Personal preference? I am not sure who/why decided on this order (you have given no context). In general one would want to look at RefSeq/SwissProt (as you point out) first since those are manually curated but nr is obviously going to encompass a much wider selection of sequences.
At the end of the day, proposed identity for a sequence is a hypothesis and nothing more. That eventually needs to be independently experimentally proven (e.g. by a knockout).

Thank you very much for your answer, but I still have one more question. There are plenty more examples for those priority order of Nr, Swiss-Prot, KEGG, and KOG/COG(following: examples 1,2,3), is it because NCBI's nr database comprised of sequences obtained from translations of GenBank coding sequences, include Swiss-Prot?

1. Nan G, Zhang Y, Li S, et al. NaCl stress-induced transcriptomics analysis of Salix linearistipularis (syn. Salix mongolica)[J]. Journal of Biological Research-Thessaloniki, 2016, 23(1): 1.
2. Iseli C, Jongeneel C V, Bucher P. ESTScan: a program for detecting, evaluating, and reconstructing potential coding regions in EST sequences[C]//ISMB. 1999, 99: 138-148.
3. Qin Y F, Fang H M, Tian Q N, et al. Transcriptome profiling and digital gene expression by deep-sequencing in normal/regenerative tissues of planarian Dugesia japonica[J]. Genomics, 2011, 97(6): 364-371.

thank you for your attention