https://ada-data-explorers.github.io/

The Formula for a Perfect Movie is a project that aims to uncover the complexity of cinematic success by examining diverse features, including ratings, revenue, genre, inclusivity, and more. Motivated by the nuanced nature of film achievement, we introduce a new Movie Score metric that considers ratings and revenue, providing a robust measure of a movie's success. We aim to offer valuable insights into the secrets of successful filmmaking. By considering diverse and valuable data sources, this project seeks to explore the value that can be extracted thanks to Data Science.

In our data story, we guide our readers through the investigative processes that were utilized on our journey to detect the key influencers of good movies. As we unravel the insights embedded in the data, we hope that the reader will be encouraged towards introspection, thinking about what they believe makes a successful movie.

In this project, we aim to tackle the following research questions:

• How can we define the success of a movie?
• Can we consider ratings and movie revenue to determine a success metric? Is there a correlation between ratings and movie revenue?
• Is there a formula that successful movies follow?
• Which features influence the success of a movie?
  • Considering: movie duration, language, geographical location of the setting, budget, genre, ethnic inclusivity, fame of actors, fame of producers, movie plot sentiment analysis, and movie sequels.

To answer our research questions, we will be using additional datasets:

• To study the effects of inclusivity on a movie’s success, we needed actors' information so we scraped the Wikidata website. The dataset generated includes information about the actors’ ethnicity.
• Movie ratings are an important part of our project, yet we have no information about it in the given datasets. Therefore, we have a bot that looks for each movie on IMDb and extracts relevant information ( e.g. ratings, number of ratings, revenue in the United States and Canada, global revenue, and revenue on the week of release) was used:
• To adjust box-office revenue for inflation, we fetched data from the US Department of Labor Bureau Statistic.
• To know which movies were sequels, we used a dataset with movie-series.

For P2: In order to realize this project, we collectively augmented and preprocessed the data. Then, when we noticed that there was little correlation between movie ratings and revenue, we established a movie success score that considered both. After that, each team member took a specialized aspect and examined if and how it shapes a film’s success or failure in terms of the defined score. Aymeric delved into temporal factors and population dynamics, understanding their correlation with financial success and viewer opinions. Yara studied genre analysis, as well as the effect of sequels. Eric cleaned and explored inclusivity. Anthony analyzed features like duration and language, while Anton focused on the influence of actors’ fame. Finally, we brought our insights together and started building the recipe for a good movie.

For P3 please view Contributions at the end of the README

To find the recipe for a good movie, we follow this methodology:

1. Data augmentation and processing

   One can’t start working with data before cleaning it and preparing it, which is why we carry the following steps:

   • Find datasets (e.g. Kaggle, scraping IMDB and Wikidata) containing additional data we need for our project and don’t have yet (e.g. movie ratings, inclusivity in movies, inflation, …).
   • Fill NaN values for inclusivity with data from more datasets (found online from scraping).
   • In the movies datasets, remove the movies without revenue and ratings data.
   • Adjust box office revenue to inflation to have comparable values.
   • In the actors datasets, remove lasting NaN values for inclusivity data.
   • Preprocess the data: convert data in each column into a more convenient format.
   • Analyze data balance and decide what to do if imbalanced, depending on the feature being analyzed.

2. Analysis of the correlation between movie revenue and rating Film directors define a good or bad movie according to its ratings and the revenue it generates. But are these two elements correlated? To answer this question, we conduct a Pearson correlation test and visualize the two distributions

3. Definition of a success score metric that considers both rating and revenue Since movie ratings and revenue are not significantly correlated, we define a movie success score metric (Movie Score), that considers them both:

   • $BORC$ : Box Office Revenue Component
   • $RC$ : Rating Component

   These components can be computed with the Adjusted Movie Revenue (we will write as $AMR$) and the movie rating (we will write as $MR$)

   We log-transform the data and define $x = \log(AMR)$ and $y = MR$

   $BORC = \frac{x - \min(x)}{\max(x) - \min(x)}$

   $RC = \frac{y - \min(y)}{\max(y) - \min(y)}$

   As such, we have $BORC, RC \in [0, 1]$

   We define a weight $\alpha \in [0, 1]$ and take the convex combinations of $BORC$ and $RC$. This weight controls the importance we give to each of our two components in our metric. Its importance will be determined in the future and we will adjust it accordingly to the study. Multiplying by 100 gives us a final score

   $$Movie Score = 100\left(\alpha BORC + (1 - \alpha) RC \right)$$

   Only then can we define two classes of good and bad movies by defining a percentage threshold on movie scores. We chose alpha = 0.5 to consider both components equally.

4. Find the effect of each feature on the movie’s success score.

7. Find the importance of the grouped importance of features

The initial feature analysis gives insight on the importance of each feature individually. But this does not take into account how the features can complement each other in regards of predicting the movie score.

Our strategy is to process features further (scaling, feature creation ...) and feed them into regression models that give insights about the importance of our features by looking at the model's weights, such as linear regression. We want deep insights, not just singular patterns that influence the movie scores.

7. Finally, a recipe for good movies …

From the above tests, we identify the features that affect the movie’s success the most thus defining a “recipe for good movies”.

In light of the feedback that we got for P2, we chose to remove from our roadmap the validation with unseen data and the neural network.