In the paper 'A Simple and Strongly-Local Flow-Based Method for Cut Improvement', the authors mentioned that the strongly-local spectral methods have shown much stronger results, where the algorithm doesn’t even access most of the nodes of the graph and thus the running time is dependent on the size of the seed set or output set, rather than the entire graph. These strongly-local spectral algorithms are very fast in both theory and practice, which is supported by another paper that you've read, please provide the full name of that paper.