I applaud their efforts to improve on movie ratings, but this is not an improvement. I agree with another commenter that they've really just created a 2D histogram, which is more difficult to assess than the histograms they complained about.
What's especially difficult with the scatter plot is that it requires you to assess density, rather than a simple scalar value. The other histograms have 5 numbers to indicate "weight" for each star, and a bar next to the star visually indicates the proportion the ratings received for that star. For their scatter plot, if there are 1000 ratings, how will it look different from a film with 100 ratings? The relative proportions of the ratings will only get muddied with the scatter plot approach.
The other thing about the scatter plot is that it still essentially maps to a 5 star rating, but only makes it more difficult to asses which star. That is, we are expected to visually assess: [1 star] - greatest density in lower left quadrant, [2 star] - density greatest in middle of graph, [3 star] - greatest density in lower right quadrant, [4 star] - greatest density in upper left quadrant, [5 star] - greatest density in upper right quadrant. There are only 5 useful density assessments which brings us back to the same categories as the 5 star system. Only in the scatter plot, its much much much more difficult to assess which quadrant (star) the ratings map to. And really, what is the meaningful difference between the 2, 3 and 4 stars (in my example)? Those density groupings seem almost equivalent (or some might argue). So in reality, the scatter plot will really only be meaningful if there is very little deviation between quality and re-watchability (which isn't true), which will help to group the ratings making density easier to assess. If they diverge frequently, then the plots are just going to be ignored by users since they'll have to assess density in every plot to try and make sense of it. That's hard. That's work. Users don't like to do work.
Finally, re-watchability? The question on a user's mind is, "would I want to watch this movie?" not, "if I saw this movie, would I want to watch it again?" I rarely watch movies again. Even the ones I love and own. That seems to be true of most people. The reasons for wanting to re-watch a movie are unrelated to whether it would be worth watching the first time. I'd argue that re-watching a film is more of a personality type than anything related to a movie.
What's especially difficult with the scatter plot is that it requires you to assess density, rather than a simple scalar value. The other histograms have 5 numbers to indicate "weight" for each star, and a bar next to the star visually indicates the proportion the ratings received for that star. For their scatter plot, if there are 1000 ratings, how will it look different from a film with 100 ratings? The relative proportions of the ratings will only get muddied with the scatter plot approach.
The other thing about the scatter plot is that it still essentially maps to a 5 star rating, but only makes it more difficult to asses which star. That is, we are expected to visually assess: [1 star] - greatest density in lower left quadrant, [2 star] - density greatest in middle of graph, [3 star] - greatest density in lower right quadrant, [4 star] - greatest density in upper left quadrant, [5 star] - greatest density in upper right quadrant. There are only 5 useful density assessments which brings us back to the same categories as the 5 star system. Only in the scatter plot, its much much much more difficult to assess which quadrant (star) the ratings map to. And really, what is the meaningful difference between the 2, 3 and 4 stars (in my example)? Those density groupings seem almost equivalent (or some might argue). So in reality, the scatter plot will really only be meaningful if there is very little deviation between quality and re-watchability (which isn't true), which will help to group the ratings making density easier to assess. If they diverge frequently, then the plots are just going to be ignored by users since they'll have to assess density in every plot to try and make sense of it. That's hard. That's work. Users don't like to do work.
Finally, re-watchability? The question on a user's mind is, "would I want to watch this movie?" not, "if I saw this movie, would I want to watch it again?" I rarely watch movies again. Even the ones I love and own. That seems to be true of most people. The reasons for wanting to re-watch a movie are unrelated to whether it would be worth watching the first time. I'd argue that re-watching a film is more of a personality type than anything related to a movie.