Not being a movie buff, I really did not understand the basics of what the goal is here. (There was a problem statement, but it really didn't give the 50,000 foot view.) After having read much of the article, I think I can summarize it.
(1) Whose name comes up first when the credits roll matters in some way. Presumably people feel slighted if they don't have the right amount of prominence or something like that.
(2) Apparently, despite this being important enough to worry about, and even though there are standardized titles, and even though people have been making movies for well over 100 years, there isn't a consensus or standardized order. Seemingly every film just sort of does something they feel is appropriate.
(3) It must be a fair assumption that people who arrange credits do it with purpose, so that if you look at the order they chose, it tells you something meaningful about what the right order is.
(4) The goal, then, is to basically computationally reverse engineer what order people have in mind when they put credits on film and produce an ordering that reflects actual practice as accurately as possible.
(5) This is a messy process because the data is inconsistent and contradictory, so it is fertile ground for creatively applying algorithms to tease out the meaningful parts.
"Cheers" had an interesting solution to a conflict over who should come first, or perhaps more accurately who should not be second. Ted Danson and Shelley Long both wanted top billing.
The solution was to have Shelley Long on the upper right of the first screen with names and to have Ted Danson on the lower left of that screen. The credits were the consecutive screen type, not the long scroll type, so the two names appeared at the same time.
Long would be read first by someone who read down, and Danson would be read first by someone going left to right. The producers were able to convince both Long and Danson that they were not second.
The rest of the main cast each got a screen afterwards with just their name on it.
When Long left and Kirstie Alley joined the cast as the new female lead and there was now no question than Danson was the star of the show there were no more special screens. Each main cast member got their own screen, with Danson's being the first shown.
Is there empirical data on which one is more likely to be read first? I feel like more people would read the top one first than the bottom one, but I don't know.
(2) commonly, once the leads or "above the fold" credits, they list by department with leads listed first, then plebes last. Look at all of the 3D artists. Just a list of names. Above them, are the project managers, above them are directors, producers, etc.
*all of the above comes from personal involvement. sometimes asking a direct question of "what goes first" is responded with "I'll get back to you", but never does. "Just do what everyone else does" is common response too.
you have to remember that the person doing the credits is pretty much the lowest of jobs. nobody wants to do this job, and it is often assigned to intern level assignments. "take this Word Doc/Excel/.txt, and turn it into a graphic". Every post house I've worked in has always cried "there must be a better way".
Post author. You nailed it on 3 - 5. This is about cutting through the noise to discover the emergent conventions, then using them to help first-time filmmakers. For something like 80% of feature film producers it's their first rodeo.
1. Above-the-Line credits (cast, director, producer, other key roles) are contractually negotiated, and a mistake here has definite consequences. That isn't usually the case for all the remaining credits. But people still care that the ordering makes sense and flows well. Chronology, prestige, authorship, and org chart level are all axes you could order on -- but that's already too many ways to reconcile.
2. One of the challenges with standardizing credits order is that roles come and go all the time. I mentioned this in the post, but 18 months ago there was no "COVID Compliance Officer" in end credits. Now it's in nearly all feature films, and we know where it goes. HN might also find it interesting that roles like "Data Center Engineering" and "Deployment Services" have started showing up just in the last few years.
70 departments, 10-20 roles each. Sounds like a common sense/spreadsheet problem to me. You don’t need big data/machine learning/insert the next big thing here for solving this.
Is there any evidence that they aren't imaginary? For example, does be listed early in one project, lead to improved likelihood of getting a better placement on the next project (with corresponding higher future pay)?
If that is the case, then the perception that one is a "star" or whatever, with credit order, mind correlate with future success. Of course, that correlation might be independent of credit-order placement (such as the actor is actually better in the movie and would be perceived as such even without credit order placement).
