Vogue has been hailed as the world’s most influential fashion magazine since its debut in 1892. Yale Librarians Peter Leonard and Lindsay King are now becoming Vogue experts — not because of their interests in next season’s fashion, but because of the extensive efforts to digitize the magazine’s entire record.
Since Vogue’s inaugural issue on Dec. 17, 1892, it has published more than 400,000 pages of fashion and lifestyle articles. Now, everything from feature stories to advertisements has been digitally labeled with critical information including photographers, image features, and product brands and names.
Leonard, King, and other Yale researchers and digital humanists are now using modern computer algorithms to dive through Vogue’s culturally, artistically, and textually rich electronic record in order to ask new and meaningful questions about the evolution of fashion and culture in the media.
Starting with words
“How well do you know fashion terms of the 1920s?” asked Leonard as he compared the use of words like “frock” and “dress” over time using a system called “Bookworm.” Similar to Google Ngrams, Bookworm generates graphs to illustrate the frequency of single words in Vogue over time, and under different editors. This enables Leonard and King to track not only how the linguistics of fashion have evolved, but also how other cultural norms and ideas have morphed in popular media.
A comparison of the usage of “women” and “girls” shows that “girls” cyclically falls out of favor — today, “girls” is most often used in reference to models, notes King. While this verbal change may be indicative of a paradigm shift in the female image, she said, it offers a starting point for more in-depth analysis.
“We want to let the data organize itself,” explained Leonard, noting that he aims to create tools that will allow users to ask better questions instead of searching for a digital needle in a haystack.
Bringing words into context
Collecting trends for individual words can leave you with a Jackson Pollock-like map of evolving cultural history. To start connecting these dots in a comprehensive way, Leonard and King turned to a statistical tool known as “topic modeling,” which evaluates the frequency of words and categorizes those that are commonly associated with one another to create a topic. Using this technique allowed Leonard and King to find groups of words like “doctor,” “health,” and “women," revealing that women’s health once held a prominent place in Vogue.
“I would have never known to search for these topics, but the topic modeling algorithm tells you what to look for,” explains King. “It’s like magic.”
A picture is worth 1,000 words
Black-and-white text, however, is not what grabs your eye as you pass Vogue on the newsstand. Rather it is the bright colors and striking faces on the cover that draw your attention. Focusing on text alone would neglect the artistic richness of the magazine, notes Leonard, so he used color as his starting point for another type of analysis: color.
By the 1900s, color images had replaced their black-and-white counterparts. Focusing on the covers between 1911 and 1951, Leonard digitally analyzed the covers to identify the five most prevalent colors on each cover image. With this information, he created an iPad tool that allows users to see how the use of these colors changed over time and to see which cover images use specific colors.
Even though color images were available, Leonard found that during the first half of the 20th century, covers primarily featured more muted tones such as grey, slate, and silver. More vibrant colors such as hot pink and midnight blue were prominently used only a handful of times.
The images laced throughout the 400,000 pages of Vogue amassed to five terabytes of data and took nine days to transfer onto a desktop computer.
Analyzing the remainder of the images proved to be a more substantial technical challenge. To help analyze the terabytes of graphics data, Leonard and King turned to Holly Rushmeier, a computer science professor at Yale with an interest in computer graphics and computational tools for cultural heritage.
Rushmeier, in turn, enlisted the help of David Li ‘14 and Christiana Wong ‘14, two students working on their senior projects in the computer science department, to explore the graphics.
Their first challenge was simply getting the data. The images laced throughout the 400,000 pages of Vogue amassed to five terabytes of data and took nine days to transfer onto a desktop computer.
Li organized the image data — a more challenging task than it may seem. Unlike the Vogue covers, which took up an entire page, many of the other images and advertisements did not. Li drew upon his computer science education to identify and extract individual images and the important content tags associated with each image in XML code.
These XML code tags provide critical information, including the year of the image, the main content of the image, and classification as an advertisement or editorial content. Having text to describe each image allowed Li to create a search engine specifically for Vogue images that allows users to easily navigate the wealth of visual data.
As Rushmeier discovered, shifts in cultural opinions can happen at any time. She demonstrated the capabilities of Li’s search engine using the term “fur” applied to images published in the 1970s. Instead of exotic furs or runway images of haute couture that you might expect to find in Vogue, the images that resulted from the search were actually of furniture. Tweaking the time frame of the search to the early 1900s, however, returned numerous images of full fur coats and other fur products that you might expect.
Scanning through the images, a cigarette advertisement catches Rushmeier’s eye, and she is off to the next image search for “Marlboro.”
“As you can see, you start getting lost in the data really quickly,” said Rushmeier.
The women of Vogue
Of all the images in Li’s search engine, one type of image in particular piqued Wong’s interest: women’s faces. Although Wong earned her degree in applied mathematics, she has also become interested in the portrayal of women in the media.
The plethora of photographs of female faces in Vogue offered Wong the opportunity to hone her computer science skills and contribute to the growing investigation into “faceism” and its role in depicting gender in the media.
While we may not recognize it, research has shown that we often assume characteristics such as intelligence, ambition, and attractiveness simply based on how a face is displayed in an image. To quantitate this idea, those studying the portrayal of women have created a “faceism index” — a measure of how faces are presented and how that connects with our subconscious assumptions.
“Men are often presented with more prominent features, whereas women are shown with smaller faces,” explained Wong. The larger, generally male, faceism indexes tend to evoke more positive responses from viewers, she adds.
“The link between larger faces and the perception of more confidence really intrigued me, and I wanted to use the information from Vogue to bridge a gap in our understanding. I wanted to see how the prominence of women’s faces changed over time,” said Wong.
Building off of Li’s image database, Wong used facial recognition software to scan nearly 1 million images. The result was 287,970 images of faces. She then used her program to detect the eyes, nose, and mouth in each pictured face. With the coordinates of the predominant facial features on the page, Wong rapidly calculated the faceism index for each of the nearly 300,000 images.
Don’t judge a magazine by its cover
What Wong found may not surprise avid Vogue readers — Vogue is on the leading edge of a continually changing female image.
While the facial perspective in images was highly dependent on the editor-in-chief at the time, Wong found that the 1960s marked the beginning of a steady rise in female facial prominence on Vogue covers, or in other terms, a higher faceism index. The concurrent second wave of feminism during the 1960s may have been the impetus for Vogue’s stylistic changes, but Wong believes that more research is needed before that can be concluded.
Wong compared the Vogue cover images to advertisements in the magazine with a particular focus on female cosmetics. As Vogue’s covers more prominently featured female faces, companies like Clinique, Estee Lauder, and Almay strayed away from that trend. Over time, the female images within the body of Vogue remained constant, even as the covers evolved.
What’s next for Yale’s exploration of Vogue?
In addition to her research with faces, Wong is interested in gender and body position in popular media. A subject’s pose also speaks to a reader’s subconscious — lying horizontally evokes associations with submissive behavior, for instance, while upright angles display dominance. Historically, women tend to be pictured in more submissive roles than men.
While Wong and Li graduated from Yale in May, the field of digital humanities and the exploration of the Vogue record are only just beginning.
“These digital records are an entry point,” explains King. “It is a totally new field and offers new directions for scholars with potentially huge implications.”
“We need someone to ask the right questions, then we can come up with the tools for the answer,” added Rushmeieir.
With the start of the new academic year, Rushmeier is seeking new and curious students interested in using computer science to explore these digital archives and to help find out what is truly in Vogue.
For more information on the results of these studies, Leonard and King have compiled much of the work done at Yale in Robots Reading Vogue.