GraphNews

Ontology-grounded knowledge graphs provide a reliable and verifiable method for mitigating hallucinations in LLM-based clinical question answering. By embedding structured clinical semantics into the reasoning process, the framework enhances factual accuracy, reproducibility, and safety in biomedica …

Large Language Models (LLMs) have been extensively adopted in Knowledge Graph Completion (KGC), showcasing significant research advancements. However, as black-box models driven by deep neural...

I Built a Knowledge Graph Recommender to Understand Where Graphs Actually Fail By Kartikeya Mandhar | UCLA MSBA I’ve had a complicated relationship with knowledge graphs. The idea that you can …

VKG vs KG: The Practical Guide No One Gives You When Building Semantic Platforms A practical guide to choosing between Virtual Knowledge Graphs and Knowledge Graphs for multi-source, ontology-driven …

Gartner has declared 2026 as the year of context. Come read allabout context.

How maplib and RDF turned ~1,900 Agricola cards into a queryable, interconnected data model, and what I learned along the way.

If you have both unique domain expertise and know how to make it usable to your AI systems, you’ll be hard to beat.

And brews up a nice glass of sherry as well ...

Enterprise search, foundation models, productivity AI, and HR analytics are all racing to own the enterprise stack and all hitting the same wall. The missing layer is not a feature any of them will ship. It is a category of its own.

We interviewed a number of individuals who work with graphs or in the world of data to get their perspectives on how graph and connected data are evolving—and where they might be headed.

