While learning about network graphs, I was intrigued by the way that they could graph relationships between fictional people. My bachelors degree is in Writing and Literature, so the idea that one could use a visualization to analyze creative work was fascinating to me. It’s something that I could see doing in my own research as an academic, as well as something I could see doing as a high school librarian in conjunction with an English teacher, so for this lab I created a character network graph. I decided to make my own dataset rather than use an already existing one for a couple of reasons. One, it would give me a better idea of how datasets for this type of visualization are constructed, and two, if I had a student group do a project like this I would want them to use familiar material so that the graph, and the process of making it, could be a new level of analysis on top of what they had already done.

I chose Brian K. Vaughn and Fiona Staples’ multiple-award winning comic series Saga to work with. Saga, a space opera/fantasy, is the story of two soldiers, Alana and Marko, from opposites sides of a long-running war who have fallen in love, run away, and are having a baby. Leaders from both sides of the conflict fears the reaction of their war-sick people if news got out about the family, and have decided that they must be wiped out. The story begins with Marko, Alana, and newborn Hazel on the run and dodging death, all while trying to reach somewhere in the galaxy where they can be free. Saga is a story rich with characters, all with different motivations, alliances, and stories, and Vaughn, the series writer, deftly spins together separate narratives into one interlocking epic. Due to the large cast of characters and the complexity of their relationships, I felt that Saga was the perfect piece of literature to turn into a network graph.

While preparing for this visualization, I decided to seek out other visualizations that dealt with fiction, comics, or other types of media. The first one I found was The HBO Recycling Program graph, which shows the connections between actors who have appeared in 3+ episodes of HBO original series and the shows they have been on. While this graph contains a lot of names, it is very readable and due to the size of the nodes it is easy to see which shows have utilized a large number of actors on the list. However, sometimes the colors are too similar to easily track the edges from actor to show, and the show layout may have been more interesting if it was ordered differently (date of the show’s premier, number of actors used, etc.). The second is the Hamlet Character Graph from the Stanford Literary Lab, a digital humanities research group. This graph neatly lays out all of the play’s character, differentiates visually between main characters and minor characters, and has clear edges. However, there is no way to tell how frequently the characters interact with one another and without accompanying text, it is very unclear what the edge colors mean. The third is the Marvel Comic Book Artist Collaboration Networks series, from All Things Graphed, which graphs the relationships between comic creators who worked on different Marvel comics. These graphs were especially helpful to look at, since they were made with Gephi. They portray a lot of information without becoming very overwhelming, though sometimes the colors are dark enough that it becomes difficult to read the names. It would be also interesting if there was a search function, so it would be easy to zero in on a specific creator who might not be one of the larger names.

For my own visualization, the first step was to create the dataset. At this time, Saga is composed of thirty issues, approximately 21 pages each, and split into five distinct story arcs. As this was a lot of material to review, I felt it best if I didn’t assign directions to interactions. I also decided that I would only count interactions between characters once per issue to simplify the process, and also because often one interaction takes place over the course of several ‘scenes’ of an issue due to the way the narrative is structured. I did make an exception for interactions taking place in flashbacks/dreams/hallucinations/visually rendered memories, as these moments are usually significant to the story and character development, as well as being separate from the action taking place in the present. I only recorded interactions between characters who were named and appeared more than once in the story. The only exceptions were if a named character appeared once but was clearly significant to the story, or if it seemed they were likely to appear later on in the series.

I imported the data into Gephi, which showed that I had recorded 46 distinct nodes (characters) and 190 edges (interactions). I used the ForceAtlas2 layout to create the graph, and checked “Prevent Overlap” so that each node would be visible and easy to label in the final version. I ran statistics for both Average Degree (8.261) and Modularity (0.453). For Modularity, I left the resolution at 1 and it showed 5 unique communities. I experimented to see if it would breakdown the nodes into a greater number of communities, and to do this I had to bring down the resolution to 0.25, where it showed 10 unique communities. However, thinking of my own analysis of the character relationships in Saga, 5 communities made more sense and I returned the resolution to 1. I then portioned the nodes by Modularity class, and ranked the node sizes by number of degrees. The original degree range was 1-22, but in order to make the nodes on the mid-lower end of the degree range more visible I set the maximum size at 11. I also reduced the edge weight influence in ForceAtlas 2 in order to make the edges between characters that interacted frequently more distinct. In Preview, I added in the character labels, adjusting the font size so that names did not overlap, and slightly lightened the node border color so that some of the mid-sized node labels were easier to read.

