Tags: * need new reviewing criteria * change citation style * NLP reviewers do not have sufficient ML expertise * reviewer performance: flag bad reviewers
"Make it possible to flag reviewers as bad. ARR compared to ICLR, ICML or NeurIPS has very opinionated
reviewers that often lack ML understanding and insist things should be named after NLP specific publications
they are aware of, even if the actual concept was taken from CV or ML and simply renamed in NLP for
novelty. Especially when the novelty is on the ML side. ML conference reviewers are significantly better
at providing relevant references, experiment suggestions and overall understanding of ML concepts while
NLP reviewers focus too much on NLP trends without checking if the underlying assumptions are useful.
To that end. Remove the year and author name annotation from the citation style. Recency or gravitas
are not a good indicator nor incentive for the relatedness of work and produce hubness issues in citations
and research directions — aka, the incremental work issue every lab is complaining about. Focus reviews
on content and deep analysis, less on fame and attention attribution to remove the stagnation."