Mining Visual Web Towards Real-time Fashion Trend Forecasting

Not so long ago, perhaps two decades back, paper booklets such as fashion magazines, celebrity newsletters, or style entertainment booklets defined the content that readers consumed and desired. With the rise of the Internet, high speed 4G networks, and smart-phones, most of these paper booklets have disappeared and are taken over by web publishers and social media. Indian websites and social media publish more than 500,000 fashion related photos and videos every single month. This is equivalent to 5,000 fashion booklets per month. This deluge of visual content has flummoxed both fashion consumers and fashion producers. 

In the previous post, we looked at making millions of fashion looks shoppable so that fashion consumers can delightfully shop fashion of their choice anywhere on the web. We learned about our AI-driven fashion intelligence toolbox that can automatically parse large number of photos and videos. How can we apply the toolbox on 5,000 booklets worth of fashion in the wild to extract meaningful information for fashion producers? Can we help fashion producers to design exactly those patterns/colours/styles that consumers seek, at the right time and in the right market?

It is common sense to see that if the fashion designers design styles that are completely opposite of the fashion trends and consumer sentiments, then it can have significant effects on retail bottom-lines and fashion supply chain. For instance, overall sales of apparels and accessories will decline, some fashion items will go out of stock far sooner than expected while others will go unsold; simply put, consumers will have to turn away from stores due to lack of fashion that they desire.

In the past, fashion designers relied on fashion shows and fashion events to seek inspirations for next generation fashion trends. Seasonal colours and fabrics apart; for instance, light colours in summer versus premium clothing in winter, designers also sought inspirations from movies and TV shows. The underlying hypothesis was that consumers too, consciously or sub-consciously, are attracted to the high-end fashion, and the next generation clothing and accessories can be designed according to expected sentiments of fashion consumers.

Time, however, has changed quite dramatically in last one decade. The social media in the form of Pinterest, Facebook, Instagram has become integral part of consumer lives. Fashion consumers today have instant access to wealth of fashion inspirations in the form on influencer websites and celebrity social accounts. At Infilect, we wished to understand the implications of this shift and the effects it can have on the entire supply-chain of fashion retail. Utilising the unique data from fashion discovery sought by users on Huew, and by applying our fashion AI toolbox on the wealth of social media content, we did a careful study of fashion trends from 1st Jan, 2018 to 24th Feb, 2018. Here are some interesting results.

Q1: What were the trending photos on social media, and what kind of impact did they have on consumer purchases?

Each of these photos received more than 1M likes, and potentially many million views. It is quite conceivable that colours, patterns, styles expressed in these photos had tremendous impact on consumer mindset. However, we wished to confirm this hypothesis through a statistical evidence. Using our affiliate relationships with large e-commerce players (as part of Huew) and tracking consumer purchase patterns, we indeed witnessed spikes in “mellow yellow” colour from Deepika’s dress above, and white checked shirt for women from Priya Prakash’s style above. There was at least 145% lift in the purchases for these colours in the above two months as compared to the last quarter of 2017.

Q2. What was the distribution of fashion clothing and accessories as seen the most by social media users?

As seen by the above charts, more than 50% of women wear on social media is dominated by just three apparel types: dresses, tops, and jeans. As compared to the first quarter of 2017, there is 32% lift in the amount of distribution that these three apparel types occupy. This is a clear signal for designers and manufacturers to focus their designs and styles around these three apparel types. We did similar analysis on fine-level fashion attributes such as colours, patterns, sleeves, and neck types. Reach out to to get free access to a detailed report.

Q3. Do movie releases have influence on fashion purchases, and can we reliably predict the demand of movie styles in advance?

The much awaited Bollywood flick, Padmaavat, was released on 25th Jan, 2018. The film gained public attention even before the release due to the controversy surrounding its historical accuracy. However, the movie had dramatic impact on the demand for “red lehengas” from “Ghoomar” song. Tracking the number of impressions of the below photo, and using data from our partner websites, we uncovered a whopping 350% lift in the purchases of red lehengas in the month of Feb as compare to Jan. Thus, if we parse movie teasers, movie trailers, and social media wallpapers before the movie releases, we can help different stakeholders in the fashion supply-chain to anticipate the consumer demand over 3 months to 6 months time period.


If you are a fashion retailer, these results show that doing accurate trend analysis and forecasting can truly transform your bottom-lines. If you are a fashion designer or manufacturer, our forecasting tool can predict short-term (3 months) to long-term (1 year) forecasts on the colours/patterns/styles you should be choosing in order to meet the distribution demand. As more and more fashion content getd available on social media, it is quite interesting to see its impact on the demand and purchases of women and men fashion. Fret not, you can stay relevant with the current and future fashion trends. Write to us now on to claim your free copy of a detailed trends report.



This entry has 0 replies

Comments open

Leave a reply

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>