Based on this graph, the characters in Saga can be broken down into five major groups. I named these groups The Family (Purple), The Robot (Blue), The Revolution (Red), The Freelancers (Green), and The Journalists (Yellow). These groups suggest that Saga, by this point, has focused on five separate narratives that I have come together to create one overarching story. Some of these narratives are closely connected, like The Revolution and The Family. This is because in the 5th story arc the majority of The Family is constantly interacting with the members of The Revolution. Meanwhile, the narratives of The Freelancers and The Journalists are mostly self-contained, occasionally linking to the other narratives via one or two edges. The characters in these groups frequently interact with each other, but rarely interact with other people and sometimes never do. For example in The Freelancers Group, Sophie is constantly interacting with the other characters in her hub, as seen by the weight of the edges that connect her to other nodes, but she herself has never interacted with anyone outside her group. While The Freelancers narrative as a whole is tentatively connected to the other narratives through characters like Gwendolyn (connected to the Family through several single interactions) or The Brand (connected to The Journalists), Sophie’s story remains self-contained with no influence from the other narratives of Saga.

This graph can be used for individual character analysis as well. According to Gephi, the average node has a degree of 8.261. This means that on average, the 46 identified characters in Saga interact with eight other characters. The characters with the highest degrees are Alana (22) and Prince Robot IV (21), which can be easily seen in the visualization, as those are the two largest nodes. However, despite the fact that these characters have almost identical degrees, their position in the network and the weight of their edges shows that they have very different stories and different kinds of interactions. Alana, at the center of the The Family group, has strong relationships with the majority of characters she interacts with. The edges that connect her to the the characters in her group have heavy weights, and the majority of relationships she has with characters outside her group have heavier weights as opposed to light weights caused by one or two interactions. By looking at this, we can assume that Alana is an extremely social character whose story arc revolves around her relationships with the other characters. Her family unit, which is at the center of the story, defines her character and the majority of her interactions are with those within that unit or with characters who belong to a closely linked group like The Revolution. Prince Robot IV, on the other hand, has many interactions but the edges that link him to other characters, even the characters in his group (The Robot) have fairly low edge weights. The nodes in his group are pushed away from him, and are even drawn closer to other groups like in the case of Yuma and Ghus, who belong to The Robot group but are more closely connected to Marko in The Family group than to Prince Robot IV. This suggests that Prince Robot IV is a solitary character whose narrative focuses less on the number of interactions he has with other characters, but more on his own personal story arc that is advanced through many single interactions with characters throughout Saga.

In future versions of the visualization, I would like to add explanatory features, such as a title, so that the visualization could be circulated without accompanying text. It could also include a legend labeling the groups to make the coloring more clear to users who are interested in literature analysis, but who may be unfamiliar with reading network visualizations. Also, as Saga is not a finished story, it would be interesting to add new character data to the network after the completion of each story arc until the series is finished. Perhaps new plot developments would lead to an increase in the number of groups, cause groups to merge, or cause characters to shift from one group to another. The shape of the network as a whole might change as well, moving closer together or becoming more spread out. Creating visualizations after each story arc could also give users another way to analyze the events of each story arc and how the story arc fits into the larger narrative.

This visualization could also be used as part of a larger project to analyze the writing style of a particular creator. For example, Brian K. Vaughn has written several other long-running comic book series, all with distinctly different stories. Character network visualizations could be created for these series as well, and then the visualizations could be compared to analyze Vaughn as a writer. Does Vaughn use similar narrative structures throughout his work despite the difference in content? What is the same about the networks? What is different? How does the specific content of a story influence a network? A series of visualizations could provide an intriguing alternative way to analyze not only a single piece of literature, but a group of literary works.