Hailong, I would still suggest following MathSciNet, but you don't necessarily have to go there every time. Presumably the journals you cite come from a small finite set. This may be a pain at first, but you will get them all very soon.
Actually, I use BibTeX and so I have a growing database of papers. You can go to MathSciNet and ask it to display references in BibTeX format. Then you can copy and paste. After a while most papers you cite are already in your database, except recent ones, but those you can get when needed. It's pretty convenient and easy. You could also use John Palmieri's bibweb script.
As for the arXiv links, I am not sure that you need to keep them there. If the author and title are the same, which is usually the case, then it is fairly easy to find on arxiv, so if someone does not have access to the published version, then they can easily find it on arXiv.org. If you want to be thorough you can include the doi information and/or the MR number.
