In film storytelling, a character’s name is rarely just a label; it often serves as a subtle cue to their personality and role in the movie. Our project tackles the intriguing question: "Can a character’s archetype be predicted from their name ?" Through this data analysis project, we aim to decode connections between specific name characteristics — such as length, structure and phonetics — and the characters' features including gender, age, the movie genre, its origin and even its morality.
Additionally, we will investigate how naming conventions differ across contexts, with a particular focus on comparing movie productions from the United States. Specifically, we aim to identify strong correlations between the portrayal of "nice guys" and "bad guys," focusing on potential differences in the depiction of Soviet characters versus American characters.
About discovering characters' features based on their first name:
- Is there a correlation between a characters' name and their gender ?
- Can we observe a relationship between a character's name and their age ?
- How does a character's origin influence their name ?
- Are character names linked to the movie genre they appear in ?
- Is there a pattern in character names that distinguishes Nice guys from Bad guys in movies ?
About how naming conventions differ across contexts:
- How do naming conventions in U.S. movie productions during the Cold War reflect ideological narratives, particularly in the portrayal of Soviet characters as "good guys" or "bad guys" ?
The CMU Movie Summary Corpus is a dataset provides a detailed view of over 42,000 movie plot summaries, extracted from Wikipedia and aligned with metadata from Freebase:
-
Movie Metadata:
- Box office revenue
- Genre
- Release date
- Runtime
- Language
-
Character Information:
- Names of characters
- Metadata aligned with the actors who portray them, including:
- Gender
- Estimated age at the time of the movie's release
-
Processed Data Supplement:
- Stanford CoreNLP-processed summaries, which include tagging, parsing, Named Entity Recognition (NER), and coreference resolution.
Source: CMU Movie Summary Corpus
The Name Ethnicity Dataset, sourced from Kaggle, provides a collection of names alongside their corresponding ethnic origins.
Dataset Features:
- Name Data:
- First and last names
- Associated Metadata:
- Country of origin
- Ethnicity information
Source: Name Ethnicity Dataset on Kaggle
├── data # Contains preprocessed data or intermediate data
│ ├── Image
│ ├── Plot
│ ├── cleaned.csv
│ ├── Ethnicity_on_character_names.csv
│ ├── movie_character_ethnicity.csv
│ ├── name_ethnicity.csv
│ ├── sentences_by_character.csv
│ └── sentimental_analysis.csv
│
├── plots # Plots for website
├── predictive_model # Predictive models developped to predict a Character ID
│
├── src
│ ├── MovieSummaries # CMU Movie Summary Corpus
│ ├── new_dataset # Name Ethnicity Dataset
│ ├── scripts
│ │ ├── load_some_dataset_and_save_it_in_data_directory.py # Preprocessing of CMU Movie Summary Corpus
│ │ ├── preprocess_ethnicity.py # Preprocessing of Name Ethnicity Dataset
│ │ └── sentimental_analysis.py # Sentimental analysis script
│ │
│ └── utils # Functions used in the data preprocessing, result notebook or for models
│ ├── __init__.py
│ ├── data_utils.py
│ ├── ml_utils.py
│ └── results_utils.py
├── test
├── .gitignore
├── predict_fct.py # Final function that calls the 4 models and returns the Character ID
├── README.md
├── requirements.txt
└── results.ipynb # Jupyter notbook containing all the results
We began with data cleaning, removing irrelevant columns and rows, particularly those missing the character name. For preprocessing, we isolated first names by:
- Stripping out prefixes (e.g., “Dr. Alison Parker” becomes “Alison Parker”)
- Eliminate commun names like “taxi” or “waiter” using
nltk.corpus.words. - From the filtered data, restore words identified as valid names using the
nltk.corpus.nameslibrary, which provides a list of English names. - Keeping only the first name from full names (e.g., “Alison” from “Alison Parker")
The cleaned data was saved in the cleaned.csv file for streamlined analysis.
We extracted the followinf name characteristics:
- Length, vowel and consonant counts
- Structure :
- first and last letters
- n-gram
- Phonetics (Soundex)
We proved significant differences in the structure of female/male names.
To better understand the relationships between character names and movies, we started by looking at the data to identify patterns and formulate hypotheses. Our goal was to find trends in character features, such as the distribution of names across different movie genres, countries and ages.
We identified statistically significant associations between certain names features and character features from the CMU Movie Summary Corpus Dataset (Genre, Age and Movie Genre). Then we created 3 models to link the Character Names to a Genre, Age and Movie Genre.
4. Creating a Model from Name Ethnicity Dataset to predict the Origins of the Characters from the CMU Movie Summary Corpus Dataset
We identified statistically significant associations between certain Names Characteristics and their Origin in the Name Ethnicity Dataset. Then we created and model to link the Character Names from the CMU Movie Summary Corpus Dataset to an Origin.
We classified the Characters from the CMU Movie Summary Corpus Dataset as "Nice guys", "Bad guys" or "Neutral" with a Sentimental Analysis performed on the Movie Summaries. Then we created and model to link the Character Names to a Morality.
Answering the Research Questions and finding patterns.
We aim to develop a “Character ID” model that predicts the most likely characteristics of a movie character based on their first name. The model would generate a character profile, including these attributes : Gender, Age, Movie Genre, Origin and Morality
- Luca: Models and Plots
- Ainhoa: Statistics and Plots
- Fanny: Sentimental Analysis and Website
- Zacharie: Models and Website
- Amaury: Models and Final Predictive Model