I think the primary goal here is to create an exercise for exploring different graph theory concepts, not to actually be useful. You're thinking too much :)
Really interesting write up. I hate that this website hides my scroll bar, though. And overlaying the ordering right on top of the image of the credits as you scroll is irritating. If you want to compare the ordering and the credits themselves, you have to scroll up and down repeatedly. There is plenty of room to put the ordering next to the credits, no "fancy" styling required.
It's a testament to how long it's been since I've been in a movie theater, but I was thinking "What about scrubbing to the end?" Oops, can't do that on the silver screen!
Doesn't this presuppose that there exists a single canonical way to order the credits, rather than them being reordered per-project? Perhaps the Lighting Electrician role in Love Hard was less impactful than in Black Is King, leading to it being intentionally ranked lower.
That's my thinking too. It also ignore any and all social reasons for changing the order.
Shaking out a fairly optimal list that jibes with the instincts of people who have read a lot of credits has its uses, though; it feels like a good starting point from which to deal with the fact that this production had some incredible demands put on the Balloon Tech, and that the Basecamp Electrician's father is one of the studio executives and cut a deal to have their kid's name higher up in exchange for greenlighting the show. Or whatever.
No, it acknowledges that common practice varies such that there is no canonical ordering, and tries to get at an ordering that, when applied, minimizes difference with common practice.
yes - but I think the idea was to reduce the perceived level of arbitrariness of their ordering by appealing to mathematical analysis. Technical analysis is used for this purpose all the time.
"The tool you bought said X, I just went with that to get something to you quickly" is a great way to Get Things Done in any political industry. If stakeholders care, they'll override it placing less blame on your shoulders than they would otherwise. And tool makers that optimize for minimizing surprise can be great allies in this.
Definitely. For most on-screen talent, part of the ordering is even contractual. I think this was more of a fun exercise to see if they could come up with a model for how that political process averages out to create a technical one.
Why? What's the point of trying to come up with a single optimal order? Every movie does it differently and probably has reasons for doing so. This seems like investing a huge amount of work to solve a problem that no one really needs the solution to.
> What's the point of trying to come up with a single optimal order?
On the frontpage of Endcrawl, the company behind this blog post:
> Ace your end credits and de-stress post production with the solution used by over 2,000 films and series.
Having a standard template seems like a way to easily "de-stress post production", and maybe even "ace your end credits". If this actually becomes an industry standard, the company behind it would also get some publicity. I've personally never even thought about how credits were made, and never knew that there were companies focused on just this point. Now I know that Endcrawl exist, and have a positive opinion of them because they wrote an intersting blog post, that could be used as a reference for other ordering problems. They're also trading this standard template for your email address: https://endcrawl.com/template/
What's more interesting that Endcrawl actually started with the process of sending over an Excel file over email. Seems like they already got their main use (no the editor/graphic/motion don't want to spend an hour trying to change 1 name) and keep finding ways to improve credit-making.
Probably not. They do credits for movies of all sizes. If you're doing a small student film, do you really want to think about the order of your credits? This gives you a great default choice.
As students, they probably have limited insights into what a role is worth. With this tool they can base it on data generated by the industry and get a fair starting point.
Does anyone know of a resource where you can search for peoples' names that appear in any movie/tv credits? The list of credits can be massive, and I don't think sites like IMDb are trying to create archival coverage at that level of detail.
Once you get past the major names and into the scroll, the text usually moves from the bottom of the screen upwards. This works naturally with the way we read English. But I've seen at least one movie do it in reverse with the text moving downwards. I thought it was the original 'Repo Man' but the only instance of those end credits I can find on YouTube seems to indicate that's not it.
Whatever movie did it, it makes me wonder if the screen could simply be split in two, with major credits scrolling upwards on the left half and more minor roles scrolling downwards on the right half, perhaps at a different speed. Major stars, directors, producers, etc. would still get their full screen credits before the scrolling started. Deciding who ends up on with side of the screen would of course be another complication.
I just checked my copy of the Criterion Collection version of Repo Man and the credits do go in reverse, moving downward. (I also saw it not too long ago in a theater and remembered it going in reverse, for what that's worth.)
