By Ali Shahed, Reinhard Cate, Saad Elbeleidy, and Jan Doesselmann
Here at Protagonist, like everyone else we’re waiting impatiently for next year’s release of the final Game of Thrones season on HBO. Since we can’t wait for the epic conclusion to the series, we decided to put our impatience to good use and ran every volume of George R.R. Martin’s “A Song of Fire and Ice” series through our machine learning model to see what we could learn. After ingesting the text, we trained the algorithm to predict the keywords associated with the major Houses of Westeros across all five books to see if just maybe we can try and unpack some deeper level insight, about the narrative around each house in every volume.
We’ve taken a stab, (no pun intended) at the narratives behind House Stark across the five books using our understanding of the model below. First here’s how to understand the visual: each node or circle represents a word. The words present in the same cluster and share the same color are most closely related to the central or largest node in that cluster. However, the direct relation of each node also highlights the interrelatedness of the entire network from cluster to cluster. Each network selected by the drop-down menu is built around a specific House of Westeros. We based our analysis primarily on the cluster around Stark in each book, while also taking into account the direct relationship of Stark to the other clusters (represented by different colors).
Interact with the model, look at the words most closely related to every major house across the five books using the drop-down menu for house and book volume to see what you can find. Thanks to the nature of the text, a fictional story, we’ve been able to find that the data provides somewhat of a predictive roadmap of the narrative around House Stark in each book that is easily interpreted. What can you discover about each house that could provide insight on the final books of the series? If you’re worried about spoilers or haven’t watched or read the whole series, stop now. This analysis is not for you. If you’re ready, then keep reading and we’ll walk you through how to do your own analysis.
Protagonist’s machine learning model and narrative analytics can, not only uncover deep insight about a subject, but the data can also be predictive. Here’s a quick walkthrough on how to start: first select a House from the drop-down menu, (it defaults to House Targaryen and volume one), and a book volume numbered one through five. We recommend starting at volume one “A Game of Thrones”. Once you’ve done that, find the primary node for the House you’ve selected in the drop-down menu by hovering over the network. Then starting from the primary House node, we’ll use Lannister as an example, investigate the keywords associated with the cluster around Lannister, (all the nodes will share the same color), but also examine the direct connections Lannister has to other clusters. Then try picking different book volumes, from “A Game of Thrones” to “A Dance with Dragons”. If you investigate how the clusters and connections around Lannister change, you’ll ultimately see what critical word associations with Lannister change from book to book. While we did our own analysis on House Stark below, we hope you’ll find some unique insights about House Lannister, House Baratheon, House Tyrell, and House Targaryen. And if you’re interested in checking out how our Data Science team built this model visit our Gitgub here. Serious Spoilers in our analysis below!
Tragedy for House Stark in Volume One “A Game of Thrones”
The words associated with Stark in this book need little interpretation. First within the Stark cluster, “die” is probably the most negative word you can get. Additionally, Stark is directly connected to the word “traitor”. If you have no experience with the books, these are strong warning signs alone for what’s to come for House Stark. The word “die” is closely related to the coma of Brandon Stark who is referred to as a “child” which is also in the Stark main cluster, and the resulting discussions about his potential fate from both friends and foes. The word is also related to discussions of Ned Stark’s potential fate at the hands of King Joffrey later on in the book. The words “truth”, “farce”, “brother”, “sister”, and “command” all within the Stark cluster capture the conflict between Ned Stark and Cersei Lannister eventually leading to Ned’s execution.
Revenge, Succession, and Red Wedding Foreshadowing in Volume Two “A Clash of Kings”
For our book readers and series-watchers, the execution of Ned Stark has a monumental effect on every house of Westeros. “Avenge”, is one of the keywords within the Stark cluster, which makes perfect sense. Robb Stark subsequently leads northern forces south against the Lannisters to avenge his father’s death. “King” is also a crucial word in the cluster, as Robb is declared King of the north. This gets interesting as we also see Stannis Baratheon and Renly Baratheon declaring themselves the rightful heir to the Iron Throne. Add in the word “brother”, which is also in the Stark cluster, and we see it being indicative of the Stark involvement in the conflict between the two Baratheons. Together with the word “Heir” which is also a part of the cluster, “King” and “Brother” all highlight the deep involvement of House Stark in the greater political drama over who is the rightful ruler of Westeros.
