This is the source code for movievsmovie.datasco.pe
Watch the talk on vimeo
Slides on slideshare
This is a very hard question to answer. Research has shown that when we rate movies, the rating depends a lot on our mood, what we've seen/rated lately, etc.
Rating movies is sensitive to environmental factors. Our memories cannot evalute the entirety of all movies we have seen, so every time we rate a movie, even though it feels like an ultimate rating, we are really comparing it to a limited set.
As an example, if we rate a lot of movies in succession and they are all movies we are fond of, our grading becomes a lot more strict, without us noticing it. This is because the human brain is great at adaptation, pattern recognition, and setting a baseline.
This is generally very useful, but in this case it becomes harder to find out which movies you truly like best, or to have a complete, continuous ranking that doesn't depend on different conditions of each rating session.
A star rating (1-5 stars out of 5) is a good starting point to get a general idea of where the movie is: towards the top, the middle, or the bottom of your rankings.
However, it does not have enough resolution for a continuous ranking, and it is sensitive to factors like our limited memory. In a moment of rating, we cannot analyze a movie in the context of our entire library of movies. Instead we think of a much smaller subset.
Trying to use a high resolution grading system (like out of 1-100 or 1-10 with a decimal place) makes ratings even more sensitive to such problems. Let's say you rate two movies. You may give a 7.8 to film A today, and 7.6 to film B a few weeks later, even if you would have said that you like B better than A on either day.
For a movie, we start by asking you to give a star rating out of five. Then, we get a better idea of how much you really like it from one on one fights against other movies.
To rank movies, we need high resolution scores. Directly rating on such a scale is unreliable. Instead, we get there through one to one comparisons.
When you determine the winner (or draw) of a fight, scores change according to the result. The score for each movie starts as your star rating. Changes after a fight depend on the previous scores.
Let's say Casablanca had a score of 5.4, and The Matrix had 4. If The Matrix beats Casablanca, its score goes up quite a bit and Casablanca's score drops significantly. However, if The Godfather has a score of 4.7, and Troll 2 has a score of 1.4, and The Godfather wins, the scores don't really change much, because that was already expected.
Also, Movie vs Movie learns much more from the first fights of a movie than its hundredth, since we do not have much uncertainty left by then.