Rhizome

CategoryMetadata

Storytelling -> Data -> Storytelling

In Metadata
6 Min Read
S

HOW TO DESIGN A DIFFERENT MOVIE METADATA MODEL?

Truth, as the saying goes, is in the eye of the beholder. This is only one side of the coin, because truth is also in the hands of the storyteller. Both sides of the truth are exemplified by the movie Gone Girl (David Fincher, 2014), which revolves around Nick and his wife Amy who live in Missouri. On the day of their fifth wedding anniversary Amy goes missing. The days count up while the police try to figure out what may have happened. Something is clearly off and even though there is no corpse, evidence of possible foul play starts to point to Nick. That’s when the story pivots to a different perspective. As it turns out the whole missing person act has been a carefully crafted and executed master plan by Amy to frame Nick.

Gone Girl is about subjectivity and it effectively shows that we, the audience/beholders, just like the characters in the movie (even the police), construct a unique and personal truth from the available information. It also shows how the perceived truth is manipulated by Amy at the story-level and by the filmmakers at the meta-level. As such the movie is clearly also about storytelling.

It isn’t the subjective truth of the storyteller (intent) nor the subjective truth of the beholder (interpretation) that make Gone Girl interesting to me. It’s about what lies in between. A culture of storytelling where similar stories have similar recognisable characteristics. This is explicitly referenced in the movie during a scene in the second act that explains how Amy’s storytelling domain knowledge is learned by analysing books and tv shows about crime and murder. Ultimately she comes to understand the building blocks that are needed to build her story by first deconstructing comparable stories.

This is similar to the journey that I’m on. I’m trying to understand what the essence of movies is and how this can be consistently captured as data. I believe that Gone Girl shows that the gateway to capturing the essence of movies is understanding storytelling. How can storytelling be envisaged as data?

Let’s start with a definition.

Storytelling happens at the intersection of three components:

  1. the message/story conveyed (what)
  2. the intended audience to convey it to (who)
  3. the medium used to convey the message (how)

Amy understands this storytelling trinity. She uses this expertise to convey her message that Nick is guilty of murdering his pregnant wife, and adapts her execution (medium) to the best fit for each separate audience. For the befriended “local idiot” neighbour she focuses on personal conversations about pregnancy and Nick’s violent temper. For the police she stages a factual evidence storyline, with a half burned diary, credit card debts and poorly cleaned blood traces in the kitchen.

Her storytelling is effective, because it has a well-executed design. Effective in this context means that the audience picks up on the intended message. Amy understands how the beholder eyes the truth and uses this to tell her story. A storyteller needs to understand the message, the audience and the medium to be able to use that knowledge in designing the storytelling experience. This obviously applies to movies, books, tv shows and speeches, and maybe less relatable, it also applies to data.

Data storytelling uses a ‘data medium’ to convey the message to the audience. For instance a dashboard or a transaction receipt. A company executive can create a dashboard to convey a message of success to the CEO and a retailer can use the transaction receipt to convey the message of a particular purchase to a customer. As said these forms of data storytelling can only be effective if they have been designed.

This design process is called data modelling. It starts with defining the conceptual model that intends to accurately capture the business requirements and it ends with the physical model (the final technical implementation). Designing the conceptual model means translating the end goal into requirements, structure and definitions. If this translation fails the final physical model will fail as well.

For instance, designing the data model behind a transaction receipt means working back from its business requirement and creating a data model that captures the data needed to print an effective transaction receipt. The purpose of a transaction receipt is to convey the message of a specific transaction, at a specific time and for a specific price. If the timestamp isn’t taken into account it will be impossible to convey the intended message. This applies to any type of storytelling, because if the intended message isn’t understood it will be hard (or even impossible) to convey it.

That is why I believe that defining a method for capturing the essence of storytelling of movies works best if it’s also treated as a data modelling process. In other words, determining the conceptual model behind movie storytelling by capturing the elements and structure needed to tell an effective story. By modelling this way, the actual tagging of a movie can be consistent, reproducible and will have the same data structure as the tagging of any other movie, making it easier to compare movies with each other. Similar to how every logged transaction in the above example will have the same structure.

The intent of this data modelling endeavour is to create a movie metadata standard that differs from conventional movie metadata such as genres. The above elaboration on the story of Gone Girl hopefully shows the value of such a standard. Simply tagging this movie as either Thriller or Drama (or both) fails to capture its complexity and offers little to compare it to other movies.

In more or less the same way as Amy did, I need to capture the movie storytelling elements that are essential for telling a (specific) story. I believe that the descriptive analysis of this (specific) message/story should be the core of the ‘Movie Meta Model’. It’s what captivates us (the audience) most about movie storytelling. It’s the component we’re most likely to discuss with friends and family.

