Harry Potter is a novel and movie series about a boy who finds out he is a wizard. He goes to a wizarding school called Hogwarts, has many adventures, and has to defeat an evil Wizard at the end of the novels. Many different characters shape Harry’s life, and the overarching theme in this series considers the power of friendship, love, and loyalty between many complicated people in the face of evil and oppression.
It is the third most read book series globally, just behind the bible and quotations from the works of Mao Zedong.
The boy who lived. Main character.
One of Harry’s best friends.
Harry’s first and best friend at Hogwarts.
Potions Master at Hogwarts. A complex character who is both an antagonist and protector of Harry.
Harry’s bully/most disliked person at Hogwarts.
Harry’s enemy. He killed Harry’s parents and was thought to be destroyed, but wasn’t.
The Harry Potter series, known for its rich storytelling and dynamic characters, has captivated audiences worldwide. From Harry’s loyal friendships to his intense conflicts, every interaction contributes to the series’ themes of love, loyalty, and resilience. But what if we could analyze and predict Harry’s relationships and interactions throughout the series using machine learning techniques? This project aims to explore that possibility.
This project draws inspiration from creative AI experiments, such as the “Harry Potter written by artificial intelligence” article, where neural networks attempted to generate a continuation of the Harry Potter series. These humorous yet insightful attempts highlighted the potential of AI in storytelling, particularly in modeling and predicting character interactions based on established narratives.
The Harry Potter Dialogue (HPD) project serves as a foundation for our work. HPD used scenes from the series to train a predictive model capable of generating responses to scene prompts. It analyzed:
For example, HPD could predict how Harry might respond to a conversation with Draco Malfoy in a given context, based on their rivalry and the tension built over time.
This type of work can be applicable to many different industries and deliver key insights. While this project looked at the idea of responsive AI in terms of video games, where a player makes choices that affect npcs and the environment, we wanted to analyze it in the broader context of ML in literature and socialization networks such as dating apps.
Here are some related projects and research that provide context and inspiration for this work:
Dating Apps’ Darkest Secret: The Algorithm – An exploration of how dating app algorithms shape user interactions and preferences.
Predicting Relationships Using ML Models – A research paper on applying machine learning techniques to predict interactions and dynamics in social networks.
The Paradox of Suspense – A philosophical analysis of suspense in storytelling, providing insight into narrative structures and audience engagement.
This project seeks to answer questions like:
7 Excel Files (Attributes) → all_books_combined_attributes.csv
7 Excel Files (Relations) → all_books_combined_relations.csv
Input Data
Combine Data
all_books_combined_attributes.csv.all_books_combined_relations.csv.Standardization
name, book, chapter).lineage column (e.g., “Pure blood wizard” →
“Pureblood”).Handling missing values by replacing NA with
"Unknown" to ensure completeness.
Dropped unnecessary columns
Mapped and Consolidated Character Names
Merge Datasets
all_books_combined_attributes.csv and
all_books_combined_relations.csv into a single file,
merged_data.csv, using shared keys: name,
book, and chapter.Final Refinement
updated_merged_data.csv.In this section, we explore Harry’s relationships with characters in the dataset. We analyze trends in affection, lineage, gender, and friendship through visualizations and statistical models.
Visualizations:
Unfortunately, when trying to go about a kNN model for answering this question, we ran into issues that in the end would require significant data overhaul and cleaning - pretty much row by row translations, etc.
Below is our initial attempt at answering this question with a model, linear regression. Here are a few characters. These are still interesting to look at - for example, Snape may be more likely to be considered a classmate rather than a teacher.
This project faced several notable challenges, largely stemming from the nature of the data and the complexity of modeling human behavior. Key limitations include:
Defining numerical attributes for intangible concepts such as “closeness” or “familiarity” presented significant obstacles. These are inherently subjective and difficult to quantify, which may introduce biases or inaccuracies in the analysis.
Translation and data integrity issues were a persistent challenge. Given the project’s origins in Hong Kong, much of the project involved cleaning, consolidating, and correcting mistranslated or inconsistent information to ensure the dataset’s usability and reliability.
Modeling human connections, such as friendships or romantic relationships, is inherently complex and prone to error. Human interactions are influenced by numerous external and internal factors that cannot be fully captured in a dataset, especially one derived from fictional narratives.
The dataset is based on a fictional universe, which introduces constraints in generalizing findings to real-world contexts. The relationships and behaviors in the Harry Potter series may not reflect realistic or nuanced human interactions.
A significant portion of the project was dedicated to cleaning and consolidating the data. This effort highlights the inherent messiness of datasets created through non-standardized processes, particularly when human judgment and interpretation play a role.
Predictive models, while insightful, cannot fully replicate the nuances of human relationships as portrayed in a book series. Comparing model predictions to the actual story outcomes emphasizes the limitations of such models, even as they provide valuable insights.
While the project offers interesting findings, its broader applicability is constrained by the specific context of the Harry Potter series. However, as digitalization increasingly influences areas like finding friends, roommates, and romantic partners, efforts to analyze and derive insights from relational data remain essential.
By recognizing these limitations, the project demonstrates the challenges of working with subjective, messy, and context-specific data. These challenges also underscore the importance of carefully interpreting results and acknowledging the constraints of predictive modeling.
We can potentially have the ability to predict book endings, character relationships, and responsive outcomes with machine learning models.
Harry’s closest friends are Ron, Hermione, Neville, Ginny, and Hagrid - mostly expected, but when breaking down their relationships in different categories, it’s exciting to see!
We may be able to predict connections between people based on previous data and lived experiences, which offer interesting possibilities in the realm of digitalization - things such as forming teams in the workplace, dating applications, and social networking.
Looking forward, this project opens the door to several exciting extensions and applications. Some of our future plans for this project include:
Interactive Guessing Game: Develop an interactive tool where users can input specific qualities of a character (e.g., gender, lineage, appearance) and receive predictions about Harry’s affection level toward them. This tool would use the existing predictive model to simulate Harry’s potential reaction to new or hypothetical characters.
Character Identification Game: Create a game where users receive clues about a character’s attributes (e.g., lineage, looks, or affiliation) and guess which character is being described. This would make use of the dataset to generate engaging trivia-style questions.
Relationship Prediction Tool: Expand the dataset and predictive model to allow users to guess the type of relationship Harry would have with a character based on their attributes. This could include categories such as “friend,” “teammate,” “opponent,” or “neutral.”
Improved Predictive Modeling: Refine the predictive models used in the project to better account for subtle nuances in relationships. This might involve experimenting with more advanced machine learning techniques or incorporating additional data sources.