Rate Attributes eg Director(s), Cast, Length
Treat attributes like tags and allow for the ability to mark as "like" about attribute or dislike that attribute about movie. This would have a catastrophic effect on improving predicted ratings if used correctly. For instance, being able to specifically rate / tell movielens that a rom-com was enjoyable in terms of the romance / drama genres but the comedy was off would be great or vice-versa. It would also likely increase "important" tags [statistically significant / even if tags are not used, attribute ratings could be more easily used / most beneficial to community as well regardless of if ML uses data], clicking a thumbs up / down or drop-down menu [not great UI implementation with ample screen real estate] like with tags is far easier than the perceived / real effort required to type and add tags.
Most people when rating a hotel, or restaurant will at a greater rate happily fill out an additional attribute ratings such as location, value for money, customer service than actually WRITE a review on top of their general star rating. Those attribute ratings are also far more easily used from a statistical analysis standpoint than no-restrictions tags. Even if recommendations are not geared by them, still extremely useful for the community, instant visual / numeric snapshots rather than wading through piles of reviews to be able to come to a properly informed opinion that the restaurant is over priced. The same would very likely apply to attribute ratings rather than current tag situation which is rather anarchist without much guiding structure. Many rom-coms are really romantic chick flicks which do not cut the drama romance mustard while others are comedies trying to widen appeal to a variety of demographics but about as romantic as a bunch of guys in their boxers in a room farting.
Short story to the above long story, many movies are not IDEALLY marked with the correct genres, the issue is anybody who edits a movie is subjectively forcing their individual view on the community if they do not think a movie suffices as a comedy / other genre. By allowing attribute ratings, the community can decide if the movie should be identified as a comedy and the quality of such.
Another great example is when movies go past around the two hour mark that is where they can become tedious, obviously everyone has different temperaments / tolerances but this again could be used for actual research purposes, benefit the ML stakeholder while also benefitting the community. A fair idea could be drawn from even five to ten movies being attributed rated negatively for length by an individual user, taking the average might very well find that an individual specifically starts to get ancy once the movie hits 135 minutes, while others might be around the 150 minute mark. Again, if the data is not used or at least in the short term, it still benefits the community as even some 110 minute movies can be a real drag which informs the community the movie may be a bore while those that are worried about a 135 minute long movie would be able to see a neutral or maybe even slightly positive rating as a reassurance that the movie is not too long.