Abstract
In the early 2010s, while text-based social media platforms like Twitter were extensively researched, Instagram remained largely unexamined. This study presents the first in-depth analysis of Instagram content and user behavior. By integrating computer vision techniques with human qualitative coding, researchers identified eight primary photo categories and five distinct user archetypes. A key finding of the study is that a user's popularity, measured by follower count, is statistically independent of the specific content categories they post, suggesting that audience growth is driven by factors beyond simple content classification.
Introduction
While platforms like Twitter have been studied extensively, Instagram remained a "black box" for researchers during its early years of massive growth. The authors argue that because "a picture is worth a thousand words," Instagram offers unique insights into social and cultural life that text-based platforms cannot. This study tackles three fundamental questions: the categories of photos people post, the grouping of users into distinct "types" based on habits, and the potential correlation between content type and follower count.
Methodology
The researchers employed a hybrid approach bridging raw data and human meaning. Using the Instagram API, they identified 13,951 "regular active users" by crawling networks starting from celebrity seed accounts.
To categorize content, they first used the Scale Invariant Feature Transform (SIFT) algorithm to extract local features, followed by k-means clustering to generate 15 initial visual clusters. Human coders then refined these into eight meaningful categories. For user analysis, each user was represented as an 8-dimensional vector based on their posting proportions across these categories, which was then subjected to further k-means clustering to identify archetypes.
Content Analysis and User Archetypes
The study identified eight core photo categories, revealing that nearly half (46.6%) of all content consists of "Selfies" and "Friends." This confirms that Instagram serves primarily as a tool for self-promotion and social networking.
The 8 Photo Categories: 1. Selfies (24.2%) 2. Friends (22.4%) 3. Activities (Outdoor/Indoor events) 4. Food 5. Gadgets 6. Captioned Photos (Memes/Quotes) 7. Pets (<5%) 8. Fashion (<5%)
Based on these categories, the researchers identified five user archetypes:
- C1 (Generalists): Diverse content with a slight lean toward food.
- C2 (Caption-Seekers): Primarily post quotes and memes.
- C3 (Socialites): Focus on friends and activities.
- C4 (Selfie-Lovers): Almost exclusively self-portraits with the lowest content entropy.
- C5 (Balanced Socializers): An equal mix of selfies and group photos.
The Follower Paradox
Perhaps the most surprising finding is the "Follower Paradox." A two-tailed t-test ($p = 0.171$) failed to show any significant relationship between user clusters and follower distributions. Whether a user is a "selfie-lover" or a "generalist," their audience size appears to be driven by factors other than the specific category of photos they share.
Limitations and Future Work
While the manual coding process achieved high accuracy (Fleiss’ kappa of 0.75), the sample size was limited to 50 users (1,000 photos). Future research could scale this analysis using advanced automated image recognition and incorporate metadata like hashtags and comments for sentiment analysis to further decode the social dynamics of visual platforms.
Summary updated