The cluster in closest proximity to the primary Stark cluster is centered around the word “Confide”. This word likely captures the internal conflict between Robb Stark and his mother Catelyn Stark over a potential alliance with House Fray through marriage. However, what we found incredibly interesting is the node “Robb” links directly between to the main “Stark” cluster and the “Confide” cluster but, also to the nodes “Liar”, “Marriage”, and “Die”. While there are certainly other aspects of the plot that these nodes could capture, like Sansa Stark’s arranged marriage to Joffrey, it could also be providing a bit of predictive foreshadowing for a certain bloody wedding that occurs in the next volume.
Volume Three “A Storm of Swords” or House Stark’s No Good Very Bad Day
The presence of the word “murder” in the Stark cluster for this book, should be no surprise for fans. Together with “honor”, “wife”, and “Lannister”, the data produced from the model reveals one of the most notorious events in the series, the Red Wedding. By not honoring his pledge to marry a Frey daughter, Robb Stark is betrayed by House Frey together with House Bolton who conspire with House Lannister to murder the young King of the North. The model also produces some interesting insight over control of the Stark castle Winterfell. The word “Winterfell” is central to the Stark cluster, readers and watchers know that the castle is captured first by the Greyjoys and then the Boltons. The word “bastard” is also critical in the model, as it highlights the importance of Jon Snow’s struggle beyond the Wall, and is indicative of the importance of Ramsey Snow, the bastard son of Roose Bolton who captures Winterfell. Are you following all of this?
Lady Stoneheart, Sansa, and Dorne in Volume Four “A Feast for Crows”
One first-degree connection to the Stark cluster, “Stoneheart” reveals a critical plotline missing from the HBO adaptation. A character central to the books and specifically the Stark family was the reanimated corpse of Catelyn Stark, known as “Lady Stoneheart” who sought vengeance on those responsible for her and Robb Stark’s murder at the Red Wedding. The direct connection from Stark to “Baelish” and the nodes “innocence” and “alive” speak to the plight of Sansa Stark who is in hiding with Petyr “Little Finger” Baelish in the Eyrie. The young Stark girl’s storyline on the run from the Lannister family is central to the Stark narrative. While Arya Stark’s training in Bravos is also a critical aspect of the Stark story, her decision to conceal her identity plays into the ability of the algorithm to model words around her storyline.
Interestingly enough the House of Dorne is a crucial aspect of the Stark cluster as well, and this plotline differing drastically from the HBO adaptation. Arrianne Martell’s attempt to crown Myrcella Lannister Queen of the Seven Kingdoms under Dornish law as a challenge to King Tommen and is captured in the node “Arrianne”. While the plot is broken up by her father the Prince of Dorne, this storyline is associated with the Starks and captured in the model by his name “Doran”.
Volume Five “A Dance with Dragons”: Threats and Red Wedding After Effects
The model describes yet another plotline separating the books from the HBO series, with the words “wedding”, “Castellan”, “Aegon”. The direct connection of “Castellan” to “Stark” represents attempts by Davos Seaworth, the right hand to Stannis Baratheon, to convince the northern houses to fight alongside House Baratheon against the Boltons. Seaworth reaches out to what’s left of the leadership of northern houses, specifically for House Manderly. The word “wedding” is also significant in a plot by the Ramsay Bolton to marry a “false” Arya Stark, in order to rally northern support to his family’s control of Winterfell. The direct connection to “Dreadfort” captures the imprisonment of Theon Greyjoy, who is sadistically tortured by Ramsay, eventually escapes alongside the fake Arya Stark, again differing from the plot shown in the TV adaptation.
One more major shift from the television series is the story of “Aegon” Targaryen, who show watchers know is Jon Snow, but for the books is an entirely separate Targaryen. Aegon was in hiding and came to Westeros with a mercenary army in “A Dance with Dragons”. The last interesting insight we saw from the model is from the words “Behead” and “Slynt”. We think these represent another subtle bit of critical foreshadowing. We see this also being indicative of the eventual betrayal and murder of Jon Snow by his brothers of the Night’s Watch at the end of the book. Snow’s decision to kill Janos Slynt drove a further wedge between himself and his opponents at the Wall and eventually lead to his own death at their hands.