Why is it so difficult to choose which movie to watch (and what could be done)

This is the answer to the post on the burning topic of "what to see . " The author asked himself to raise the right question, but for some reason he reduced everything to a strange psychological analogy of “stress seizing”, saying: “Until you yourself know what you want to see, no one will help you. You will look stupidly for the sake of psychological comfort. ”



In my opinion, such a formulation of the question is excessively speculative. Yes, it is partly true, in the sense that watching a movie is a spontaneous thing and initially does not imply a serious mental analysis in the process of this entertainment.





Keanu also wondered which of his films to watch.



But (everything written below is the personal opinion of the author and his reasoning) as a rule, personally I can, albeit in my own terms, explain what exactly I want from the film. And since there are personal criteria and at least a qualitative assessment of them, in addition, there are formalized concepts like “script”, “directing”, “sound”, which can also be criteria, it is already possible to build a certain model - in the form of algorithms, a classifier, or learning with a teacher.



I am not a programmer and I can’t demonstrate a certain working method "just right". But I am a film enthusiast, and I will try to clearly articulate those ways of understanding and evaluating films that you can pay attention to, and the attendant problems that may not be obvious if they are not clearly identified. Since this is the answer, I set the tags-hubs as in the original post, don’t beat, Friday .



I apologize if the presentation turns out to be messy, it is written a bit in a rush and in the course of thoughts. In addition, the problems of cinema perception and the problems of its classification are so closely related that it is easier to analyze them than to propose solutions. But this does not mean that these difficulties are unsolvable in principle.

What is a movie?

Cinema is a type of visual art. Of the remaining forms of art, the tasks and methods of perceiving cinema are closest to painting. But on one important property - development over time - cinema is rather closer to music (I didn’t come up with it, but read it with the well-known Robert Mackey, who I will refer to below). That is, a movie is a kind of dynamic picture that has two dimensions - a visual component and a plot. Both components can enjoy and give the viewer pleasure. Accordingly, to separate these concepts in the article I will use the completely conventional terms “ quality ” (according to the picture) and “ interesting ” (according to the plot) cinema, which I coined.



The task of the cinema is its so-called “magic”. In English, magicians are called "magician" because they use a rather crude and deceitful manipulation to create the appearance of magic (and as you know, any sufficiently developed technology is indistinguishable from magic - Sir Arthur Clark). So the two-dimensional projection of the acting game should draw us into the screen so that we forget about the wallpaper along its edges and the ending beer in a glass. This is the magic - to follow the images of people, their actions, forgetting about conventions like changing scenes, flashbacks, voiceover and the fact that we know these actors from a dozen other films.

The problem "What to see?"

However, all the parameters of films, from length and country of origin to genre and director, are a matter of audience taste. And our task is to cater to a particular taste. The compulsive overeating example is good, but one important point has been missed: does someone get stuck in stress with unloved food? No, it does not lead to a feeling of comfort. So it is with the movie. Yes, the average viewer can relax after work, any series about bandits will go in the background. But the author of the original post positioned himself as a sophisticated movie fan, to which, however, I myself modestly rank myself. And it would seem that I should be well “calibrated” for some AI-Advisor, but in reality I just reviewed all the suitable films that interested me. And also my tastes are wide and depend on the momentary mood - and here it is, the movie screening crisis.



The following important thing should also be noted. Bad films still go on sale, otherwise they would not have failed in it. And millions of people still manage to go to the failed film. So any movie for the viewer is a black box with some probabilistic characteristics until the viewer watches it. It turns out that running into a bad film in principle is a statistical inevitability .



In this regard, I would solve the “what to see ?? 77” problem not in the form calculated by the neural network based on the preferences of the optimum-Film, but in the form of the Fermi method to reduce the probability of choosing a “bad” movie. Of a million films, you need to cut off everything that the viewer is definitely not interested in. As a result, we will have a pool of films at the output, from which it will be quite possible to choose something for the evening. Actually, I personally did this with the help of Kinopoisk, cutting off non-American films until 2000, comedy / television / musicals and other crap, as well as everything below 7 on IMDB. But alas, these films quickly ended. So, you need to speculate on more sophisticated and less formal search algorithms.

