I know this question is a bit old, but I read a paper recently which (claimed) to answer this question.
I think the answer to this question is suprisingly controversial. I don't really have a formed opinion yet of which answer is correct, but I will try to discuss the different viewpoints here.
Argument 1: Sweeps are rare
On one hand, we have the people who are sceptical of how common sweeps are. This, in my understanding is most researchers. Many people believe that polygenic is the predominant form of selection in humans.
There is the paper by Hernandez et al (2011), who find scant evidence of selective sweeps in recent human history. Essentially, they looked at the reduction in diversity (something which is associated with sweeps) around amino acid substitutions and synonymous substitutions and found that the two were statistically the same. They say it's probably more likely that this is caused by strong purifying selection at linked sites rather than sweeps.
Szpak et al (2019) suggested that most of the time, the criteria for designating something a sweep in human populations was probably too lax, and many of the previouly claimed sweeps were probably artefacts of some kind. In fact, they only find compelling evidence for 2 sweeps in recent human history.
Argument 2: Sweeps are more common
Alternatively, there are some studies which have apparently discovered a relative abundance of sweeps in human history. Some might make the argument that sweeps are likely to be common, since humans have a relatively small effective population size (~20,000) and low mutation rate.
Souilmi et al (2020) claimed to have discovered 57 (fpr<10%) sweeps using genomes from ancient Eurasian individuals. 35 of these 57 were inferred to have occured before 35kya. The reason they claim to have found many more sweeps than previous studies is because they used ancient DNA. It is well established (Hellenthal et al 2014) that admixture is virtually ubiquitous within modern and ancient human populations. Admixture likely obscures the signal of sweeps by introducing non-selected haplotypes from the admixing population, making hard sweeps look like soft sweeps, or perhaps not even sweeps at all. They confirm this using simulations and show the power to detect sweeps is massively reduced when you are looking at admixed populations.
However, their method uses ancient DNA, which is highly prone to various artefacts, which they didn't appropriately test with simulations. It also hasn't been peer reviewed, so there may well be other potential issues.
Hernandez et al 2011
Szpak et al 2019
Hellenthal et al 2015
Souilmi et al 2020