I’m a huge Sherlock Holmes fan, and one of my favorite TV shows is the BBC’s Sherlock, which re-imagines Sherlock Holmes in 21st century London. Naturally, I couldn’t resist having an excuse to delve back into the show for this Gephi lab!
The following network visualization depicts interactions between main/recurring characters for season (or series, as it’s referred to in the U.K.) 1, season 2, and season 3 of Sherlock. Each season has three 90-minute episodes. I filtered out the minor characters because I was only interested in the “main” characters and their interactions. Thus, the visualization will show interactions between protagonists, antagonists, and recurring secondary characters (some of which are not recurring, but play a significant role [e.g., the “baddie”] in particular episodes across all seasons).
My major issue when generating data for the visualization was how, precisely, to qualify when characters’ interactions stop and start. An example that requires some operationalization is if a scene begins in the morning, then cuts to later in the day. If the characters are still in the same room and continue a conversation concerning the same subject as the conversation earlier in the day, is this considered two interactions or one? Literature Geek suggests that interactions are “chunk[s] of communication such as a dialogue, a speech directed at one or more specific characters…’chunk’ means that we’ll think of an interaction as ending…when communication ends, other events intervene….” I still had great difficulty trying to figure out a systematic way to generate data because the interactions were not clear-cut. In the end, I could only use my own judgment.
First, I tried watching (skim-watching, rather) season 1–I must have recorded and re-recorded data a few times, but I wasn’t satisfied with the results. I eventually remembered coming across fanmade transcripts of the episodes and found them on LiveJournal (the transcripts I used are courtesy of Ariane DeVere). It was easier to digest actions and dialogue in text rather than on screen.
The data I generated is certainly not perfect but since I am quite familiar with the show I can see that the visualization accurately reflects the strength or weakness of interactions between characters. I defined interactions as dialogue exchanged in person or electronically. In some cases, I recorded nonverbal interaction. Most interactions are undirected, with the exception of unanswered text messages or character A not hearing character B addressing him/her.
The viz layouts I used was Fruchterman Reingold, then Expansion and Label Adjust so the nodes (characters) weren’t overlapping. In order to see “relative connectedness” within the network, I placed the node size (by degree) at a minimum of 10 and maximum of 50 which shows who has the most, second-most, third-most, etc. connections in the network–the larger the node, the more general relations the character has. The edges are either emboldened or thinly weighted, which illustrates greater and lower amount of interactions between two characters, respectively. Lastly, I partitioned the nodes by degree, and the colors relate to certain numbers of relations.
If you are interested in knowing which characters have similar presences in the show, looking to the colored degree could be a way to determine this. For example, Mycroft Holmes and Molly Hooper are both neon green, and thus each have 6 connections.
From the viz, a few observations can be made:
- As the largest node, Sherlock Holmes has the most centrality (of course); in a way, the viz would look nearly the same as an ego network, with Sherlock as the focal point. Only one character (Ella Thompson, who is John’s therapist), you may notice, does not interact with Sherlock Holmes.
- Since they have the thickest edge weight, Sherlock Holmes and John Watson interacted the most with each other.
- As the node that is not quite as large as Sherlock and John but is larger than other characters, D. I. Gregory Lestrade has the third most interactions.
- Measuring by edge weights and interactions with the two protagonists, Sherlock and John, it can be determined that Mary Morstan, Mrs. Hudson, Mycroft Holmes, Gregory Lestrade, and Molly Hooper are the most present characters in the show.
- There are little to no female-female interactions (Sherlock woefully fails the Bechdel test).
Going forward, I think adding an interactive filter by season and a subfilter by episode would be useful if users wish to analyze interactions at more granular levels. I ultimately think working with a larger dataset would have allowed me to do more with my visualization (e.g., delineating communities).
Sarah Hatoum