| Student's name | SCIPER |
|---|---|
| Néhémie Piero Frei | 346850 |
| Adrian Martinez Lopez | 396379 |
| Clara Irina Hamousz | 326814 |
| Arthur Taieb | 361195 |
Milestone 1 • Milestone 2 • Milestone 3
10% of the final grade
This is a preliminary milestone to let you set up goals for your final project and assess the feasibility of your ideas. Please, fill the following sections about your project.
(max. 2000 characters per section)
We will use the Letterboxd (Movies Dataset) from Kaggle. This comprehensive resource is actually composed of 10 interconnected datasets (e.g., movies.csv, themes.csv, actors.csv), containing metadata for over 950,000 films. Data Quality and Preprocessing: Overall, this is a high-quality dataset with a large portion of its data already well-populated. The largest tables, such as those detailing actors and movie countries, are very complete. While there is some missing data in more granular fields—for instance, the specific character role is null for about 25% of the actors—the core attributes remain extremely robust. For the movies themselves, essential columns like duration, description, and release year are available in more than 80% of the cases. Our preprocessing strategy will consist of straightforward data cleaning: we will drop the incomplete rows to keep only the clean data. Some columns that are not relevant for our work will also be dropped. Because the initial volume is so massive, even after this filtering process, our working dataset will remain extremely large, varied, and highly interesting to look at for our visualizations.
Our motivation for this project comes first from a shared interest in cinema and its history, but also a regular use of the platform Letterboxd. Our goal is to show data from a social media site (where movies and information about them are discovered through the activity of other users) in a completely different way. Our visualization is meant for people with a general interest and curiosity for cinema, but not necessarily any prior knowledge about film history. To achieve this goal, we plan to present the data from letterboxd through three main lenses :
We want to bring out the differences and specialities of local cinema across the world (highest rated films, most prolific directors, most popular genres, …) but also allow people to discover movies and directors from smaller countries.
We want to highlight information about how cinema evolved over time : movie length, genres distribution, size of cast and crew, …
We want to present the relationships between the people who appear and participate in the production of movies. We will try to highlight people who often collaborate together or artistic movements in cinema where important figures tend to be highly connected.
The dataset contains 941,597 movies from Letterboxd organized in a star schema: a central movies table joined by id to 9 related tables.
This is a really short EDA showing only the core features of the dataset. A more detailed EDA can be found in extended_EDA.md.
It is interesting to see that 90% of movies do not have an average rating or tagline! We have seen in the data that only more recent movies (from 1990-2000 onwards) have a rating. This might be problematic when trying to provide rating-related visualizations. The remaining gaps (runtime 19%, description 17%, date 10%) are moderate and manageable.
The distribution is left-skewed: users on Letterboxd tend to rate positively, with the bulk of ratings between 3.0 and 3.5. Very few movies fall below 2.0.
Movie production volume seems to grow exponentially over the decades, with a sharp peak in the 2020s. There is a noticeable dip around 2020 (probably caused by COVID-19).
The length distribution reveals three distinct groups:
The short film segment is surprisingly large (37%). The extreme outliers (up to 72,000 min = 50 days) must be filtered.
Drama dominates with 232K entries, followed by Documentary (164K) and Comedy (141K). Each movie can have multiple genre tags.
English dominates with 473K entries (45% of all language assignments) as expected whereas the rest do not surpass the 10% mark.
The USA accounts for 174K movies, 3.8x more than France in the second place (46K). The top 5 (USA, France, UK, Japan, Germany) represent the bulk of cataloged films.
Beyond personal analytics dashboards (tools that visualize a single user's watch history in terms of top genres, favorite directors, or rating habits), the most notable existing uses of this specific dataset are two Kaggle notebooks: "What the World be Watchin?" by Mike DeLong, which produces geographic visualizations: showing movie production counts and dominant genres by country, and "Classify genre using posters" by スワミ, which uses poster images for ML-based genre classification. More broadly, film network analysis is a well-studied area: the Hollywood co-stardom graph (where actors are linked by films they share) has been explored since Watts & Strogatz (1998) and has been visualized using tools like Gephi and Sigma.js. On the poster side, Vijay Pandurangan's 2012 project analyzed color trends across 35,000 movie posters over time. However, all of these remain static, single-dimension analyses with no interactive or narrative component.
Our approach is original in that we combine four different lenses on the same large-scale dataset: geographic, temporal, social, and visual, in a single interactive experience. We want to go beyond ranking top countries or genres, and instead make world cinema genuinely discoverable: surfacing lesser-known national cinemas, showing how genres have emerged and spread across the world map over decades, highlighting how filmmaking has structurally changed over time, mapping the collaborative networks that define artistic movements, and exploring how poster aesthetics have shifted across eras and genres.
For design inspiration, we look to The Pudding for its storytelling approach, and to Gapminder and Information is Beautiful for the idea of animated, interactive charts that make complex temporal data (more) intuitive. For the network component, we draw inspiration from existing Hollywood collaboration graph visualizations, which we aim to extend to the full dataset including non English and lesser-known cinema.
10% of the final grade
80% of the final grade
- < 24h: 80% of the grade for the milestone
- < 48h: 70% of the grade for the milestone