For data analysis, we carried out a heavily exploratory process as we had a large amount of data to play with: 17,477 tweets, or 3.88 megabytes. A variety of analysis methods and programming languages were used to extract information from this raw data source.
Scope of Analysis
Overall, we focused on descriptive and diagnostic analytics, inspecting our data to describe the happenings at Glossier, based on raw Tweets, and interpret why these things are happening, based on results like numbers, patterns, and visualizations. Our analysis focused primarily on text rather than on images, though we did conduct a general examination of images posted on the Glossier Instagram account, images associated with most retweeted and most favorited tweets (if present), and images associated with popular Glossier hashtags. Our emphasis on text stems from two reasons: one, Twitter is based on expression and communication through messages rather than images; two, we lack the time and technical tools to perform reliable image analysis.
The scope for our analyses included two of Glossier's official social media accounts of interest to this project, which are Instagram and Twitter. As of November 25th, 2017, Glossier's Instagram account consisted of 845K followers, with average likes of 25K to 27K per post, and a total of about 2320 posts, while the Twitter account consisted of 47.8K followers, with average favorites of about 60 per post, and with a total of 8721 tweets. These numbers alone tell a story of and support the knowledge that Instagram is Glossier's main social media outlet and that Glossier's main marketing tactic is through highly created and curated images, as opposed to Twitter and it is chiefly text-based posts (Forbes). The number of followers and average engagement through likes/favorites endorse this point as well, as the Instagram account had about 790K more followers and about 25K more likes per post. Interestingly, there were roughly 6,000 more tweets than there were Instagram posts up-to-date, despite more content on the latter platform, further suggesting that there is more content engagement through Instagram than through Twitter.
To understand the number of posts versus engagement issue, we performed extensive scroll-throughs of the official Glossier pages on both accounts. We opted for exploratory analysis in the form of scrolling due to our lack of access to the stringent Instagram API, and thus inability to mine Instagram data. Though lacking in technical and mathematical accuracy, we felt it was still necessary and imperative to inspect the two platforms in order to understand our limitations and possible reasonings for our results.
Glossier's Instagram account comprise exclusively of photographs and snapshots, with an overall hue of pink, red, pastel, and white. The pictures are simple and aesthetic, and consist of models, products, snapshots of customer posts and images, cute animals, and random photos featuring flirty colors as well as Glossier's signature color: glossier pink. This account is very focused on customer engagement through a distraction-free feed focused on beaming Glossier-selected visual content at users; the thematic images achieve the goal of grabbing user attention and keeping that attention (Jackson).
On the other hand, the Twitter account comprises of tweets and retweets of customer tweets, blog posts, videos, and images, just to name a few, often combined with links to external sites (Jackson). Examples of common tweets include customers raving or complaining about products, as well as images of makeup products joined with purchase links to the Glossier website and blog, Into the Gloss. Compared to Instagram, this account is a lot noisier and a lot more text-focused, thus a bit harder to gain and keep user attention. Based on many articles, including ones on The Business Journals (bizjournals) and National Public Radio (NPR), visual content marketing proves to be much more effective than other forms of marketing, as they are easier to take in and digest, and oftentimes more real and more memorable (Patti 2013; Pyle 2017).
Though we were unable to access content on Glossier's main social media platform, we had identified the benefit of mining and analyzing Tweets: understand direct text-based communication between Glossier and its customers. Through Twitter, we were able to perform considerable text analysis, which some may argue is less subjective than image analysis.
Analysis Methods
Our methods included performing sentiment analysis using Python's Natural Language Toolkit (NLTK) package, in order to determine the overall emotional reaction towards Glossier and its products and culture. We also used Python to perform basic tweet analysis (ex. number of tweets, average number of favorites/retweets per post, max favorited tweet and associated user, unique key values) and manual text analysis (ex. textual contents that are most often shared and by whom, frequency of certain keywords like product names or lifestyle-related descriptors) using both JSON and CSV files. Python was our main programming language of choice for data analytics as our team members were all familiar with it, and both JSON and CSV files were easily manipulable in its environment. R and Javascript were also languages we used for some analysis, but mainly for constructing dynamic, web-based visualizations. The visualizations were completed using D3.js and Tableau, and served as both additional methods of analysis (in a visual rather than numerical and textual perspective) and reporting of results.