Welcome to our deep dive into the world of anime - not through streaming platforms, but through networks, graphs, and communities.
This project explores how anime, genres, and users connect inside MyAnimeList using network analysis. Instead of asking “What is this anime about?” we asked:
- How do genres cluster?
- How do anime bridge communities?
- And how do users actually move through this ecosystem?
A previous study by Sosa et al. (2024) analysed anime communities using NLP on synopses. Cool idea... But after exploring their research we wondered:
🧐 Do summaries really drive user behavior? Or do genres, ratings, and completion habits shape communities more strongly?
So we rebuilt the analysis using the same Kaggle dataset, but with a different focus:
- 🎭 Genre co-occurrence
- 📺 Anime-genre bipartite mapping
- 👥 User preference structures
Disclaimer: 3/5 files from the dataset can be found in the data folder of this repo, the other two are very big so we suggest to open them using Row Zero, a spreadsheet tool for big data.
From the full dataset, we focused on:
anime_synopsis.csv→ for genre tagsrating_complete.csv→ only anime users marked as completed
Why only completed? Because someone who finishes an anime probably had a more stable viewing experience than someone who dropped it halfway.
To handle scale:
- The original dataset had 100+ million rows.
- We sampled 10,000 users using
numpy.random.choice.
Open the Live Networks before continuing:
We built a weighted network where:
- Nodes = genres
- Edges = genres appearing together in the same anime
- Weight = how often they co-occur
📊 Results:
- 41 genres
- 726 connections
- Very dense network (low modularity ~0.12)
Both Greedy Modularity and Louvain detected 4 consistent communities:
- ⚔️ Action & Adventure
- 💕 Romance & Slice of Life
- 🌑 Dark & Psychological
- 🎵 Niche
Low modularity isn’t a weakness! It shows that anime genres heavily overlap.
Examples:
- Attack on Titan → action + military + psychological
- Your Lie in April → romance + drama + slice of life
- Sword Art Online → action + game + romance
👉 Genres don’t exist in isolation, anime is inherently hybrid.
We assigned each anime to the genre community most represented in its tags.
If an anime connected evenly across communities, it became a bridge anime.
The strongest bridge wasn’t between Action & Dark.
It was between:
💕 Romance & Slice of Life 🎵 Niche
This suggests niche genres often blend into softer, character-driven storytelling.
✔️ Over 50% of anime had strong confidence (>0.75) in their assigned communities, supporting the structure.
Instead of simply assigning users to communities, we calculated:
What percentage of a user’s completed anime belongs to each community?
This revealed two viewer types:
Watch across multiple communities.
Have >40% preference in one community.
- 💕 Romance & Slice of Life had the highest overall engagement.
- 🎵 Niche had the highest percentage of specialists.
👉 Fewer people watch niche anime, but those who do are highly committed!
Anime, genres, and users form a tightly connected ecosystem.
- Genres shape anime structure.
- Anime structure shapes user patterns.
- Users move across communities, but not equally.
Unlike synopsis-based NLP clustering, genre networks:
- Reflect platform structure
- Align with user behavior
- Avoid noisy keyword artifacts
- Data from 2020 (snapshot in time)
- Only MyAnimeList users (mostly Western / English-speaking)
- Based on Kaggle’s scraped dataset
Future expansions could include:
- Cross-platform comparison
- Longitudinal analysis
- NLP on user comments
- Image/video recognition of fan culture (AMVs, edits)
- Python
- Pandas
- NumPy
- NetworkX
- python-louvain
- Google Colab
Recommendation systems often treat media as isolated items.
But anime isn’t isolated.
It’s:
- A cultural ecosystem
- A hybrid genre web
- A dynamic fan network
And sometimes, the strongest communities aren’t the biggest ones but the most devoted ones.
If you’ve ever:
- Gone from a mecha anime to a romance unexpectedly
- Fallen into a niche genre spiral
- Or insisted you “only watch one type” but somehow don’t
You’re already part of the network.
This project was developed by a group of five students in the Network Analysis course (2025/26) taught by Prof. Saverio Giallorenzo at the University of Bologna for the MA degree in DHDK:
✨ Built with curiosity, critique, and a lot of graph theory. Check out the full report here!