It’s Been a While Luffy! :: Predicting the Return of Past Characters in Eiichiro Oda’s One Piece
2023-12-13
Rhys Griffiths, Lux Donaldson, Soren Rusnak, Brenden Kobliha, Ben Shingledecker, Eric Garceia, Aierken Aikebaier, and Andrew Henderson
Fundamentally our goal is pretty simple. See if we can use edge list data networks to predict which characters will return in the popular manga and anime One Piece.
For our project, we imputed every character in One Piece before the time skip (The first major half of the story), and connected them to everyone else they have either fought or had a meaningful conversation with. We did this in order to see if we could predict which characters will return in later parts, by examining three different sets of data. First is all the characters, with none removed. Second is all the characters, minus Luffy, the main character. The final dataset is all of the characters except for the main character Luffy and the ensemble cast.
Legend (For the lower three networks): Color: Modularity Class Name Size: Eigenvector centrality
This network helps confirm just how central Luffy is to One Piece’s story. He has both the highest eigenvector score and degree by a huge margin.
Without Luffy, the balance between the highest ranking eigenvector scores evens out a lot alongside the network itself becoming much less centralized. Unsurprisingly to those who’ve read/watched One Piece, the next highest scores belong to the rest of the Straw Hat Pirates. The ensemble cast that leads the show alongside Luffy, the captain of said pirates.
Thirdly, after removing all of the Straw Hats, we get the clearest look at possible returning characters. The network also supports a theory we had running while collecting the data. That while connections to the Straw Hats are obviously important. If that’s your only important connection, you’re unlikely to turn up again. Characters with high eigenvector scores, and more specifically, characters with connections that bridge the gap between one or more modularity classes.
Major returnees include: Marshall “Blackbeard” D. Teach, a character set up as one of the series most central antagonists. Shanks, Luffy’s first and most important mentor. Edward Newgate, technically dead as of the end of Pre-time skip but they show up in major flashbacks and were one of the most important characters in the setting. Crocodile, the first major antagonist who connected past their story arc.
Speaking of modularity class. I believe it’s important to discuss what they represent in the actual story of One Piece. One Piece is separated into many different story arcs. With each one commonly taking place on a different island. As such, most of the modularity classes represent the stories’ separate major arcs. While eigenvector remains the foundation force when determining a character’s likelihood of returning, possessing connections that cross these classes is the equivalent to having connections that go beyond a character’s initial story. Thus drastically increasing their likelihood of returning.
A great example of this is Dracule Mihawk. While his eigenvector isn’t crazy high, he possesses a connection to almost every different modularity class. Which makes sense given his presence in the story. A character who doesn’t interact with others much, but is a massively important person in the setting itself.
Our data has also been shown to work across not just One Piece’s 1st and 2nd parts. But 2nd to 3rd as well. As a bit of context, One Piece is seemingly separated into three major parts. Pre-time skip, the part where all of our data comes from. Post-time skip, the part we are testing to see if our data successfully predicts returning characters. And the Final Saga, the very recently started, final portion of One Piece. While Rob Lucci, Kaku, and Kalifa did all return here and there Post-time skip, they very recently re-entered the spotlight in the Final Saga as temporary allies of the Straw Hats in the most recent story arc.
Obviously this data isn’t perfect. Our criteria for a “connection” isn’t as solid as it could be and our method for determining returnees isn’t as quantitative as I would like. However I think this is an excellent first use of this data and method. In the future we hope to take both these networks and statistics further. To hopefully map enough of One Piece to be able to predict where the series might end.