i would like to identify a set of reads (that i will later assemble regarding their taxonomic affiliations), and i've tested sequence alignment (BLAST) and read mapping (Bowtie2). The thing is, i'm not sure which one i should trust for my read classification.
Well, blast is a bit more sensitive than bowtie2, however bowtie2 is vastly faster than blast. So, what to choose is based more on how much data you have to go through and how many computers you can throw at the problem (unless time isn't an issue, in which case enjoy just using blast).
BTW, both blast and bowtie2 produce alignments (in addition to mapping sequences).
I agree with the Devon's answer. An other thing to take in account is which kind or read you have, pair-ended and long reads, or short reads for example.
You have here a benchmark (2015) about NGS short read mappers, it may helps you to determine which one you can use, and what are the differences between these tools.
I hope this helps you! ;)