Will the Mother of Dragons Survive or Die? Which Prediction Algorithm Is More Accurate?
By analyzing information from the American TV series "Game of Thrones", the survival probability of the main characters was predicted from a data perspective. This research method for fantasy stories will also bring inspiration to similar application cases in real life.
There are only two kinds of people in this world: those who watch Game of Thrones and those who don’t.
The Game of Thrones that you've all been longing for has finally ushered in its final season. The family motto of the Wolf Starks, "Winter is coming", has finally come true, and the war between humans and the White Walkers is about to break out.Can you guess the life and death of the main characters?
Is it only after watching each episode that you can know whether the Arya you care about will survive, and whether the Queen of Lust will be killed by Arya?
Faced with this torment, there are always people who can't hold back. For example,Programmers who love code used algorithms to analyze the stories in Game of Thrones and predicted the characters' survival probabilities and who was most likely to ascend the Iron Throne.
Interestingly, different algorithms and teams predicted very different results. For example, the algorithm of the Technical University of Munich (TUM) team predicted that Dragon Mother would eventually ascend to the Iron Throne, while another algorithm predicted that Dragon Mother would not survive the winter.
In addition, there is another magical "algorithm": Men all die.
So, which company's predictions are accurate? Each company has its own basis. Let's see how they do it.
Machine learning predicts that Dragon Mother will survive to the end
Students at TUM did an interesting and bold experiment in a computer science seminar:Use data science and machine learning methods to predict the final chance of survival of characters in Game of Thrones. Before the eighth season of Game of Thrones aired, algorithms predicted the fates of the characters.
The full list of Game of Thrones characters and detailed predictions can be found at
Get it online at https://got.show
(Image source: https://got.show/)
In this conclusion,Dragon Mother has the highest chance of survival (99%), her King's Hand, Little Devil, also has a 97% survival rate. Could it be that Dragon Mother finally led all the major families? And Bronn, who is cunning and good at self-preservation, ranks first in the order of death with 93.5%.
Where does this result come from?
TUM teamThe analysis data was collected from the books of "A Song of Ice and Fire", the lines of Game of Thrones, and the content of the fan circle on Wikipedia.Information used for predictions includes: the character's family, whether they are married, and their allies.
They extracted a dataset of hundreds of character traits, combing through information about the characters in the story. In addition to data such as gender and status, they also took into account metadata information: such as whether someone is a major or minor character, and how often they are cited in channels such as Wikipedia.
Through this data, using algorithm analysis, some trends about the mortality rate in Game of Thrones were revealed, such asMen are more at risk than women(The male mortality rate is 22%, while that of females is 11%).
To make predictions about the fate of each character while forecasting trends, the team used two separate models:The first is the Bayesian inference method, use the MCMC method packaged in pymc3 to train the Bayesian survival analysis model;The second involves machine learning and neural networks, carried out under the Keras framework in Python.
Their work may have some effect. Back in 2016, before the sixth season aired, students from the same course at TUM created an algorithm that accurately predicted Jon Snow's resurrection.
Using text analysis: Is the Mother of Dragons going to die?
You have to believe that there is definitely more than one person who has the same curiosity.
Another data scientist, Peter Vesterberg Through text analysis of the original novel, we can predict the plot direction of the final season.
Peter believes that the direction of the plot is supported by the relationships between the characters. By analyzing the five published books of "A Song of Ice and Fire", he used network theory to calculate the relationships between the characters, presented them in a visual way, and judged the final survival rate based on the importance of the characters.
He used a dot to represent a character, and based on the text in the Ice and Fire story, he used methods such as part-of-speech tagging and similarity measurement to define the degree of "closeness" between the characters. The closer and more frequently the two names appear, the more connected they are.
The method of judging the relevance of a role to other roles mainly uses four key concepts:
- Degree centrality - the proportion of nodes directly connected to the node to the total number of nodes;
- Closeness centrality - for a node, the closer it is to other nodes, the higher its centrality;
- Betweenness centrality – quantifies the number of times the node acts as a bridge for the shortest path between two other nodes;
- Eigenvector centrality — The importance of a node depends on both the number of its neighboring nodes and the importance of its neighboring nodes.
Through the analysis of these concepts, we can getThe "weight" number of the relationship valueThe final result is a graph of character relationships, where the thickness of the lines indicates how closely the characters are connected, and the size of the final nodes represents the importance of the characters. This importance index also shows the possibility of being "removed" by the author.
The specific indicator rankings are as follows:
Jon leads in the ranking of the four analysis indicators.
In this analysis, Jon Snow is undoubtedly the most important character. Will this suspected RMB player, who was born in the Dragon family and raised in the Wolf family, eventually ascend to the Iron Throne? Following closely in importance are the intelligence of Tyrion and the Kingslayer Jaime.
From the perspective of network theory, Daenerys, the Mother of Dragons, seems to have spent too much money in the early stages, and probably did not have enough gold in the later stages. She is just on the edge of the relationship network, and it seems that there is a high probability that she will be doomed.
Predictions are not just for entertainment, they also have practical significance
We don’t know which algorithm is better with different results, and maybe we can only know it when the finale comes. But the methods they use are not just for fun, but have practical value.
The survival probability algorithm developed by TUM comes from a serious learning project. The main purpose of setting up this course is toStudents will learn how to design, develop, and deploy intelligent computer systems.
Dr Guy Yachdav, lead supervisor on the project, said: “While predicting the survival chances of Game of Thrones characters relies on data taken from a fantasy world, using exactly the same AI techniques in the real world could have a significant impact on our everyday lives.”
"The combination of passion and teaching is a great way to create new tools. In our courses at TUM, we have found interesting ways to teach students how to use this technology," said Professor Burkhard Rost, who heads the Department of Informatics at TUM.
Currently, in the real world,Similar algorithms can be used in medicine and finance, for example to predict health outcomes using combined information analysis. This technique is similar to analyzing the effects of treatments or complications on cancer patients.
As for what data scientist Peter did, in addition to analyzing the survival of the characters, he also analyzed the various writing rules of the "A Song of Ice and Fire" novels from a digital perspective, using data to analyze the elements of a novel.
Perhaps, after mastering this pattern,In the future, novelists and screenwriters will be able to use artificial intelligence to quickly generate new content.This way, you won't have to watch Martin make up for the mistakes he makes.
Want to predict which hero will survive the Avengers?
The lively month of April seems to be another season of ending. In addition to the final season of Game of Thrones, The Avengers will also usher in the finale next Wednesday. I believe that many people have not yet recovered from the snap of Thanos' fingers, that ruthless man who insisted on watching the sunset quietly and randomly destroyed half of the universe.
This willful death rate is probably something that AI cannot predict. However, if you are still too curious and want to try it yourself, the TUM team's project open source address is here:
But in fact, there is no need for AI prediction. Avengers 4 should also be the best ending. After all, Doctor Strange used an exhaustive method, or trial and error method, to leave hope to the Avengers world: he tried 14,000,605 possibilities with the Time Stone and chose the only one that would lead to victory.
This spirit of Doctor Strange would probably be great for writing code.