The number of stars has been not very reliable on its own since a long time.

For example, a repository from Google or a random Chinese university always had much more stars than other similar repositories.

I prefer to look at the number of stars with other metrics, such as the number of issues. 3000 stars and 50 issues? Something is very fishy. 12000 stars and 6000 issues? Perhaps the software is not very good. You have exceptions of course.

I also like to look at the contributors graphs. After some time, you recognize the patterns, and you can quickly notice when the number of stars doesn't match with the graphs you see.