Sifting poor quality

It turns out that the first task of the algorithm, designed not to spoil your evening, is within the power of the simplest classifier. This is to weed out "poor-quality" cinema. A movie that breaks its own magic with poor shooting, ugly special effects, powerless scenery.





Yes, you encroached on the sacred, - ST fans will tell me. Fans already know what to see, ”I will answer. Old and classic cinema can “go out” on the second component - an interesting plot, which we will discuss further. Or maybe it’s not to go in because of obsolescence. Yes, each era in the cinema has its own purely technical and visual techniques, such as combined shooting, signs of technology that are one way or another striking the modern viewer. Therefore, let us bow to the talent of James Cameron, who invested his soul in blowing off the Los Angeles layout with a building hairdryer.



Yes, if you have a love for a famous film, a good actor or a literary basis, poor impressions of mediocre visual techniques (such as accelerated reproduction of car chases to make speeds seem higher, or frame-by-frame animation of AT-AT walkers), you are leveled out by general pleasure. But in modern cinema, the quality bar is set, and the only excuse for a bad picture can only be a budget of $ 100:





Well, seriously, the task of eliminating low-quality cinema is easily solved by the same IMDB rating. People will watch comedies with Adam Sandler, but not poorly shot films with wooden actors, and will vote accordingly. But this is the task of eliminating a false positive result (you will come across a low-quality movie).



And what about false-negative results (you are a fan of beautiful, stylish, juicy, atmospheric shots, and missed such a movie)? How to evaluate a film with a well-known or well-worn storyline, but with outstanding production and artistic value?



The visual and artistic component comes to the fore in historical, costume, and epic films. Here she is the first means and the main advantage of the film, because there can be several adaptations of one story, of varying degrees of saturation and sophistication. But the plot is unlikely to interest the masses and cause intrigue in the viewer. Well, who does not know how Waterloo ended, did Jeanne D'Arc and Marie Antoinette survived, what was in the Trojan horse? For connoisseurs of such a movie, the winner will be the one that is more beautiful and scaled off, but the ratings will not reflect this so much. Yes, of course, the peplums of the 1960s will be rated and spectacular today, even despite their venerable age. Here, however, the effect of the classics works: a must see for the viewer and an example for equalization among professional filmmakers create a positive feedback for themselves.



But there are completely different genres and subgenres of cinema with the main emphasis on the visual component, which are much more subtle and “tasteful”. Examples: noir "Batman", both Tim Burton and Christopher Nolan. Retrowave "Throne" and "Blade Runner 2049." The now forgotten diesel-punk "Sky Captain and the World of the Future" with sepia colors, like a military newsreel. Naive and pathos like an anime, and at the same time incredibly stylish “Forbidden Reception”, which infuriated me after the first viewing, but revised with interest and revealed to me the charming Abby Cornish and Jena Malone.





In addition to, and somewhere, instead of the plot, these films capture their style, create an original and unique atmosphere, a color scheme that creates a certain mood. There is no intrigue in Irreversibility with Kassel and Bellucci, because this film was shot in reverse time sequence and reveals its own finale in the first minutes! But the camera, constantly darting to dizziness, leads us from one episode to another in a pseudo-continuous plan. She flies like a fly from a disgusting BDSM club into a taxi with the main characters going there and back in time - a unique and exciting technique that carried me away with all my wariness to the arthouse.



Recognizing a visually outstanding movie that can bring a connoisseur pleasure is a serious problem for our AI, neural network, Advisor - whatever you create and whatever you call it. For now, we just designate it and move on to the next component of the cinema.

And what is an “interesting movie”?

Evaluating the plot is even more ungrateful than arguing about tastes regarding a cinematic "picture." When someone prefers the Transformers CGI, and someone enjoys the idea that the “Road of Fury” car tricks are authentic, then both will agree that they want a good action. But when it comes to history, character motivation, plot moves, scenario adaptation (and the book is always better ) - the same opinions practically no longer exist.



