There may be reviewers who do not generally fit the bimodal distribution, but if they are the minority, why design your whole rating system around them? Better to optimize your system for the way most of your users behave.
The reason online ratings tend to be bimodal is because the respondents are self selecting sample rather than randomized. Since product ratings are not typically compulsory, the extreme likers/dis-likers tend to fill the survey question.
When implemented as compulsory, it drives poor user experience and having to fill a survey to get something anyway introduces a bias (just fill something to get it over with). So no easy answer.
This is also the reason traditional survey firms haven't really gone out of business even with the wealth of online data from Facebook, Twitter and product reviews available, though better/easier access to actual behavioral signals are certainly replacing surveys in some areas.
Just curious if I am an outlier with how I do my ratings.
1 - Terrible. Good Lord, who thought making this was a good idea?
2 - Poor. Would not watch again
3 - Fair. Neither for or against. Unlikely to watch again, but I don't grudge the time spent.
4 - Good. I enjoyed it. I would probably watch again.
5 - Exceptional. Probably in my top 10 now, and foresee enjoying watching for years to come.
3 and 4 are used a lot. 2 is used occasionally. 1 and 5 are rare.
1 - Did not like it
2 - It was ok
3 - Liked it
4 - Really liked it
5 - It was amazing
The reason being is that most books won't end up being 1 or 2, and there really isn't much difference between them. But it sucks giving 2 stars to something that wasn't bad. Perhaps emoji arranged in a non-star-like pattern would be the best.
Ultimately the key reasons to have these ratings in the first place is not to let people express their feelings, it is to get data for the recommendation modeling to predict better. Perhaps showing a simple like/don't like increases the number of people rating something which gives you more training data and thus better business outcome from your model without compromising on midel quality since you had to anyway model it bi-modally before.