This is not the place for a pedagogical explanation of field theory and perturbative calculations. That will come with the lectures.
First, one should understand that we calculate scattering probabilities based on a perturbative expansion in a small quantity. In fixed-order QCD calculations, we expand in the strong coupling at large momentum transfer, g_s. Since Feynman diagrams are squared, we see g_s**2/4pi=alpha_s. For Electroweak calculations, we expand in g_em, with g_em**2/4pi=alpha_em.
The leading or lowest order (or Born level) calculation is the one with the smallest power of the coupling that is possible. For q q~ -> W production, LO is 1 power of alpha_em and 0 powers of alpha_s.
Next-to-leading order (NLO) is one power of the coupling beyond LO. For q q~ -> W production, NLO in QCD is 1 power of alpha_em and 1 power of alpha_s. Since alpha_s >> alpha_em, the QCD correction is considered first.
With 1 power of alpha_s, there are 2 types of Feynman diagrams to consider.
Virtual diagrams lead to infinities when the virtual gluon comes close to going on the mass shell. The real diagrams become infinite when an emitted gluon becomes soft (low energy) or collinear (parallel to one of the quarks). Fortunately, the two sets of infinities can be cancelled in such a way to yield a finite result.
The main virtue of NLO over LO is that the answer is:
NLO Monte Carlo tools, as described here, are not event generators. Since they require a cancellation between real and virtual diagrams, it is not even possible to ask if there are exactly zero gluons in the final state. However, when suitably averaged, useful distributions can be made.
A technical point is that NLO calculations are made using weighted Monte Carlo methods. In practice, cross section weights come out positive and negative. If one chooses a suitable observable, and generates enough events, the positive and negative weights will cancel in a histogram, yielding a physical prediction.
The next order in perturbation theory is called NNLO, for next-to-next-to-leading order. For Electroweak processes, NLO is the first time that alpha_s appears in the problem, so one cannot be confident that the result is converging. For these cases, NNLO is desired to study the scale variation. Indeed, for those processes calculated at NNLO, the scale variation is minimal (with the exception of Higgs boson production).