrating technologies so that they allow organizations to interact with their knowledge assets without neglecting the underlying principles of KM and its understanding that people’s use of the technologies lies at the heart of next-generation KM.

These are the constructs that strike at the core of what KM has come to mean for knowledge workers everywhere who simply “want the right assets, at the right moment, to do their jobs,” commented Alex Smith, iManage’s global product lead–knowledge, search, AI. This simple desire has informed the next generation of constructs for managing and surfacing relevant enterprise knowledge, oftentimes at the point of its consumption.

Knowledge Graph Relationships

The knowledge graph framework has emerged as one of the most utilitarian for supplying an overarching understanding of what resources are available to organizations to accomplish business objectives. More than providing rich metadata descriptions about enterprise content, they also contextualize the processes and personnel responsible for them to “relate the content together to workflows, and to people, and to how they use them,” Smith continued. Semantic knowledge graphs are particularly helpful in this regard because they integrate entire systems—including respective databases—for a comprehensive understanding of how they pertain to each other and specific objectives.

Organizations savvy enough to construct retrospective knowledge graphs for HR purposes or projects employees have worked on finding value in linking together products, IT systems, and people to enable query across them with remarkable specificity. “You can find the people that know the most about GenAI [generative AI] agents that were successful products around the design of systems for this kind of project and see if we’re paying them market salary,” pointed out Jans Aasman, Franz CEO. “Now, we’ve got a query that touches at least four databases.”

GraphRAG

Knowledge graphs may prove even more advantageous for facilitating the graph retrieval-augmented generation (GraphRAG) paradigm. RAG is one of the numerous forms of prompt augmentation for adding to the context language models receive when performing natural language question- answering, semantic search, and vector similarity search. Many consider the GraphRAG option an improvement on RAG because “it has the basic ability to ground your AI to say, ‘Here is the answer; it relates to this,’” Smith mentioned. GraphRAG accomplishes this goal since, unlike its non-graph counterpart, the entirety of the graph’s contents can be used for metadata filtering.

This approach not only increases the accuracy of search results, but also decreases the amount of content models search, which provides cost benefits. According to Aasman, “What GraphRAG people do is annotate each document with the important terms. Then when you ask a question, it finds all the terms that might be related to your question and does a structured query for what documents might apply. Then you do RAG.”