The main acting credits order in films is an interesting question as well which I guess could maybe be answered in a similar manner. Some films order actors alphabetically. Some by appearance. And some order the first N by some contractually reasoning, but then presumably the rest will be ordered less strictly, possibly somewhat randomly. I've always wanted to tackle the problem of trying to standardize the lists somewhat.
Just an example to explain what I mean. The film Bard Wire has an alphabetical cast list (so Pamela Anderson Lee, as she went by at the time, is mid way down the credits). Interestingly the "three actors" listed by IMDb has Anderson first, but then the next two are just the top two on the credits. It would be useful, to a degree, to try and figure out who those second and third people should be based on some metric.
I think of Black Adder here, where the credits ordering is different in every episode. In one episode it's ordering by disappearance, in another it's by precedence in the royal lineage, in another it's by Geographical Order, and so on.
Cool article, it's fun to see how many situations can be mapped to graph theory.
The description of the cost function seems strange, since it's described in terms of the distance from "the correct order". It's clear that there is no single correct order. If two movies disagree on the ordering, it is not necessarily the case that one of them is doing it "wrong".
Mathematically, it seems like it would be better to see any given movie's ordering as a sample from a statistical distribution. That suggests that computing cost in confidence terms, as in the probability of generating that ordering given your assumed distribution, might make more semantic sense. So for example you could maybe use the frequency graphs from the article and sum up the surprise of each path from the first to last entry in your list. (Where "surprise" here is the inverse of how frequently one node follows another.) That's linear. Or you could do it quadratically by making a matrix of A-follows-B frequencies and then summing up all pairs of entries in your list (normalizing by the length of the list). The latter takes more of the graph structure into account.
Which is also the other thing that seemed a bit odd -- it seems like the "A follows B" relationship is getting a little mixed up with "A immediately follows B". As in, clumping the generator-related roles together isn't the same thing as saying an intern should follow a principal, and the cost function shouldn't treat those constraints the same way. I don't know how much noise it introduces, but intuitively it seems like the algorithm probably ought to do an ordering and then a clumping. Or perhaps the opposite: do ordering within clumps ("everything with 'generator' in the name"), then treat the clump as a single component for the main ordering pass.
The last thing is that the article seems to take NP-hardness too seriously. Sure, if you really had to consider every possible permutation, it would take too long. But there's way more than enough structure in the problem to take advantage of. Some very very conservative heuristics would surely dramatically reduce the size of the relevant N that participates in the core NP-hard problem. Your Traveling Salesman may have to visit 50 cities in each of Oregon and New York, but you know there's no point in making him fly back and forth between the states more than the minimally required (2). Write your algorithms in such a way that you don't need to even allow the possibility of putting the Gaffer behind the Intern Electrician's Boyfriend's Dog.
If two movies disagree on the ordering, it is not necessarily the case that one of them is doing it "wrong".
(Author here.) Indeed! This is about trying to discover emergent conventions, so we can give first-time filmmakers a good starting point.
Or you could do it quadratically by making a matrix of A-follows-B frequencies and then summing up all pairs of entries in your list (normalizing by the length of the list). The latter takes more of the graph structure into account.
This is what PageRank (Experiment 3) does!
The last thing is that the article seems to take NP-hardness too seriously. Sure, if you really had to consider every possible permutation, it would take too long. But there's way more than enough structure in the problem to take advantage of.
I ask this question in a footnote [0] -- is this permutation space amenable to gradient descent? Don't know the answer! If someone knows this area well I'm all ears.
You missed one very interesting angle for the problem (speaking of games), namely Voting theory, which is an important part of game theory!
In voting theory, there is a concept called Kemeny ranking (Kemeny–Young method), which I believe is exactly what you are looking for. It is of course an NP-hard problem, but that shouldn't scare you away.
In a voting setting, each movie would "vote" for a ranking of, say, the electric unit; i.e., the gaffer comes before the other people. When you have many movies, you have many votes that you want to combine in order to rank all the candidates, while minimizing inconsistencies. A seminal paper was published in the journal of the ACM, Aggregating inconsistent information: Ranking and clustering by Ailon, Charikar, and Newman.