On the one hand, “a person is governed by love, hunger and fear of death” (c) Nietzsche, and all the plots of the movie can be divided by these extremes. On the other hand, there are about a thousand archetypes of folk tales alone.



How to classify scenarios? For battered genres? How to determine the quality of the story told? I love the famously twisted plot, an unexpected denouement, just like in the movie "Seven" with Pitt and Spacey. Here's how I put this into the algorithm?



I propose to separate the genre of cinema and the interest of the viewer in the story told. These are different qualities of the film, and they should be considered separately. To fully understand the concept of the plot - not necessarily a cinematic one - I highly recommend Robert Mackey’s book “A Million Dollar Story” (the original is simply “Story” - “Script”, it’s a little ironic that the book about the movie was localized as if the translator had bitten the translator) . Among other things, all cinema genres are listed separately, not invented for some reasons by critics and film critics, but sorted from existing films based on some general ideas, postulated values, plot moves, etc. Therefore, in addition to the genres “Love Story”, “Social drama ”,“ Crime film ”with subgenres, the book also contains“ The plot of punishment ”,“ The plot of growing up ”,“ The plot of trial ”,“ The plot of atonement ”. Such an applied classification, with some mechanics, is just more specific and useful for sorting data - that is, movie titles.



You can also talk about the quality of the script by drawing on the basics of Mackey.





Using the theory of writing a script, one can evaluate the integrity, canonicity, and consistency of the history of the film. But it is rather a job for the critic. To create algorithms, it is important for us that such an applied critic work not according to our own emotions and tastes, but according to similar formal criteria, and issue a verdict on the quality of the film received. I do not claim that someone will seriously engage in such a strange occupation, going on stream, preparing film after film for the sake of training a recommendation service. I am only claiming that this is not impossible.

General strategy for creating a recommendation mechanism

Now neural networks are fashionable. But their use is not always justified. Somewhere I want to think for myself, somewhere this tool is a cannon against sparrows. In the case of our task, it is reasonable to optimize the approach - first use the simplest tools. Before carving the sculpture, the stone must be broken out with a bulldozer.



There are countless films in the world, and even thousands of people will not have so many human hours in their lives to watch them all. That which is not worth our viewing needs to be ruthlessly thrown out, and it is convenient to do this with some binary algorithm with a multi-parameter input. In the diagram below, the numbers from 1 to 10 are the rating of films I made up by year and country of release, but in principle, you can set any parameters along the axes that are guaranteed to cut off the excess for you. Duration / language / genre. Anything to narrow your search. In this conditional example, it is clear that the films of the United States since 2000 are “worth watching”. But an unassuming viewer can expand the range of watchability even to Russian cinema of the 1990s. It is important to ensure the flexibility of this choice. That is, despite the possibility of using the simplest binary classifier, a neural network at this stage is not needed at all. You can give this stage to the setting by the viewer.







We go further. After watching several calibration films, or after passing a test / survey, you can get your genre profile, I chose a petal diagram to illustrate the principle:







At this stage, you can already apply neural networks. Do you have any preferences by genre - ready-made weight coefficients for training or cluster analysis. At the same time, all quasi-relevant data was cut off at the previous stage - the user or the algorithm excluded a huge layer of Indian cinema, and it will not mix with modern musicals or melodramas. It is important that the genres are correctly characterized, standardized, and the viewer speaks the same language of terms with the film critic, the creator of the service.



The next stage of sorting and offers is the selection of the most high-quality and interesting films from the selected clusters. For example, I presented conditional data on films grouped by directors.







The graphs are T-shaped, because, as already discussed, the plot can be interesting or ordinary, but the visual quality of the film can affect the perception of the plot both positively, enhancing the impression, and negatively. Relatively speaking, in the drama, the emphasis is on the psychology of the heroes and the conflict, so the picture is not so important and for most films of the genre it will go approximately along the abscissa axis.



