Paco N.

Join us for a talk about "How To Construct Knowledge Graphs from Unstructured Data" hosted by GraphGeeks.org https://live.zoho.com/PBOB6fvr6c Wed Aug 14, 09:00 US Pacific We'll walk you through the broader practices of building and updating knowledge graphs, while leveraging open source libraries for state-of-the-art tooling (hint: lots of transformer models, not LLMs). Techniques such as "Graph-enhanced RAG" help mitigate hallucinations in the AI apps downstream. This is especially important for use in highly regulated environments where KG construction must be accountable, explainable, based on evidence, etc. If you've been hearing terms such as "named entity recognition", "entity resolution", "lexical graph", "entity linking", "textgraphs", "semantic random walks", "embeddings", "relation extraction", "semantic layer", and so on, and ever wondered how these fit together into practical solutions ... join us, that's what we'll cover! In other words, a full deconstruction of the elements of KG construction leading into GraphRAG. Some component libraries include: spaCy, NetworkX, Pandas, GLiNER, GLiREL, ReLIK, Pydantic, LanceDB -- and if you want a sneak peek at the code: https://github.com/DerwenAI/strwythura BTW, the graph visualization shown here is what we'll construct from articles about linkages between eating processed red meat frequently and the risk of developing dementia. We'll build out a KG and vector DB suitable for GraphRAG grounding in an LLM chatbot or agent. #ERKG #KnowledgeGraph #UnstructuredData #DataScience #MachineLearning #TechWorkshops

Thank you Hugo (., the KnowHax Pitch Event was fantastic!! Congrats on the teams competing.

If this might be of interest, here's an article we're circulating for discussion, about abstraction layers in knowledge graphs: https://blog.derwen.ai/graph-levels-of-detail-ea4226abba55 Colleagues from enterprise teams are interested in the possibility of these kinds of features, i.e., abstraction layers in KGs, and some discussions at the recent K1st World led to writing this. Use cases include: enhancing graph ML, making large scale graphs more manageable, and informing visualization layouts to be more meaningful. Granted, this is early stage -- there aren't any such features available yet, but exploring cost-effective methods is needed long before that. The intention for this article is to circulate for review and gain feedback: are any definitions misrepresented, or any points being misconstrued? And to gauge interest from a broader audience.