My time using both Voyant Tools and Google Gemini turned out to be more of a playful exploration of The Count of Monté Cristo than an in-depth analysis into the text. The tool that initially caught my eye was Voyant Tools ‘Bubbles’ where the frequency of a given word (minus common words such as ‘the’, ‘and’, etc.) is displayed via circles of varying sizes. This program runs through the entire text word by word, and it thereby possible to observe how the frequency of a word’s use changes over the course of a work. Granted, there are other programs in Voyant Tools, such as ‘Corpus Terms,’ which display the changes in frequency of certain terms much better. However, ‘Bubbles’ also allows one to visualize the overall frequency of terms in a more appealing way which is easy to understand.
Another fun visualization tool is the ‘Cirrus,’ which makes it easy to analyze the frequency of certain terms through scale.
Given the monumental scale of The Count of Monté Cristo, Voyant Tools may have had a difficult time rendering networks; my attempt using the ‘TextualArc’ tool went quite slowly as the program was trying to show the interconnection between all of the words at the same time as it traced a arc of connection through words in document order. It is also quite tricky to parse which words are which in a network that is so massive. Thus I argue that though certain tools are good for visualizing networks in large texts, the way one scales such visuals is quite important if they wish to receive any workable information.
My time using Google Gemini went over much more smoothly than in using Voyant Tools, but the visualizations came out with a lower quality. I followed all of Anastasia Salter’s prompts in the order in which they were presented, and this led to a more tangible flow of understanding. With each prompt given, Gemini gave me python code which would allow me to visualize the data outside the bounds of the desktop. At the same time, a few of Salter’s prompts — such as “Many of these are common words, can you apply a basic stopwords to remove things like I, the, do, is, our, etc?” — seemed unnecessary as none of such words appears in the most frequent visualizations. Gemini knew what I wanted and what needed to be avoided without me correcting it — and that is a bit intimidating, seeing it come this far. This said, the visualizations were not only a bit basic but also difficult to parse.
When I asked Gemini if it could ‘visualize the network of character relationships in the text’, it gave me this network seen above. Having read The Count of Monté Cristo, not only can I say that this is not an entire network of all of the characters in the first eighteen chapters (only the major ones), the connection between Dantés, Villefort, and Faria makes no sense. Gemini’s explanation of this as a connection between the three of them, though technically true, does not account for the strength of Dantés and Faria’s connection between one another — almost as though Villefort is an intermediary between the two, which he is not.
While we may be concerned by the speed and accuracy of Ai’s capabilities in some respects, it seems that at least Gemini (or the version of Gemini we have access to as students) needs some work before its data visualization becomes anything of wonder or on par with Voyant Tools.
Hi Max, this is interesting. I don’t know much about the book, but I think the Gemini-produced network analysis is very interesting. I had similar results with Gemini, where it tried to give me what I wanted but did so in a similar text-based visualization with inaccurate information. For me, this was when I tried to make a word cloud, and it actually told me the word frequencies were simply estimations and not actually based on the .txt file I provided, and I wonder if it’s doing the same with your character network. It may just use what information in its database it has about The Count of Monte Cristo, which isn’t necessarily as accurate as the text itself. Because of this, I really agree with your analysis that AI, or at least, Gemini needs improvement before it’s anywhere near on par with Voyant Tools.