Yeah, there are some good survey articles that review various methods. Here are some links you can check out: https://cloud.r-project.org/web/packages/correlat

Yeah, there are some good survey articles that review various methods. Here are some links you can check out: https://cloud.r-project.org/web/packages/correlation/vignettes/types.html https://cloud.r-project.org/web/packages/correlation/vignett... https://juniperpublishers.com/bboaj/pdf/BBOAJ.MS.ID.555795.pdf https://juniperpublishers.com/bboaj/pdf/BBOAJ.MS.ID.555795.p... https://www.stats.ox.ac.uk/~cucuring/Lecture_2_Correlations_Dependence.pdf https://www.stats.ox.ac.uk/~cucuring/Lecture_2_Correlations_... https://blogs.sas.com/content/iml/2021/04/28/compute-hoeffding-d-statistic.html https://blogs.sas.com/content/iml/2021/04/28/compute-hoeffdi... Essentially, cosine similarity just measures the extent to which the vectors point in the same direction in N dimensional space. That turns out to be quite discriminative when N is very high (i.e., over 1000). But it is not as good for finding complex non-linear relationships.