𝐀 𝐮𝐬𝐞𝐟𝐮𝐥 𝐭𝐚𝐤𝐞𝐚𝐰𝐚𝐲 𝐟𝐫𝐨𝐦 𝐭𝐡𝐞 𝐜𝐮𝐫𝐫𝐞𝐧𝐭 “𝐜𝐨𝐧𝐭𝐞𝐱𝐭 𝐠𝐫𝐚𝐩𝐡” 𝐜𝐨𝐧𝐯𝐞𝐫𝐬𝐚𝐭𝐢𝐨𝐧 👉 𝐎𝐧𝐞 𝐠𝐫𝐚𝐩𝐡 𝐢𝐬 𝐮𝐬𝐮𝐚𝐥𝐥𝐲 𝐧𝐨𝐭 𝐞𝐧𝐨𝐮𝐠𝐡 The more practical design pattern is a stack of context layers, each with a distinct responsibility. In recent discussions (including Year of the Graph Vol. 30 ) by George Anadiotis, the strongest point is not hype around a new term. It is the shift from “context as retrieved text” to context as operational infrastructure. For AI decision systems, five knowledge layers are emerging: 👉𝐊𝐧𝐨𝐰𝐥𝐞𝐝𝐠𝐞 / 𝐒𝐞𝐦𝐚𝐧𝐭𝐢𝐜 𝐥𝐚𝐲𝐞𝐫 𝐂𝐨𝐧𝐜𝐞𝐩𝐭𝐬:  𝙿𝚊𝚜𝚜𝚎𝚗𝚐𝚎𝚛, 𝙵𝚕𝚒𝚐𝚑𝚝, 𝙸𝚝𝚒𝚗𝚎𝚛𝚊𝚛𝚢, 𝙲𝚘𝚗𝚗𝚎𝚌𝚝𝚒𝚘𝚗 𝐑𝐞𝐥𝐚𝐭𝐢𝐨𝐧𝐬𝐡𝐢𝐩𝐬: 𝙿𝚊𝚜𝚜𝚎𝚗𝚐𝚎𝚛 𝒃𝒐𝒐𝒌𝒔 𝙸𝚝𝚒𝚗𝚎𝚛𝚊𝚛𝚢, 𝙸𝚝𝚒𝚗𝚎𝚛𝚊𝚛𝚢 𝒊𝒏𝒄𝒍𝒖𝒅𝒆𝒔 𝙵𝚕𝚒𝚐𝚑𝚝𝙻𝚎𝚐, 𝙵𝚕𝚒𝚐𝚑𝚝 𝒂𝒓𝒓𝒊𝒗𝒆𝒔_𝒂𝒕 𝙰𝚒𝚛𝚙𝚘𝚛𝚝 𝐃𝐞𝐟𝐢𝐧𝐢𝐭𝐢𝐨𝐧𝐬: 𝚖𝚒𝚜𝚜𝚎𝚍_𝚌𝚘𝚗𝚗𝚎𝚌𝚝𝚒𝚘𝚗 = 𝚊𝚌𝚝𝚞𝚊𝚕_𝚊𝚛𝚛𝚒𝚟𝚊𝚕_𝚝𝚒𝚖𝚎 +𝚖𝚒𝚗𝚒𝚖𝚞𝚖_𝚌𝚘𝚗𝚗𝚎𝚌𝚝𝚒𝚘𝚗_𝚝𝚒𝚖𝚎𝚗𝚎𝚡𝚝_𝚕𝚎𝚐_𝚍𝚎𝚙𝚊𝚛𝚝𝚞𝚛𝚎_𝚝𝚒𝚖𝚎   👉𝐆𝐨𝐯𝐞𝐫𝐧𝐚𝐧𝐜𝐞 𝐥𝐚𝐲𝐞𝐫 𝐏𝐨𝐥𝐢𝐜𝐲: “Do not expose passenger PII to operations recommendations” 𝐑𝐮𝐥𝐞: “Compensation above 300 EUR requires Revenue Approver” 𝐂𝐨𝐧𝐬𝐭𝐫𝐚𝐢𝐧𝐭: “Any rebooking recommendation must include a valid contract check and risk score” 👉𝐎𝐩𝐞𝐫𝐚𝐭𝐢𝐨𝐧𝐚𝐥 / 𝐄𝐯𝐞𝐧𝐭 𝐥𝐚𝐲𝐞𝐫 𝐄𝐯𝐞𝐧𝐭𝐬: FlightDelayed(F1001, +45m), PassengerCheckedIn(P123, F1001), GateChanged(F1002, G12) 𝐒𝐭𝐚𝐭𝐞 𝐝𝐞𝐫𝐢𝐯𝐞𝐝 𝐟𝐫𝐨𝐦 𝐞𝐯𝐞𝐧𝐭𝐬:  itinerary_status = at_risk, estimated_missed_connections=27 👉𝐃𝐞𝐜𝐢𝐬𝐢𝐨𝐧 / 𝐏𝐫𝐨𝐯𝐞𝐧𝐚𝐧𝐜𝐞 𝐥𝐚𝐲𝐞𝐫 𝐃𝐞𝐜𝐢𝐬𝐢𝐨𝐧𝐏𝐫𝐨𝐩𝐨𝐬𝐚𝐥 𝐃-𝟏𝟎𝟒𝟐:“Rebook Passenger P123 to F1010” 𝐄𝐯𝐢𝐝𝐞𝐧𝐜𝐞:  delay event E-7781, missed-connection score 0.91, seat availability snapshot 𝐂𝐡𝐞𝐜𝐤𝐬: policy gate passed, no PII violation, compensation threshold not triggered 𝐀𝐩𝐩𝐫𝐨𝐯𝐚𝐥: Ops Controller approved at 10:42 𝐎𝐮𝐭𝐜𝐨𝐦𝐞: Passenger rebooked successfully; arrival delay reduced by 3h 👉𝐀𝐜𝐭𝐢𝐨𝐧/𝐏𝐞𝐫𝐦𝐢𝐬𝐬𝐢𝐨𝐧 𝐥𝐚𝐲𝐞𝐫 𝐑𝐞𝐛𝐨𝐨𝐤𝐢𝐧𝐠 𝐀𝐠𝐞𝐧𝐭: can propose rebook_now for low/medium risk 𝐎𝐩𝐬 𝐂𝐨𝐧𝐭𝐫𝐨𝐥𝐥𝐞𝐫:  can approve low risk, can escalate medium/high 𝐏𝐨𝐥𝐢𝐜𝐲/𝐑𝐞𝐯𝐞𝐧𝐮𝐞 𝐀𝐩𝐩𝐫𝐨𝐯𝐞𝐫:  required to approve compensation or high-cost alternatives. 𝐒𝐲𝐬𝐭𝐞𝐦 𝐞𝐧𝐟𝐨𝐫𝐜𝐞𝐦𝐞𝐧𝐭: action blocked if role tries to execute outside its rights. 🤔 𝐖𝐡𝐲 𝐭𝐡𝐢𝐬 𝐦𝐚𝐭𝐭𝐞𝐫𝐬 𝐧𝐨𝐰 👉 The hard part is no longer only retrieval quality. 👉 The hard part is temporal validity, decision traceability, and enforceable control across the full decision lifecycle. So yes, context graphs are valuable. But for enterprise AI, the real architecture is multi-layered: meaning+governance+operations+provenance+permissions. A new era of enterprise architecture is emerging: technology, data, and business architecture are converging into one decision-ready context foundation.

