Are you sure you're talking about Steam here? You don't rate a game by stars, you either recommend it or don't recommend it.
Regardless, your suggestion won't actually do anything to prevent review-bombers. Your system is still prone to abuse and would invalidate a lot of legitimate reviews.
I'm talking about a rating system for user reviews, not the game. The game is rated by recommend or don't recommend, but user reviews have no parsing method to check for troll or not.
By introducing a tag system, Valve can auto-check for troll reviews/filter reviews to aggregate a more valid score. Troll reviews hurt game sales if the overall recommendation of the game goes from "Overwhelmingly Positive" to "Very Positive" to "Mostly Positive" to "Mixed" so by being able to filter reviews for specific tags, you can create a more validated score.
I'll give you an example of what I mean:
The precedent example now is Firewatch. The game got review bombed for the PewDiePie incident and almost all the negative reviews (aka
Do Not Recommend) referred to the dev as "whiny" "SJW" "DMCA" The theoretical tag system would auto-generate tags for the reviews based on usage (similar to how Steam already does for games genres) and it would allow Valve to remove the impact those particular reviews have on the overall recommendation status of the game. This would prevent trolls from nuking game reviews/hurting sales and you wouldn't be affecting the review pool at all as the blacklisted tags wouldn't affect reviews that don't carry the negative content.
Sure someone can review bomb something and not use specific keywords, but with the timeline functionality, you can cross-reference review updates with patch releases/content releases to negate any reviews that were altered outside patches/content pushes that utilize specific keywords/phrases.
Not saying it's easy, I'm saying it's doable and even though it may not be 100% accurate, it can help diffuse the nukers.