Separating interpretation, transformation and retrieval into layers is the clarification this conversation needed. Building the search for a scoping review is where it bites: a beautifully formed Boolean strategy built from a slightly misread question returns thousands of clean, precise, irrelevant records, and you do not find that out until you are deep into screening. Being able to inspect and edit the generated query before it runs is the part that actually matters.
I've been reading most of you post now looking at, and slowly better grasping, how Ovid Discovery work. It claims to use both vector and RAG in two different of its AI search tools - which are part of its package rather than an add on.
When I teach our users how to best use it, I am clear about using this only as a scoping tool, and the need to determine the referenced lit accuracy against the search question. I then show they how to use the results to either find additional articles, or (if there are as far too many articles no longer have subject headings) build Boolean searches from these.
Far too many of my users use OpenEvidence and Consensus without ever validating the results, so I am hoping that I am at least giving them some pause now in those searches as well.
I wpuld say the vector search and RAG pieces are doing somewhat different jobs.
The vector component is about retrieving/ranking potentially relevant material by semantic similarity, while the RAG component uses retrieved material (whatever the method eg lexical or vector search or both) to generate an answer or synthesis.
Yes, this is what happens. First is uses the vector component when the prompt question is asked to retrieve literature from all the journals within the databases collected in the Discovery platform. Then once one has selected articles from the result it provides "Essential Insights" from those that address the prompt.
It provides an Overview from the vector search that includes linked references, meant to include that above.
Separating interpretation, transformation and retrieval into layers is the clarification this conversation needed. Building the search for a scoping review is where it bites: a beautifully formed Boolean strategy built from a slightly misread question returns thousands of clean, precise, irrelevant records, and you do not find that out until you are deep into screening. Being able to inspect and edit the generated query before it runs is the part that actually matters.
Thanks Aaron!
I've been reading most of you post now looking at, and slowly better grasping, how Ovid Discovery work. It claims to use both vector and RAG in two different of its AI search tools - which are part of its package rather than an add on.
When I teach our users how to best use it, I am clear about using this only as a scoping tool, and the need to determine the referenced lit accuracy against the search question. I then show they how to use the results to either find additional articles, or (if there are as far too many articles no longer have subject headings) build Boolean searches from these.
Far too many of my users use OpenEvidence and Consensus without ever validating the results, so I am hoping that I am at least giving them some pause now in those searches as well.
I wpuld say the vector search and RAG pieces are doing somewhat different jobs.
The vector component is about retrieving/ranking potentially relevant material by semantic similarity, while the RAG component uses retrieved material (whatever the method eg lexical or vector search or both) to generate an answer or synthesis.
Yes, this is what happens. First is uses the vector component when the prompt question is asked to retrieve literature from all the journals within the databases collected in the Discovery platform. Then once one has selected articles from the result it provides "Essential Insights" from those that address the prompt.