performance for RAG with Neo4j knowledge graph depends from how texts was transformed to graph
do yo have survey/ comparison for different algorithms how to convert text to graph ?
for example
1
sparse ( tfidf or count vectorizer ) vs dense (LLM bert or not bert GPT) embeddings
2
hybrid : both sparse ( tfidf or count vectorizer ) vs dense (LLM bert or not bert GPT) embeddings , with different waits
3
different prompts to convert text to dense LLM embedding
4
big text (many files , less files but big files ) and small text
5
synonyms
shallow nodes similarity ( only similar to next node ) and deep nodes similarity
6
etc
to avoid
https://contextual.ai/introducing-rag2/
A typical RAG system today uses a frozen off-the-shelf model for embeddings, a vector database for retrieval, and a black-box language model for generation, stitched together through prompting or an orchestration framework. This leads to a “Frankenstein’s monster” of generative AI: the individual components technically work, but the whole is far from optimal.
see also
https://www.linkedin.com/pulse/data-science-machine-learning-thoughts-quotes-sander-stepanov/?trackingId=IUH7lVdxTPS%2BJcZX%2FYf7oA%3D%3D
performance for RAG with Neo4j knowledge graph depends from how texts was transformed to graph do yo have survey/ comparison for different algorithms how to convert text to graph ? for example 1 sparse ( tfidf or count vectorizer ) vs dense (LLM bert or not bert GPT) embeddings 2 hybrid : both sparse ( tfidf or count vectorizer ) vs dense (LLM bert or not bert GPT) embeddings , with different waits 3 different prompts to convert text to dense LLM embedding 4 big text (many files , less files but big files ) and small text 5 synonyms shallow nodes similarity ( only similar to next node ) and deep nodes similarity 6 etc to avoid https://contextual.ai/introducing-rag2/ A typical RAG system today uses a frozen off-the-shelf model for embeddings, a vector database for retrieval, and a black-box language model for generation, stitched together through prompting or an orchestration framework. This leads to a “Frankenstein’s monster” of generative AI: the individual components technically work, but the whole is far from optimal. see also https://www.linkedin.com/pulse/data-science-machine-learning-thoughts-quotes-sander-stepanov/?trackingId=IUH7lVdxTPS%2BJcZX%2FYf7oA%3D%3D