I've been using ResearchRabbit for several months now, primarily for literature discovery in my FinOps and cloud cost research. While the visualization and author mapping are strong features, I've found the core 'similar work' recommendation engine to be surprisingly inconsistent.
My methodology involves testing its suggestions against known seminal papers in a domain. For instance, when starting from a well-cited paper on AWS Reserved Instance pricing optimization, the suggestions often drift into tangential areas after one or two hops. The connections seem to be based heavily on keyword overlap or shared authors, rather than a nuanced understanding of the paper's core contribution. This leads to:
* **Surface-level similarity:** Suggests papers that mention "Reserved Instances" but in the context of general capacity planning, missing the specific financial analysis angle.
* **Rapid topic drift:** The second-degree recommendations can veer into unrelated infrastructure or even different cloud providers without the same cost model, which isn't helpful for focused research.
* **Missed key works:** It occasionally overlooks highly relevant papers that are methodologically aligned but use slightly different terminology.
This is problematic for a tool marketed on discovering "similar work." The algorithm appears to prioritize lexical matching and co-citation networks over semantic understanding of the research problem. For academic research, where precision in literature review is critical, this can create significant noise and require extensive manual filtering.
I'm curious if others in technical fields have experienced this. Have you developed specific workflows or filters to improve the relevance of the suggestions, or do you primarily use the tool for its visualization capabilities instead?
—EK
Your bill is too high.