I got interested in KG’s when first learning vibe coding bc I thought it could be a powerful and efficient way for an LLM to map a user’s interests and personality. Maybe I’ll go back to that.
Every project is different, including ML. I just start at the start and gradually turn up the volume to full. I’m not super impressed w graph databases, and have felt that way for a long time, but graphrag made them relevant to me. But I have been enjoying learning about them. I just prefer my programmatic approaches like on earlier days of this series.
I did without a graph databases even at nearly trillion scale, so design really impacts outcome, and really deep understanding of how to work with graphs can lead to some shortcuts and opportunities.
I miss ML. Lol. Gosh I miss ML. Don’t get to do it as much anymore.
Haha then we should collab on a finance one. I’m pretty sure the previous ML ones for the market didn’t try all the parameters I have in mind. But i need to review the details of the literature.
Excellent. In my next articles, I am going to be working on creating Knowledge Graphs from scratch, trying different things. Good timing! The recent Cypher articles gave me the learning I needed, and now I am interested in pushing some large networks to Neo4j and testing out some of its graph algorithms.
Yeah, they are a powerful way to map a user's interests, personality, and other things. There is a lot of room for creativity. The problem is there is probably too much room for creativity, and so that is going to be fun to find the right balance and design for optimal AI use.
Yup those connections can add up quickly eh? What are the processing needs like compared to the average ML project? Not like training an LLM but the more moderate ones like analyzing finance or sports.
Yup those connections can add up quickly eh? What are the processing needs like compared to say the average ML project? And I guess I mean not like training an LLM but the more moderate ones like analyzing finance or sports.
Brilliant walkthrough of the GraphDB-to-visualization workflow! Your approach to building edgelist DataFrames with those flexible relationship patterns is really elegant,especially how you can pivot between different topic filters to explore various collaboration networks. The fact that you immediately saw structural clarity in the 13k edge network with Cosmograph's latest release speaks volumes aboutthe tool's capability. I'm curious whether you've experimented with any graph metrics onthese collaboration networks before visualization, or if you find the visual exploration sufficient for pattern discovery at this stage?
In my earlier days in this series, I don’t use graph databases. I wanted to explain how to do things programmatically. The work is much more powerful. I use a lot of network science metrics and SNA methods. I wrote a book on the subject as well. I am much more comfortable without a graph database, and have done at nearly trillion-scale.
Thanks for the comment! Nice to have some discussion!
I got interested in KG’s when first learning vibe coding bc I thought it could be a powerful and efficient way for an LLM to map a user’s interests and personality. Maybe I’ll go back to that.
Every project is different, including ML. I just start at the start and gradually turn up the volume to full. I’m not super impressed w graph databases, and have felt that way for a long time, but graphrag made them relevant to me. But I have been enjoying learning about them. I just prefer my programmatic approaches like on earlier days of this series.
I did without a graph databases even at nearly trillion scale, so design really impacts outcome, and really deep understanding of how to work with graphs can lead to some shortcuts and opportunities.
I miss ML. Lol. Gosh I miss ML. Don’t get to do it as much anymore.
Haha then we should collab on a finance one. I’m pretty sure the previous ML ones for the market didn’t try all the parameters I have in mind. But i need to review the details of the literature.
Sounds fun! Happy to help!
Excellent. In my next articles, I am going to be working on creating Knowledge Graphs from scratch, trying different things. Good timing! The recent Cypher articles gave me the learning I needed, and now I am interested in pushing some large networks to Neo4j and testing out some of its graph algorithms.
Yeah, they are a powerful way to map a user's interests, personality, and other things. There is a lot of room for creativity. The problem is there is probably too much room for creativity, and so that is going to be fun to find the right balance and design for optimal AI use.
Yup those connections can add up quickly eh? What are the processing needs like compared to the average ML project? Not like training an LLM but the more moderate ones like analyzing finance or sports.
Yup those connections can add up quickly eh? What are the processing needs like compared to say the average ML project? And I guess I mean not like training an LLM but the more moderate ones like analyzing finance or sports.
Brilliant walkthrough of the GraphDB-to-visualization workflow! Your approach to building edgelist DataFrames with those flexible relationship patterns is really elegant,especially how you can pivot between different topic filters to explore various collaboration networks. The fact that you immediately saw structural clarity in the 13k edge network with Cosmograph's latest release speaks volumes aboutthe tool's capability. I'm curious whether you've experimented with any graph metrics onthese collaboration networks before visualization, or if you find the visual exploration sufficient for pattern discovery at this stage?
Same dataset but two years ago. https://open.substack.com/pub/100daysofnetworks/p/day-19-of-100daysofnetworks?r=du0vp&utm_medium=ios
In my earlier days in this series, I don’t use graph databases. I wanted to explain how to do things programmatically. The work is much more powerful. I use a lot of network science metrics and SNA methods. I wrote a book on the subject as well. I am much more comfortable without a graph database, and have done at nearly trillion-scale.
Thanks for the comment! Nice to have some discussion!