It is clear that we need to focus on films from the top-right, beautiful and interesting. But what if the best films already seen? Or do you want to plunge into a beautiful arthouse anti-plot? My opinion - it is necessary in each case to give the viewer something like a priority slider - aesthetic pleasure against a plot that is in suspense. A counseling service cannot be a deaf seer, but must have feedback.



I summarize: the most productive approach to the recommendation algorithm in the choice of cinema, as it seems to me, should be the sequence: 1) we cut off all unnecessary; 2) select a group of optimums by the neural network based on the preferences of the viewer; 3) choose from the proposed in accordance with the priorities of the viewer in the direction of the plot or the colorfulness of the movie.



In this case, the first stage should also be controllable. The viewer should not be aggravated in his desire to watch more and more thriller thrillers thanks to the recommendations. He should be able to optionally exclude a heavy movie at the time of depression, or discover anime and GDR westerns.

Unsolved problems

I affirm that there will always be an opportunity to stumble upon a movie that you don’t like. It can be minimized as much as you like, but good films that fall under false negative selection may begin to suffer from this.



Another important point is that the film can be treated differently - it depends largely on the viewer's expectations of the film. In my practice of pleasant viewing, there were films that I did not accept the first time because of unjustified expectations. But remembering them - he came back and reviewed, because I suddenly wanted just such a movie. For example, sitting down to watch the movie "Survivor", I was waiting for a cruel thriller like "I spit on your graves", but realized in time that I was watching thrash and successfully "reconfigured". From that moment on, I no longer had any complaints that the heroine had burned the wound with a beer can and that an eagle with a label was printed on her belly, which was not even mirrored, as well as to a man wrapped in stretch film so that his intestines would not fall out. “Gruz-200”, for my taste, is disgusting as “author's”, “festival” and other such films, but it’s also good to watch it like trash. Another thing is that no algorithm can predict such a human reaction.



Enhanced specificity of the displayed results would help extended tags, sort of like on Yandex Market. “Viewers love this film for: (unexpected ending) (soundtrack) (love line) (supporting role). But who will write them? Spectators - there will be anarchy and srachi.



Moderators or film experts? For little-known films to no avail. Using the names of actors and directors as tags is a tempting idea, and I myself tried to watch films by the directors I liked and my favorite actresses. That's just Christian Bale so many times did not lose and gained weight, how many jumped up and down the rating of films with the participation of Nicolas Cage. Numi Rapas is genius in the role of Lisbeth Salander, but critics rightly said that she had absolutely nothing to play in the second Sherlock Holmes. A stylish film about a driver with a toothpick in his mouth has nothing to do with the early films of the same director about Danish gopniks, one of which was played by the future bond villain Mads Mikkelsen. Another example is our compatriot Yuri Bykov. He shot three seemingly absolutely identical films: “Major”, “Fool”, “Plant” (a person wakes up with a heightened sense of justice, but he is opposed by the servants of the system and the shadow oligarch), but all three are occupied by the same actors with too characteristic shchshchi, and it hurt my eyes when watching, as if you are watching some kind of self-parody.



Yes, there is another aspect of the perception of cinema, which is not subject to either your mind or artificial. Your tastes, your views and your perception of the movie changes over time. And you can’t do anything about it - you can only watch the movie again. And I mean far from just the fact that the film began to seem banal, boring and familiar by heart. Ceased to “cling”. No, it can even begin to be perceived from a completely different angle. If you understand what I mean - you can skip my personal example under the cut.





In general, I consider the recommendation system a reasonable and achievable thing. Its main properties should be minimal customizability, which does not allow to go into positive feedback, and working out the basic parameters of the quality of the movie (how to implement this is a separate issue). Just don’t expect too much from such a system. In the end, you want to watch a movie tonight, not her.

Yes, by the way, you can freely exchange for comments on this article, and you are also absolutely not obliged to press the donate button down there. Since I’m writing a response anti-article, then anti-plot in everything.