Annie Lee



FLUIDNOTES: GENDER AND FRAGRANCE

Exploring Gender and Fragrance Over Time


Live Site    Github   Summary



ROLEResearch, Data, Design & Development DURATION5 Months (Jan-May 2025) TOOLSD3.js, Python, Github Copilot, Scrollama, Figma, HTML/CSS, ChatGPT VisionSKILLSWeb Scraping, Data Cleaning, Data Analysis, Scrollytelling, Interactive Data Visualization, Machine Learning PROJECT OVERVIEW
What smells “feminine” or “masculine” isn’t chemistry — it’s about culture, time, and storytelling. They change — just like our ideas of gender. My thesis, Fluidnotes, maps over 25,000 perfumes and 100 years of gender history to explore how fragrance culture reflects - and sometimes challenges - our evolving understanding of gender.

Using machine learning and computer vision tools, I was able to analyze 250+ vintage advertisements, map fragrance notes into categories, and connect visuals to scent compositions. While the dataset does not represent the entirety of the global perfume market, it covers a broad span to reflect the Western industry’s dominant trends.

4 MIN PROJECT WALKTHROUGH
THE VISUALIZATION
This project is best experienced by exploring the live site on desktopFluidnotes is designed for exploration. Rather than telling viewers what to find, it provides an interactive overview that encourages curiosity, comparison, and discovery across more than 100 years of fragrance history.
 


(1) Summary and NavigationYear 1900 to 2025 are grouped into 4 main eras by common gender approaches: Idealized Feminity, Binary Boom, Fluidity Emerges, and Age of Expression.(2) Fragrance Release by Gender and YearThe doughnut chart gives an overview of the ratio of women:men:unisex perfumes released in each year.(3) Key Gender MilestonesA compiled list of key gender milestones is layered to provide context - sourced from multiple sources including historical archives, academic databases, and cultural timelines.
(4) Top 10 Notes by Gender and YearAs you scroll through each year, the top 10 notes reorder. You can filter by gender, note type, or search and click into any note to discover its rise and fall in each gender category. (5) Popularity Trends of each noteDetailed. expanded view of the rise and fall of each note in the different gender categories. 




(6) Ad Subject Treemap The three interactive treemaps are also filters for the ads gallery. They show subjects of each ad — women, men, both, or no one — and who the ad targets. I was curious not just who shows up, but how they’re portrayed. Clicking into any cell reveals their expressed emotions color coded by gender.
(7) Fragrance DetailsFragrance details page displays the name, brand, year, target gender, ad copy, and scent profile grouped by top, middle, and end notes.
(8) Fragrance Ad GalleryI collected over 250 vintage fragrance ads and they can be explored by subject gender, expressions, and year of release. 

BEHIND THE SCENES
DATA COLLECTIONData was collected through web scraping and cleaned for consistency across fragrance, advertising, and historical sources. Machine learning and computer vision, including ChatGPT Vision, were used to explore patterns in vintage perfume advertising and prepare a research-ready dataset.
Data Description Access
Fragrantica Dataset Over 24,000 entries with notes, ratings, accords, perfumers, etc. Link
Parfumo Year of Release Counts of discontinued/current perfumes; “popular fragrance of the era”. Link
Fragrantica Notes Scraped notes + images for mapping. Link
Ad*Access (Duke) 7,000+ US/Canada print ads (1911–1955), beauty & hygiene. Link
All Ads A collection of vintage archives and recent social media posts. Link
Advertising Archives UK The advertisements are arranged by decade and can be searched by product. Many American advertisements are included. Link
GPT Vision Automate visual analysis of vintage perfume ads using ChatGPT Vision - test.py file in github Link
Key Gender Milestones Compiled list of key gender milestones from 1900 to 2025 by aggregating information from multiple web sources, including historical archives, academic databases, and cultural timelines. Link
    
CHATGPT VISIONI used ChatGPT Vision and Python to automate the analysis of hundreds of vintage perfume advertisements. By prompting the model to identify visual attributes such as perceived gender, facial expressions, poses, mood, and interactions, I generated a structured dataset that enabled large-scale analysis of how gender representation in perfume advertising evolved over time.




Fluidnotes