Learn what an agentic workflow is, how it differs from traditional workflows, and the 4 core patterns to build reliable systems in production.

Only 7% of enterprises have AI-ready data. Learn how a semantic layer built on knowledge graphs pushes accuracy from 80% to 95%+.

Imagine you’re an explorer searching an old mine for abandoned treasure. You have in your possession an old map that shows all the chambers and the tunnels connecting them. Each chamber is a node in our graph, and each tunnel… Read more →

Discover what’s new in graph tech this week: 2 graph updates from Microsoft, an RDF knowledge graph creator, and a graph visualization of the Epstein Files.

Security is entering a new era, one defined by explosive data growth, increasingly sophisticated threats, and the rise of AI-enabled operations. To keep pace, security teams need an AI-powered approach to collect, reason over, and act on security data at scale. At RSA Conference 2026 (RSAC), we’re unveiling the next wave of Sentinel innovations designed to help organizations move faster, see deeper, and defend smarter with AI-ready tools. These updates include AI-driven playbooks that accelerate SOC automation, Granular Delegated Admin Privileges (GDAP) and granular role-based access controls (RBAC) that let you scale your SOC, accelerated data onboarding through new connectors, and data federation that enables analysis in place without duplication. Together, they give teams greater clarity, control, and speed. Come see us at RSAC to view these innovations in action. Hear from Sentinel leaders during our exclusive Microsoft Pre-Day, then visit Microsoft booth #5744 for demos, theater sessions, and conversations with Sentinel experts. Read on to explore what’s new. See you at RSAC! Sentinel feature innovations: Sentinel SIEM Sentinel data lake Sentinel graph Sentinel MCP Threat Intelligence Microsoft Security Store Sentinel promotions Sentinel SIEM Playbook generator [Now in public preview] The Sentinel playbook generator delivers a new era of automation capabilities. You can vibe code complex automations, integrate with different tools to ensure timely and compliant workflows throughout your SOC and feel confident in the results with built in testing and documentation. Customers and partners are already seeing benefit from this innovation. “The playbook generator gives security engineers the flexibility and speed of AI-assisted coding  while delivering the deterministic outcomes that enterprise security operations require. It's the best of both worlds, and it lives natively in Defender where the engineers already work.” – Jaime Guimera Coll | Security and AI Architect | BlueVoyant Learn more about playbook generator. SIEM migration experience [General availability now] The Sentinel SIEM migration experience helps you plan and execute SIEM migrations through a guided, in-product workflow. You can upload Splunk or QRadar exports to generate recommendations for best‑fit Sentinel analytics rules and required data connectors, then assess migration scope, validate detection coverage, and migrate from Splunk or QRadar to Sentinel in phases while tracking progress. “The tool helps turn a Splunk to Sentinel migration into a practical decision process. It gives clear visibility into which detections are relevant, how they align to real security use cases, and where it makes sense to enable or prioritize coverage—especially with cost and data sources in mind.” – Deniz Mutlu | Director | Swiss Post Cybersecurity Ltd Learn more about SIEM migration experience. GDAP, unified RBAC, and row-level RBAC for Sentinel [Public preview, April 1] As Sentinel environments grow for enterprises, MSSPs, hyperscalers, and partners operating across shared or multiple environments, the challenge becomes managing access control efficiently and consistently at scale. Sentinel’s expanded permissions and access capabilities are designed to meet these needs.  Granular Delegated Admin Privileges (GDAP) lets you streamline management across multiple governed tenants using your primary account, based on existing GDAP relationships. Unified RBAC allows you to opt in to managing permissions for Sentinel workspaces through a single pane of glass, configuring and enforcing access across Sentinel experiences in the analytics tier and data lake in the Defender portal. This simplifies administration and improves operational efficiency by reducing the number of permission models you need to manage. Row-level RBAC scoping within tables enables precise, scoped access to data in the Sentinel data lake. Multiple SOC teams can operate independently within a shared Sentinel environment, querying only the data they are authorized to see, without separating workspaces or introducing complex data flow changes. Consistent, reusable scope definitions ensure permissions are applied uniformly across tables and experiences, while maintaining strong security boundaries. To learn more, read our technical deep dives on RBAC and GDAP. Sentinel data lake Sentinel data federation [Public preview, April 1] Sentinel data federation lets you analyze security data in place without copying or duplicating your data. Powered by Microsoft Fabric, you can now federate data from Fabric, Azure Data Lake Storage (ADLS), and Azure