AI has the capability to identify city-specific fashion trends. This tech could be deployed with a ranking model to give fashion brands insight on how to best serve their customers.
Artificial intelligence (AI) is making a big debut in the fashion industry. While holding many use cases for the future of fashion, AI’s power in trend prediction is one that will shape the future of the industry.
Until the rise of social media, trends were typically born in 1 of 2 ways: top-down or bottom-up. The top-down cycle begins when a designer chooses a color, fabric, or style to design a collection after. Upon its popularity, a trend will be created–making its way through other designers, filtering itself through department stores, and finally being diluted through adoption of the general public.
Trends that are a product of the bottom-up cycle occur when designers take inspiration from the general public. For example, blue jeans, once exclusive to the working man, can now be seen in the collections of nearly every major fashion house.
Social media has altered these cycles, creating a medium for new trends to start, and allowing for more room for niche trends to begin. Trends in the digital age have cycled much quicker than those in previous decades–making them harder for individuals and retailers to quickly adopt.
The adoption of social media to display one's fashion choices has created a huge labeled database; with individuals labeling their photos with hashtags, geotags, and timestamps. Deploying AI and machine learning (ML) algorithms to these databases has made identifying trends, on macro and micro labels, easier than ever before.
Millions of photos are uploaded to the internet everyday. These photos communicate much more than just what an individual is wearing. Factors like the weather, subcultures, activities, etc. can all be deciphered through photos shared online. Using AI algorithms to analyze these photos can be a powerful tool to predict upcoming trends.
Taking in mind the impact of location, weather, and societal events, GeoStyle aims to identify fashion trends across 37 cities around the world and decipher what that communicates. The abstract to their research states “our method automatically segments the map into neighborhoods with a similar fashion sense. Our approach further allows discovering insights about a city.”
The process begins using Instagram and Flickr to generate a database containing 7 million images of people from around the world. Both Instagram and Flickr images identify when and where a photo was posted. GeoStyle uses convolutional neural networks (CNN) from ImageNet and GoogLetNet to train their algorithm. CNNs learn an image by assigning weights and biases to different aspects of the photo. This allows the algorithm to be able to differentiate certain photos from one another. GeoStyle’s CNNs were programmed to decipher both the clothes in the photo and city-identifying background images to demonstrate how trends can be exclusive to region and neighborhood.
Images from similar regions and of similar style were then clustered together, giving researchers an organized database of popular styles emerging in cities around the world. Researchers of the project were able to identify neighborhood-specific trends and draw educated inferences on relationships between styles of dress across certain cities.
GeoStyle’s findings give us a good idea of what trends are emerging and what style of dress is appropriate for certain regions. One of the findings highlighted in the project was that along with food and architecture, neighborhood specific trends were a powerful predictor of the culture of the neighborhood (ethnicity, socio-economic status, etc.).
This kind of technology insight can allow businesses of any size, not just the ones who can afford an expensive research department, to understand what trends are popular for a specific time and region. This gives them the opportunity to compete with larger companies who have historically been ahead of the trend cycle. In practice, data generated from this algorithm could be applied to another layer of AI to further transform fashion businesses.
The future of fashion, or at least a large percentage of it, exists online. In 2020, E-commerce made up 30% of retail fashion sales in the US. That number continues to grow both domestically and worldwide. As more brands begin to establish an online presence, a strong push towards AI and ML deployment is being made to customize the user experience.
GeoStyle’s algorithm could provide brands with data which they could build a ranking model off of. Using varying types of data, ranking models aim to customize the user experience.
Generating data is essential for building an effective ranking model. Many brands struggle with generating a large amount of individual user data as customers may not be making enough frequent purchases. Making it even more essential to recommend relevant products when a customer enters the site.
Tools like GeoStyle could strategically be used to first identify neighborhood and regional specific trends emerging. These findings can then be applied to a ranking model, giving more weight to clothes that fit within what’s in-style for a user of a particular area.
This benefits the buyer and seller, as the buyer is able to quickly find the clothes they are most likely to look for, and the seller is seen as a reputable source for finding what is in style.
Like any technology, GeoStyle’s algorithm doesn’t come without its concerns. Backed by Amazon and Meta (previously Facebook), algorithms like this raise concerns about how technology is commodifying personal data. It is unknown whether or not people who posted images used in this dataset consented to having them used in this type of algorithm. This technology has concerns to also further exacerbate the rate and specificity at which targeted ads are displayed.
The findings of this study have shown to be a powerful step in using AI to predict fashion trends. The trend prediction process has carried an extensive process of research and development, either doing field research or manually scrolling through social media to identify emerging trends. GeoStyle’s technology is a step to automating this process, allowing those in the industry to quickly identify emerging trends.
AI technology holds great potential in being widely used in the fashion industry. It’s power in regional and neighborhood specific trend prediction can help businesses fine tune their business production strategy.