Alto Data Analytics went back in time to research to what extent Instagram data from the past could be used to predict how future fashion trends evolved.

Using the predictive analytics capacities of our big data software, Alto Analyzer, and our team of data scientists, we were able to successfully spot key emerging trends by pin-pointing patterns in the most-shared, liked or commented content by Instagram users. We then researched if those patterns eventually would match what was published later as trends of the season by major brands and retailers. Interestingly, the historic data and what was published matched.

Instagram’s user activity proved to be useful data for predicting emerging trends thus, revealing how retail brands might use predictive analytics to anticipate trends in fashion.

Network Analysis of User Interactions

Our approach was to retrieve all data related to “street style” and other similar concepts such as “outfit of the day” in order to gather an initial data set of images, comments, and hashtags and map the interactions. We analyzed Instagram from February 6th to 23rd, 2015 with no restrictions in geography or language.

Hashtags and keywords used to formulate visual data
“Outfit of the Day” brought along a number of adjacent concepts organically used by Instagramers such as “StreetStyle”, “StreetSnap” and “Today’s look”
Subset of images from data set
Subset of images from data set

Making Sense of Visual Data: Algorithms and Strategies to Discover Trends

In order to start identifying trends, we did not initially focus on the images. Instead, we built a network out of the interactions users created when posting images. We filtered out users that had no interactions connected to them – no likes, comments or replies – as it was unlikely that these users and the images they shared could create trends. We then we applied Alto’s algorithms to uncover communities and patterns with users whose content had at least one interaction.

Process of analysis
Process of analysis

Community Analysis: Surfacing the Shape of Discussion

Once we laid out all the interactions between users, we first noticed how the resulting networks were hyper-connected with a high density of interactions. Likes, comments and replies were abundant showing user engagement was strong:

Hyper-connected network with high level of user engagement
Hyper-connected network with high level of user engagement

Through a deeper analysis, our data algorithms clustered the different communities and we were able to identify 154 different distinctive communities with diverse interaction levels. By filtering the most relevant communities, we determined the two main nucleuses generating interaction. They are found in the centre and composed of several different communities:

Main nucleuses in centre and blue (top right) generating interaction
154 different community clusters with diverse interaction levels

Swipe to view network

Communities and Evolving Trends

Our data scientists then isolated the top 5 communities and were able to start identifying patterns based on visual similarities within the content of the posts.

  • 1 – Street Style Amateurs
  • 2 – White Lovers
  • 3 – Models and Celebrities
  • 4 – Urban Style
  • 5 – Alternative Style
Communities based on similarities within photos posted - Street Style AmateursCommunities based on similarities within photos posted - White LoversCommunities based on similarities within photos posted - Models and CelebritiesCommunities based on similarities within photos posted - Urban StyleCommunities based on similarities within photos posted - Alternative StyleCommunities based on similarities within photos posted
Communities were formed based on core similarities within the images posted

Mouseover titles to view communities

Street Style Amateurs

The largest community was formed by amateur young women posting pictures wearing casual street style. The most popular were black trousers, jeans, jackets, boots and flat shoes. Scarves and handbags were prominent accessories. Black, white and earth colours were basic, although electric colours such as red, fuchsia, Klein blue and yellow were also popular.

White Lovers

Images from this community were more professional and commercial. Many users were qualified bloggers, online shops, designers or social media curators. Their pictures featured female models with a low rate of selfies. The clothes represent casual, sport and minimalist styles based on high-fashion pret-a-porter. White garments and all white outfits were dominant, while black, grey, and blue were also significant. Earth tones, yellow, red and fuchsia were stood out and plain patterns were common.

Models and Celebrities

The third community represents high-fashion haute couture and pret-a-porter. The images mainly portray female models and celebrities. The most popular garments were oversized coats and jackets, dresses and gowns. The mixture of textures such as furs, leathers and plain patterns were repetitive. This community used a darker range of colours, of which the most present were black, burgundy, earth tones. They also used electric colours.

Urban Style

Clothes in this community mainly represented street style based on high-fashion pret-a-porter, although there was a deeper focus on minimalist, oversized cuts and patterns, usually plain. The most popular colors amongst this community were white, black, grey, blue, red, fuchsia and yellow. Images feature many high-profile bloggers.

Alternative Style

Trends in this community were the most diverse and divergent. London Fashion Week style lovers such as photographers, male and female models, and bloggers expressed new extravagant styles and outfits. Multiculturalism, parody and pastiche was the soul of this trend. It included both gentleman and rakish cuts for men, sport styles, retro patterns, and even some punk and hiphop influences. They wore high contrast colours joining black with strident yellows, oranges, reds, blues or pinks.

How Did Trends Flow?

Once we identified the communities, we wanted to understand their communication flow and how it influenced the development of trends. We gathered the most-influential photos from each community and recreated a set representative of the whole conversation:

Sample of most-influential photos within 5 top communities
Sample of most-influential photos within 5 top communities

To understand the content of the pictures, we tagged them with 7 categories based on their composition:

Categorization of photos based on composition
Categorization of photos based on composition

With this categorization we were able to weigh the presence of each category and understand the importance of subjects and framing within photos. We were also able to connect those categories to the type of users interacting with them, which helped us understand user behaviour:

Relation of photo frame to user
Note how influencers tend to communicate with whole body pictures of one person

Mouseover image to view connections

Successful Prediction of Trends

Now it was time to link the trends we identified to what was actually published. Once the connections were made, it was easy to validate the results of our analysis:

Street Style Amateurs

Outfits of black trousers, jeans, jackets, boots and flat shoes. Black, white and earth colours as base, with electric colours such as red, fuchsia, Klein blue and yellow.

White Lovers

Casual, sport and minimalist styles based on high-fashion pret-a- porter. White garments and all white outfits. Black, grey, blue and earth tones.

Models and Celebrities

Oversized coats and jackets, dresses and gowns. The mixture of textures such as furs, leathers. Darker range of colours such as black, burgundy, earth tones.

Urban Style

Minimalist and oversize cuts and patterns. Most popular colours: white, black and grey, while Klein blue, red, fuchsia and yellow were also chosen as more outstanding colors.

Alternative Style

Multiculturalism, parody and pastiche. Rakish cuts for men, sport styles, retro patterns, punk and hip-hop influences. High contrast colours: black with strident yellows, oranges, reds, blues or pinks.

Key Learnings: How fashion retail brands can use predictive analytics to anticipate evolving trends

Our analysis revealed a clear ability to be able to match evolving trends spotted on Instagram to what major fashion retail brands promoted months later in their Spring and Summer collections. Through the use of predictive analytics, Alto was able to forecast the season’s trending textures, colours, styles and outfits. Here are our key insights divided into strategy development and trend development.

Strategy

Instagram ecosystem includes abundance of likes and hashtags, allowing brands to have genuine insights due to the association of concepts and visuals.Brands can create communities and attract users through the use of new organic hashtags that users naturally associated with the trends.

Trends

Core communities’ analysis reveals general trends by showing similar and consistent visual expression.Community analysis shows top colours to reveal trends.Categorizing posts visual expression reveals most common clothing types.Identifying influencers within communities can point to upcoming style trends.In-depth analysis of style trends has ability to reveal textile patterns and trends.

Want to learn more? Please contact media@alto-analytics.com to view the entire analysis and discover how to use predictive analytics to make data-driven decisions. To learn more about Alto Data Analytics’s articles and ongoing projects, please sign up to our newsletter below.

Article written by Clarissa Watson, Head of Marketing