Stars rating systems have been working pretty well in practice from the user sideon giving a general idea of the quality of a product, but we can't expect more than that from them.
The rating is familiar with most of the users because of:
Familiarity on the real world: It's used worldwide in hotels as a quality scale.
Familiarity on the web: It has been used throughout massive-traffic websites with consistent results (shops sites, movies and series sites,etc).
There are some features to take in count to make them work as good as they can:
And when you display products, get sure of showing the number of votes done to make the rating more meaningful.
Something to avoid: This way of showing reputation (Ebay as an example). Endless and impossible to remember colored stars.
As it is not possible to make some "coherence validation" between the amount of stars selected by the user and their comments, there always be some users that will fail with or without porpoise, but at the end it would tend to be the minority.
I think that the thing about the users' tendency of voting 1 or 5 stars is something to take in count while analyzing this model, but the Like-Dislike approach is too extremist for rating products IMHO.
For me a good approach would be a system with 3 states, positive + negative + neutral, because talking about products you could think something is good but "not good enough", or "not as expected" but still useful, and that information will be really useful for the product seller. Maybe 5 stars are too much for the average user to "decide" between them, so they end up falling on the extremes.