What do you get when you mix 10 Knowledge Graph experts, a thoughtful and engaged audience, some AI, 90 minutes of conversation, and 60 minutes of post-processing?
Lots of insights, and a quick recap.
We thoroughly enjoyed our Roundtable yesterday.
Huge thanks to our esteemed Chairs and Program Committee members as well as everyone who joined the conversation and kept it going with questions, comments and insights.
Summarizing the takeways from these sessions always takes time. It may well be the hardest, but also the most valuable part of it all. This is why we'll definitely do it, and share with everyone in due time.
However, we thought we'd run an experiment to see how all these AI-powered tools that we so often talk about can work in practice. Here's what we did, and what we found out.
1. Took the recording of the Roundtable and extracted the transcript. Speaker identification was not perfect, and the transcript needed some polishing. Total time spent: 15 minutes.
2. Created an account on Neo4j Aura, and fed the annotated extracted transcript to the newly released
@neo4j LLM Knowledge Graph Builder to Extract Nodes and Relationships from Unstructured Text tool using GPT 4o. Total time spent: 20 minutes.
3. Explored the generated Knowledge Graph and tweaked auto-generated queries and visualization to produce something representative of the conversation. Total time spent: 20 minutes.
This is a picture of the end result.
It was a fun exercise, and the idea was to strictly contain it to 60 minutes or less. None of these steps worked entirely flawlessly, except maybe for #2.
The million dollar question is - does this work? Is it useful?
We think the answer reflects the state of AI tools at this point: Yes, if you know what you are doing and can jump in to verify and enhance.
Focusing on the Knowledge Graph creation, which is the most interesting part, it's a good starting point and certainly impressive compared to what was the state of the art previously. Still, you should not blindly trust the outcome. This picture is far from perfect - there are topics, people, and interaction missing.
There was both redundancy and missing entities and relationships in the auto-generated Knowledge Graph. Domain expertise and technical knowledge are both required for a production-ready outcome. And obviously, the quality of the input matters - pre-processing was necessary.
Remember though: This was all done for free, in under 60 minutes.
Expect a podcast episode release and a writeup with our takeways coming your way in the coming days.
Happy 4th of July to our US-based friends, and a smooth election day for everyone based in the UK.
#KnowledgeGraphs #GraphAI #GraphDB #SemanticTech #DataScience #Analytics #DataViz #AMA #GenAI #AI #LLM #Freebies #Events @pacoid @PAlexop