medTint

Insights

How every #GameOfThrones episode has been discussed on Twitter

By
Wednesday, 1 June 2016

Every Sunday night, fans of HBO’s Game Of Thrones (@GameOfThrones) know that Twitter is the place to come to talk about the crazy plot twists, how they feel about the characters’ latest shocking actions, share their own memes, and more. Each #GameOfThrones season inspires millions of Tweets. The richness and variety of these conversations inspired us to explore what the audience is talking about in more depth. And here’s where that exploration landed us:

This post is unavailable
This post is unavailable.

Figure 1. Season 6 Episode 5 “The Door” (click to see interactive)

Explore the visualization in full here.

Read on to learn about how we built this visualization, and what we’ve learned about Game of Thrones - and its fans - from it. And for any of you who aren’t caught up… Warning!The rest of this post contain spoilers from Game of Thrones.

Background
We started by gathering Tweets with keywords related to Game of Thrones in the 24 hours after each episode first aired. Tweets contain many interesting aspects: distribution over time, sentiment, frequent terms, etc. We decided to focus on the characters and extract the characters that were mentioned in these Tweets to find patterns and insights. A quick look at character mentions over time (Figure 2) shows the rise and fall of each character. Seeing how often characters —no matter big or small— were mentioned, motivated us to push this analysis further and try to identify the most talked about plot in each episode.

This post is unavailable
This post is unavailable.

Figure 2. Distribution of character mentions by character and episode

See the plots of every episode from network of characters
Plots, in a way, are connections between characters. When two characters are mentioned together in the same Tweet, we infer that there is a connection. The characters then can be grouped together based on the connections using community detection algorithm. This effectively reveals the different storylines in each episode, and helps us spot the parts that caught the most attention.

We visualize these connection as a network (node-link diagram) in which each character is represented by a circle with its size proportional to how often the character was mentioned in the Tweets. Each connection uses thickness to indicate how often the two characters were mentioned together. A physics simulation is used to pull connected characters together while avoiding overlaps. Areas are drawn to surround characters that belong to the same cluster.

The example below (click the image to explore the interactive) is taken from Season 5 Episode 9 “The Dance of Dragons” where two major storyline are about Stannis Baratheon - on the left - and Daenerys Targaryen - at the bottom-right. Arya Stark and Ser Meryn Trant appear isolated in the top right, suggesting that they had a minor storyline together this episode.

This post is unavailable
This post is unavailable.

Figure 3. Season 5 Episode 9 “The Dance of Dragons” (click to see interactive)

One amazing thing to point out about this approach is that the algorithm knows nothing about the show. It does not know about relationships between characters or storyline of the show beforehand. Yet it still can reveal so much about the show entirely based on the Tweets. Of course, a perfect result would be too good to be true. One issue with our assumption is when two characters are mentioned together in the same Tweet, sometimes it does not mean a connection. Another possible flaw of this automatic [] approach is potential misclassification of characters with minor roles in the episode. For instance, Jon Snow, who has little to do in this episode, was pulled into Daenerys’ cluster. Aside from this minor mistake, the major storylines are still properly highlighted. There are many possibilities to address these issues and take this analysis further. With more human intervention, we can filter only characters or connections that appear in each episode, or use the storyline information to guide the algorithm. There are also many community detection algorithms to choose from. (Our visualization uses Louvain method, which is one of the most widely used methods.) This leaves so much potential for future work.

How emojis tell the story of each episode
Recently emojis have become a common and rich way to express opinions & emotions. Since this has become more common over time, we extracted emojis that appear with characters in the Tweets from later seasons of the show. These emojis often represent how the audience feels about the plot, but sometimes can illustrate what happened in the story.

In the example above, the strongest connection is between Stannis and Shireen, his daughter. The emojis suggest the audience’s anger towards Stannis, and sadness for Shireen. Well, this was the episode when Stannis sacrificed his daughter according to advice from Melisandre, who burned the poor girl alive (which also explains the fire emoji).

Ser Jorah Mormont, on the other hand, received the flexed biceps emoji for saving Daenerys by striking a spear straight at an assassin. Another interesting observation is when there was an act of cruelty, there will be association between that character and Joffrey Baratheon, who was a supervillain in the early seasons.

Our analysis also captures small moments in the show that spark huge interest for the viewers. In Season 6 Episode 4 “Book of the Stranger”, when Tormund Giantsbane saw Brienne of Tarth for the first time, the audience was speculating a spark of romantic interest between them.

This post is unavailable
This post is unavailable.

Figure 4. Season 6 Episode 4: “Book of the Stranger”

The examples above are only a small preview from the data we have collected, and a small fraction of what we could learn from conversations on Twitter. There are more data, more episodes and more visualizations to explore - we have visualized every episode of the show so far, and plan to update through this season. We invite you all to play with it yourself and find more stories! Share any of your favorite observations with us by Tweeting to us @TwitterData.

This post is unavailable
This post is unavailable.