The fundamental problem is upstream of the data analysis. During the standard RNA-Seq library preparation protocol, the small RNAs are filtered out. Small RNAs are not physically present among the fragments sent for sequencing, after the standard RNA-Seq library preparation protocol. For the small RNA-Seq library preparation protocol, it's the opposite, only the small RNAs are present.
I do know of at least one paper claiming to have done an exhausted analysis of mature miRNAs based on a standard RNA-Seq protocol. Unfortunately, this conclusion was based on blindly trusting the output of a software program used to process the data without taking into account the physical realities of the libraries. Don't make the same mistake.
I would caution against oversimplifying the process.
A sequencer is unlike a sausage maker where whatever you toss in on one end it has to come out on the other. It is a lot more complicated process, there are physical processes in play that require the DNA to be within a certain range and that range is not related to the number of cycles that the instrument measures (read lenght). If your sequences are too short the sequencing will not work properly in the first place.
To get back to your question when the sequences are short the mate is the reverse complement of the first read. At the same time both the read and its mate may also end up measuring the artificial sequences ligated to their ends. So only the overlapping region is complementary.