This is great guys. I've been actually working on extracting entities and relationships from a lot of web scrapped data by using LLMs. I was not really convinced of the hype of MS GraphRAG and it is too expensive just to try to find out if it fitted my needs or not so I went for creating a basic ontology of the domain. The result was quite decent vs just throwing a bunch of text into the LLM and expect it to understand all the contex by itself. This also helps a lot with entity de-duplication and makes the outcome of each iteration more repeatable. Ontologies is the key.
Would love to see more about going from Semi-Structured / Structured formats with mixed numerical + text in tabular to knowledge Graphs and also Temporal Graphs 🙂
Maybe a future episode could present a demonstration of concept of a platform for collective intelligence. The platform would keep humans in the loop through a conversational interface that merges public parts of conversations into a shared graph representation. It would report on which issues are discussed by the most people, aggregate measures for sentiment about proposed solutions, and how those measures of sentiment change over time. One big challenge is how the platform would authenticate the identities of humans at the same time that it allows people to share selected parts of conversations anonymously.
Thank you! Check out our last few episodes of Season 1 where we talk about this: neo4j.com/video/going-meta-a-series-on-graphs-semantics-and-knowledge/
This is great guys. I've been actually working on extracting entities and relationships from a lot of web scrapped data by using LLMs. I was not really convinced of the hype of MS GraphRAG and it is too expensive just to try to find out if it fitted my needs or not so I went for creating a basic ontology of the domain. The result was quite decent vs just throwing a bunch of text into the LLM and expect it to understand all the contex by itself. This also helps a lot with entity de-duplication and makes the outcome of each iteration more repeatable. Ontologies is the key.
Yes! Thank you!
PUMPED! HAPPY TO SEE YA'LL BACK!🐉
Thank you! Happy to be back!
Yeah!
Would love to see more about going from Semi-Structured / Structured formats with mixed numerical + text in tabular to knowledge Graphs and also Temporal Graphs 🙂
Maybe a future episode could present a demonstration of concept of a platform for collective intelligence. The platform would keep humans in the loop through a conversational interface that merges public parts of conversations into a shared graph representation. It would report on which issues are discussed by the most people, aggregate measures for sentiment about proposed solutions, and how those measures of sentiment change over time. One big challenge is how the platform would authenticate the identities of humans at the same time that it allows people to share selected parts of conversations anonymously.
Thank you for that!
If it possible in next episodes, please tell about the most common technics how we can use graph in RAG flow
Thank you! Check out our last few episodes of Season 1 where we talk about this: neo4j.com/video/going-meta-a-series-on-graphs-semantics-and-knowledge/
Is there a way to extract a graph that considers a well-defined ontology, including a complete class definition with properties within the nodes?
we will try to cover that next episode :)
@@neo4jplus 1. Thanks for making this series! Exactly the content I was looking for.