I have a bilateral dataset that includes countries and their the weight of their relation. I'd like to calculate the similarity of countries in 1) who their trade parterns are and 2) the weight of the relation to those partners. For this purpuse, I've calculated the weighted Jaccard similarity of the countries.

However, my data includes negative values. I know results from Jaccard similiary is supposed to range between 0-1, but I don't know whether results from the weighted Jaccard similiary should also be between 0-1. My question is if I need to normalize my data before running the weighted Jaccard similarity or if I can run it with negative values (which runs fine in code but also generates similarity results with negative values)?

If I should normalize, would min-max normalization be the best option?

Thank you!

  • $\begingroup$ The standard version of Jaccard similarity measures similarity of two sets, therefore the elements of the corresponding vectors x and y, where x_i indicates whether element i belongs to the first set and y_i indicates whether i belongs to the second set, are assumed to be 0/1, integer, or at least non-negative real numbers. $\endgroup$
    – Valentas
    Apr 16 at 20:20


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.