Databricks into Sentinel data lake. Federated data appears alongside native Sentinel data, so you can use familiar tools like KQL hunting, notebooks, and custom graphs to correlate signals and investigate across your entire digital estate, all while preserving governance and compliance. You can start analyzing data in place and progressively ingest data into Sentinel for deeper security insights, advanced automation, and AI-powered defense at scale. You are billed only when you run analytics on federated data using existing Sentinel data lake query and advanced insights meters. Figure 1: Federated data appears alongside native Sentinel tables for unified investigation and hunting Sentinel cost estimation tool [Public Preview, April 9] The new Sentinel cost estimation tool offers all Microsoft customers and partners a guided, meter-level cost estimation experience that makes pricing transparent and predictable. A built-in three-year cost projection lets you model data growth and ramp-up over time, anticipate spend, and avoid surprises. Get transparent estimates into spend as you scale your security operations. All other customers can continue to use the Azure calculator for Sentinel pricing estimates. See the Sentinel pricing page for more information. Figure 2: Plan your Sentinel investment with guided, meter-level cost breakdowns and three-year projections Sentinel data connectors A365 Observability connector [Public preview, April 15] Bring AI agent telemetry into the Sentinel data lake to investigate agent behavior, tool usage, prompts, reasoning and execution using hunting, graph, and MCP workflows. GitHub audit log connector using API polling [General availability, March 6] Ingest GitHub enterprise audit logs into Sentinel to monitor user and administrator activity, detect risky changes, and investigate security events across your development environment. Google Kubernetes Engine (GKE) connector [General availability, March 6] Collect Google Kubernetes Engine (GKE) audit and workload logs in Sentinel to monitor cluster activity, analyze workload behavior, and detect security threats across Kubernetes environments. Microsoft Entra and Azure Resource Graph (ARG) connector enhancements [Public preview, April 15] Enable new Entra assets (EntraDevices, EntraOrgContacts) and ARG assets (ARGRoleDefinitions) in existing asset connectors, expanding inventory coverage and powering richer, built‑in graph experiences for greater visibility. With over 350 Sentinel data connectors, customers achieve broad visibility into complex digital environments and can expand their security operations effectively. “Microsoft Sentinel data lake forms the core of our agentic SOC. By unifying large volumes of Microsoft and third-party data, enabling graph-based analysis, and supporting MCP-driven workflows, it allows us to investigate faster, at lower cost, and with greater confidence.” – Øyvind Bergerud | Head of Security Operations | Storebrand Learn more about Sentinel data connectors. Sentinel connector builder agent using Sentinel Visual Studio Code extension [Public preview, March 31] Build Sentinel data connectors in minutes instead of weeks using the AI‑assisted Connector Builder agent in Visual Studio Code. This low‑code experience guides developers and ISVs end-to-end, automatically generating schemas, deployment assets, connector UI, secure secret handling, and polling logic. Built‑in validation surfaces issues early, so you can validate event logs before deployment and ingestion. Example prompt in GitHub Copilot Chat: @sentinel-connector-builder Create a new connector for OpenAI audit logs using https://api.openai.com/v1/organization/audit_logs   Figure 3: The Sentinel Connector Builder agent in VS Code guides you through building a data connector end-to-end, from API documentation to deployment-ready assets Data filtering and splitting [Public preview, March 30] As security teams ingest more data, the challenge shifts from scale to relevance. With filtering and splitting now built into the Defender portal, teams can shape data before it lands in Sentinel, without switching tools or managing custom JSON files. Define simple KQL‑based transformations directly in the UI to filter low‑value events and intelligently route data, making ingestion optimization faster, more intuitive, and easier to manage at scale. Filtering at ingest time allows you to remove low-value or benign events to reduce noise, cut unnecessary processing, and ensure that high-signal data drives detections and investigations. Splitting enables intelligent routing of data between the analytics tier and the data lake tier based on relevance and usage. Together, these two capabilities help you balance cost and performance while scaling data ingestion sustainably as your digital estate grows. Create workbook reports directly from the data lake [Public preview, April 1] Sentinel workbooks can now directly run on the data lake using KQL, enabling you to visualize and monitor security data straight from the data lake. By selecting the data lake as the workbook data source, you can now create trend analysis and executive reporting. Sentinel graph Custom graphs [Public preview, April 1] Custom graphs let you build tailored security graphs tuned to your unique security scen

