How to Find a Killer Using Graphs: What Sherlock Holmes Was Really Doing with That Red String
summary
A digital murder-board called Networks Notebook converts PDFs, images and OCR into a property knowledge graph, extracts entities and relations with LLMs or NLP, then runs precise Cypher queries in Neo4j and interprets results with a two-stage analysis that combines “cold” graph pattern detection with generative summaries. Demonstrated on the Polish translation of the murder_of_Roger_Ackroyd, it compresses half a book into a small graph, surfaces contradictory beliefs, three independent info sources, and finds forensic patterns analogous to AML flows. Repository is commercial.
AI-written summary · about
chapters (10)
- 0:00Introduction to graphs and detectives
- 1:09Building knowledge graphs from text
- 3:21Tools for graph extraction
- 5:49Small language models for niche data
- 8:18From raw data to actionable insights
- 10:48Why relations matter in graphs
- 13:40Agentic approaches to graph analysis
- 15:45Pattern-based crime investigation
- 17:32Case study: Murder of Roger Ackroyd
- 22:57Q&A on tools and applications
related talks
13mCombating Gun Trafficking and Gun Violence With Graph Analytics and Visualization
45mConnecting the dots to go from tabular security incident data to behavioral graph understanding
48mHunting Stealth Adversaries with Graphs & AI
20mAttack of the Graph: Visual Tools for Cyber Analysis
29mBuilding a Graph User-Interface for Malware-Analysis
29mKnowledge Graphs in Action: Identifying Sanctions Circumvention in a Complex Global Landscape
27mHow Trustworthy is Your Knowledge Graph?
30mCybersecurity Risk Assessment Using LLM Agents and Graph Data Science