This is a continuation/follow-up to my previous post, MEASURING BRAND SIMILARITY. Please check it out first!
In my last post, I scraped and analyzed data from around 2,000 Grailed users in order to determine a quantitative measure of similarity between different brands. With the dataset I compiled in that post, it was easy to evaluate the similarity of brands, even those that would be hard to qualitatively judge.
While my approach was great for pairs of brands, I wanted to see if I would be able to categorize the different users into different clusters. Mens’ fashion has branched out into numerous subgroups over the past few years, and I was interested in seeing what groups the data could be split up into. To accomplish this, I ran some standard clustering algorithms in order to determine groups of brands that tended to appear together. After experimenting with several approaches, I found that a hierarchical clustering approach with 5 clusters seemed to produce the best results. The top 5 brands from each cluster and my best effort to categorize them are as follows:
While my cluster labels aren’t perfect, I was surprised to see that many of them, especially the hypebeast category, did feel quite intuitive. However, many smaller subgroups were not represented in these clusters - techwear and Scandinavian minimalism, to name a few. Furthermore, these clusters had some overlap, reflecting how different styles are rarely completely distinct.
Overall, I think that there is a lot of potential in analyzing fashion from a more analytical perspective. In both this post and the previous post, this perspective was able to shed some light on something as abstract as personal style, and I'm looking forward to further exploring this area in the future. Hopefully you enjoyed the posts!