To be informative requires saying something that could be wrong (if you do not have to think whether you are making a good or correct assertion, such an assertion has practically no meaning). And to do this in an interesting way requires statements/assertions with sufficiently rich structure.
A rating must rate some thing. It must represent, or measure, an at least semi-defined objective thing. Rating should produce data, and if there is no definition – no abstraction – at all, there is no data.
The easy-to-measure things can be done automatically, with sensors or algorithms etc.. The harder things must be left to humans. But the difference in harder things is not having no structure – of being entirely irrational, aesthetic, etc. – it is being semi-structured. The difference is merely in what is currently technically automatable.
* * *
And it appears that Google not only understands this, but is somewhat ahead: they have been developing a sophisticated system based on complex analysis, as an extension to their +1 button. ‘Project Argand’ (as it is known internally) is almost ready to go live, as the Google +i button.