An important insight is that your directed graph is actually what we call a tournament; for every two vertices a and b, there is an edge either from a to b, or from b to a.
In that case, you want to solve a well known and widely studied problem, namely Feedback Arc Set in Tournaments (FAST). Check out the paper in ISAAC by Karpinski and Schudy, Faster algorithms for feedback arc set tournament, Kemeny rank aggregation and betweenness tournament (it's on arxiv).
Jumping off the end of the article... Has this technique been done with org charts? Is there any place I can dump a list of role names and have an "expected" org chart generated?
There are two preconditions for performing the analysis the same way. You need a standardized set of roles between companies. And you need a corpus of existing org-charts. Between those two, you could compute the "most common" org hierarchy. It actually would be interesting just to understand how how common titles align across firms, even independent of hierarchical structure.
This is a tangent, and I know everything is negotiated via contracts but what is the deal with the ordering of names on posters not matching the order of the actors on the poster when the poster picture is a group photo.
Because of an antiquated rule where the highest paid or biggest actor's name is always left most. So you naturally end up in the situation where the biggest actor name is on the left but the picture is in the middle.
Movie posters are a complex subject. Way more complex than they have any right to be.
Look at the poster for "The Towering Inferno". Steve McQueen's name is left-most while Paul Newman's name is the highest. William Holden is lower than both of them and to the right, lining up with McQueen's last name. And Faye Dunaway is about half an line lower than Holden's last name.
Then there's fights about how big someone's name/picture can be in relation to others.
Look at the live action Beauty and the Beast Poster. Emma Watson gets a lot of poster estate and first billing. Then it gets a bit crazy. Not to mention, how people are billed is a bit contentious as well. You'll notice that it's mostly a list of names except the final two. "with Ian McKellen and Emma Thompson". That's because being a special mention is worth something. It's a way to give prominence to non-leading actors who are significant in other ways.
The Thor movies do this with Anthony Hopkins. Clearly he's not the lead, but dude has had a career and he was a big get for the movies. For Ragnarok, they even toss it to Mark Ruffalo, probably as a way to acknowledge his part's significance in the film.
Spider-man: Homecoming is another interesting one to look at as well as it gives a lot of real estate to Robert Downey Jr. Tom Holland has a smaller picture than the Spider-man suit, while Robert Downey Jr. is much larger than the Iron Man suit. Michael Keaton also gets the double up, but both of his images are much smaller.
A lot of the Marvel posters are good to look at to see the politicking that goes on as they do a lot of ensemble movies.
And then there's also the fact that sometimes the pictures are made before the credits are put on it. So you're locked in an image but contracts dictate name order.
The names are listed with the biggest star first. Billing order is part of the contract. But the posters typically have the biggest star in the middle, most visually prominent, and then the smaller stars on both sides or sourrounding.
That is the premise of the startup that funded this work. Use their tools to build the credits at both higher quality and cheaper than the current approach. Having a standard order (that is modifiable when needed) both establishes creditability and seems necessary to auto-generate credits when the list of creditable people provided is unordered.
(1) Whose name comes up first when the credits roll matters in some way. Presumably people feel slighted if they don't have the right amount of prominence or something like that.
(2) Apparently, despite this being important enough to worry about, and even though there are standardized titles, and even though people have been making movies for well over 100 years, there isn't a consensus or standardized order. Seemingly every film just sort of does something they feel is appropriate.
(3) It must be a fair assumption that people who arrange credits do it with purpose, so that if you look at the order they chose, it tells you something meaningful about what the right order is.
(4) The goal, then, is to basically computationally reverse engineer what order people have in mind when they put credits on film and produce an ordering that reflects actual practice as accurately as possible.
(5) This is a messy process because the data is inconsistent and contradictory, so it is fertile ground for creatively applying algorithms to tease out the meaningful parts.