The origins of the semantic web, the rise of enterprise knowledge graphs, their relationship to AI, and what it takes to get started.

Much of what an organization’s employees know isn’t written down. With a semantic layer, you can transform this hidden expertise into valuable assets to power AI.

Agentic architecture is the system design that enables agentic systems to work efficiently towards the goal. Learn architecture patterns that work in production.Secondary keywords:

By storing your data as a knowledge graph rather than as flat embeddings, you give your LLM the ability to reason about how entities relate to one another.

This release makes it even more easy than before to extract knowledge from documents and shortens the path to visualization and backend storage. For example, if your Neo4j is running locally and you have extracted some web page or pdf, you can use It will also create the necessary indexes and constr

Using Cursor I just wrote a plugin for the Protege Ontology Editor -- to add a feature I was dearly missing in the OWL language, namely Roles. The Web Ontology Language (OWL) supports the typical modeling elements, such as a Class (taxonomy), Properties and Relationships. But, what is missing in OWL are Roles. For example: A teacher is a role a person plays. In vanilla OWL you are forced to represent teacher as a class, but semantically, teacher is better understood as a role a person can play at times. Upper ontologies such as BFO and UFO recognize the need for Roles and formally describe their unique semantics. OWL has a great feature though: annotations -- essentially metadata associated with modeling elements such as classes. Using these I can indicate in a class definition that a protege class is in fact a role, and what class(es) can play that role. And, annotations is handled via new Role UI Tab in Protege, implemented as a Protege plugin. It's amazing how easy it now is to tailor open source tools to what we need! https://lnkd.in/dCqnaiHh Juan Sequeda Kurt Cagle Ora Lassila Dean Allemang Jessica Talisman

LoanGuard AI: Graph-based agentic AI for compliance & investigation - Part 2 of 2

LoanGuard AI: Graph-based agentic AI for compliance & investigation — Part 1 of 2

OrionBelt Ontology Builder hits v1.0.0 After weeks of telling myself "just one more feature," I finally ran out of excuses. OrionBelt is a browser-based OWL ontology workbench. No Java, no Protégé, no 2GB RAM just to open a file. Streamlit + rdflib, pip install, done. What it does well:  - Full CRUD for classes, properties, individuals, and restrictions  - SKOS vocabulary support (because apparently everyone needs a thesaurus)  - Bulk operations (because life is too short to add 50 classes one by one)  - Import with actual diff review (so you see what you're about to break)  - Undo/redo (because we all make mistakes, some of us more often)  - Interactive graph visualization (the part that makes the screenshots look good) What it doesn't do: replace Protégé for hardcore ontology engineering. It's not trying to. It's for people who want to get ontology work done without fighting their tools. Pairs nicely with its sibling project OrionBelt Analytics — an MCP server that analyzes your database schemas (PostgreSQL, Snowflake, Dremio) and generates OWL ontologies from them. So you can go from "I have tables" to "I have a proper ontology" without writing a single triple by hand. Try it: https://lnkd.in/dERFebCy GitHub: https://lnkd.in/egGZJHiP Analytics: https://lnkd.in/dkbq-aqC PyPI: pip install orionbelt-ontology-builder Built with a lot of coffee (? :D) and a healthy distrust of overcomplicated software. #ontology #semanticweb #owl #skos #knowledgegraph #opensource #python #streamlit #mcp

Evaluating LLM-based Knowledge Graph Extraction

The Data Industry’s Semantic Layer. The history of the Semantic Layer dates back to the 1990's and Business Objects. How did we arrive at today's Semantic Layer?