It’s not my intent to further elaborate on the specifics of this in this article. I believe that the best way to show how it could work is with a detailed example in a follow-up article.

WHAT IS THE DATA VALUE OF A MOVIE GENRE?

In Metadata
6 Min Read
W

Let’s democratise content discovery, starting with a different metadata model

The so-called streaming wars came and went, and did not leave us (the users) with better tools for content discovery. Over the last few years, the movie and series content offering grew enormously and got very fragmented at the same time. New streaming services tried to hone in on a piece of the market by differentiating themselves with unique and exclusive content. Even though they strive to be different through this content offering it’s clear that most streaming services aren’t unique in the way they offer their content to the user. Options for navigating content libraries are very homogenous. Every service offers a combination of landing page banners and rows, navigational categories, search and AI powered personalisation. Next to all the (talked about) advances in AI there seems to be no innovation on other methods for content discovery.

For me, the personalisation algorithm never provided any value. It did not help me find new and unexpected content. I’m sure many people will recognize this in some way, since research has shown that users are getting more frustrated trying to find something to watch. The business model of most streaming services focuses on unique content to acquire users and after that personalise the sh*t out of them to prevent them from leaving. I doubt that increased search frustration is beneficial in preventing churn. So maybe we should look beyond personalisation for an answer. Providing users with more freedom to navigate content as an additional feature next to personalisation, can diminish the search frustration by increasing findability of content. It can provide value for the user and as a consequence also for the streaming service by preventing churn.

Increasing freedom means providing informed access to content. On the one hand by adding more options for filtering and searching the content library and on the other by explaining the logic behind the options (informed as opposed to a black box).

To be able to achieve this, it is crucial to establish a qualitative metadata model. Metadata expert John Horodyski addresses this necessity in his book Metadata Matters. It ‘matters because it is both identification and discovery; it’s about access.’ Creating more freedom for the user to find the right content on a streaming service is all about navigational access and at the moment that qualitative access is lacking. Horodyski argues that a shift in thinking is needed. ‘Content is no longer “queen” … there are many in the realm, not the least of which is the user and the user experience. If you have great content and your users cannot find it, the value of the content is diminished or lost altogether.’

Of course, streaming services provide some metadata for navigating, but I want to argue that the value of that metadata is so low that it does the opposite of creating freedom. It doesn’t help in finding content. It diminishes the value of the content and traps us within a limited view of how movies can be categorised. I will explain my idea to remedy this by narrowing in on a specific category that is ubiquitous in every day movie discussions and included in the navigational menus of all streaming services in one way or another: genres.

Genre is a French word with origins in Latin meaning ‘kind’ or ‘sort’. In other words, a movie genre represents a specific kind of movie. It suggests that there’s a generally agreed-upon definition of the shared elements that shape a genre. There isn’t, and as such the current application of genres is inconsistent at best. Giving access to inconsistent data will not benefit identification or discovery, since one of the worst things that can happen to your data model is bad quality data. Anything built on that data or any conclusions drawn from that data will suffer from lack of quality because ‘garbage in is garbage out’.

This data value problem can only be solved by approaching movie classification from a different perspective. We should stop trying to see genre as something universal that can be used to tag movies. Instead we should look at the ‘hidden’ data that describes movies. A bottom up strategy that doesn’t work towards a specific output variable. This way, it’s possible to deconstruct movies to create a structured data model with clearly defined components and relations between these components, an ontology of movie metadata.

It isn’t possible to define this in one go. It will be a continuous process of development, whereby defining the model and using the model to create data will run parallel to each other. If the first logic is consistently applied to a sufficient number of movies this data can facilitate new insights to further refine the model and so on.

The above doesn’t mean that genres no longer have value. If used consistently they can help in positioning movies. As such they still have a cultural value. The new model can be used as a baseline for capturing that cultural value and also creating transparency about the underlying logic. That means that at some point the data generated by applying the model must be analysed and from that analysis specific clusters of data may become apparent. These clusters can be positioned as genres; a specific kind of movie based on metadata. A simplified example is a cluster of movies that take place in the north-west of the Americas during the 1849 – 1890 period, which can be subjectively defined as westerns.

The meaning attributed to the clusters isn’t fixed. One of the reasons that genres are hard to define is that categorising movies can be impacted by societal changes and that genres evolve over time. That subjectivity still applies. The meaning attributed to the data in the model can be adapted over time to keep up with the zeitgeist. The underlying data however stays the same, although the model can be enriched with additional components.

A model that doesn’t attempt to capture an objective truth. Even if the universal truth in art can be found, it’s not necessary to know it to build a data model. It doesn’t matter if grass is classified as green or as red, as long as the classification is applied consistently. In other words, this article is the first leg of a journey to formulate a consistent subjective method. The next step is to describe how to deconstruct movies to generate consistent metadata.

Rhizome