Source-linked AI summary
Online Popularity and Topical Interests through the Lens of Instagram
Emilio Ferrara, Roberto Interdonato, Andrea Tagarelli
TL;DR
The paper investigates how Instagram’s mixed social, tagging, and media-sharing structure reflects human behavior, focusing on network organization, content dynamics, and topical interests. Using a sampled Instagram dataset and analyses of interactions, tags, and user groupings, it finds topical communities, heterogeneous production and consumption, limited tagging vocabularies, and a mild relation between topical variety and popularity, while noting important sampling and observability limits.
Problem
The paper addresses limited knowledge about how Instagram’s social interactions, content activity, tagging behavior, and topical interests are related at scale.
Method
The study analyzes an Instagram dataset using network analysis, content-production and consumption signals, tag-entropy measures, and clustering of users represented by tag frequencies.
Results
The study finds topical community structure, heterogeneous content production and consumption, limited user vocabularies, and a mild popularity relationship in which broader interests tend to yield higher popularity while popular users show extreme topical breadth or specialization.
Takeaways & Limitations
Instagram’s network, content, and tagging signals provide a combined basis for studying how topical interests, user interactions, and popularity are interrelated.
Takeaways & Limitations
The findings depend on a sample centered on users and media engaged in a public, competition-driven Instagram initiative and do not cover the full ecosystem or latent interactions.
Abstract
from arXiv · showhide
Online socio-technical systems can be studied as proxy of the real world to investigate human behavior and social interactions at scale. Here we focus on Instagram, a media-sharing online platform whose popularity has been rising up to gathering hundred millions users. Instagram exhibits a mixture of features including social structure, social tagging and media sharing. The network of social interactions among users models various dynamics including follower/followee relations and users' communication by means of posts/comments. Users can upload and tag media such as photos and pictures, and they can "like" and comment each piece of information on the platform. In this work we investigate three major aspects on our Instagram dataset: (i) the structural characteristics of its network of heterogeneous interactions, to unveil the emergence of self organization and topically-induced community structure; (ii) the dynamics of content production and consumption, to understand how global trends and popular users emerge; (iii) the behavior of users labeling media with tags, to determine how they devote their attention and to explore the variety of their topical interests. Our analysis provides clues to understand human behavior dynamics on socio-technical systems, specifically users and content popularity, the mechanisms of users' interactions in online environments and how collective trends emerge from individuals' topical interests.
1. INTRODUCTION
The paper studies Instagram as a socio-technical system combining social interactions, tagging, and media sharing to investigate network structure, content dynamics, and topical interests. It frames five questions about network features, content production and consumption, tagging diversity, topical clustering, and the relationship between popularity and topical interests.
- Scope: Instagram is examined as an unprecedented environment for studying human behavior and social interactions at scale.The platform combines social structure, media sharing, and user-generated tags.
- Contribution: The paper’s framework integrates network-, semantic-, and topical-based analysis of Instagram users, media, and their interrelations.The work is presented as a first study of Instagram interactions, tagging activities, and topical interests.
- Research questions: The study asks which structural features characterize the network built from users’ interactions.It also examines whether community structure reflects users’ self-organization around interests.
- Research questions: The study asks how users produce and consume content and how they interact with content produced by others.This addresses engagement across the production and consumption cycle.
- Research questions: The study examines how diverse users’ tags are and how users can be grouped by the tags they use.These questions address tagging behavior and topical clusters of interest.
- Research questions: The study asks how topical interests affect popularity and how the variety of topics differs across users and media.It also investigates whether popularity relates to topical interests.
2. METHODOLOGY
The study constructs an Instagram dataset by crawling users and media associated with public, competition-driven communities, then models participant relations as a directed weighted network. The resulting sample contains broad activity data but excludes some temporal and latent-interaction information.
- Data collection: The dataset was collected through Instagram APIs because direct access from network administrators was unavailable.The crawling strategy sought consistency in relationships, topical variety, and time coverage.
- Crawling strategy: Researchers shifted from geolocated media to users participating in Instagram’s Weekend Hashtag Project contests.The contests supplied thematic channels for sampling users and media.
- Crawling strategy: About 2,100 users from 72 popular contests were selected, and all media uploaded by these users were collected.Collected fields included media identifiers, users, timestamps, tags, likes, and comments.
- Network construction: The Relational Instagram Network is a directed weighted graph representing follower-followee relations and interaction intensity through likes and comments.The network was built from contest participants used as seed nodes.
- Dataset scale: The dataset contains over 2,000 users, almost 1.7 million media, about 9 million tags, 1.2 billion likes, and 41 million comments.Data were crawled over approximately one month from January 20 to February 17, 2014.
- Limitations: The sample is centered on a competition-driven Instagram community and lacks fellowship creation timestamps needed for some temporal network analyses.It therefore does not represent the full Instagram ecosystem.
3. ANALYSIS AND RESULTS
The analysis addresses Instagram’s network structure, content cycle, tagging behavior, and topical organization. It examines whether interactions and interests shape communities and popularity, while clustering users by their tagging behavior.
- Q1: Network structure: The analysis asks whether social relations and interactions shape the structure of the Relational Instagram Network.It specifically considers whether communities reflect self-organization around interests.
- Q2: Content production and consumption: The analysis asks how Instagram’s cycle of content production and consumption is characterized.It compares the driving mechanism of production with consumption measured through social interactions.
- Q3: Social tagging dynamics: The study examines tag adoption patterns, the emergence of popular tags, and whether users focus attention on few or many contexts.This combines user-level and global-level social-tagging dynamics.
- Q4: Topical clusters of interest: Users are clustered by tagging behavior to determine whether topical clusters emerge.The clustering question treats tags as signals of users’ interests.
- Q5: Popularity and topicality: The study tests whether user popularity relates to the variety and specificity of topical interests.It hypothesizes that popular users may exhibit different attention patterns and topical interests.
3.1 Structural features of the Instagram Network
The Relational Instagram Network shows heterogeneous connectivity and community sizes, with communities separating clusters of closely connected users from more isolated groups. These patterns are interpreted as consistent with preferential attachment and self-organization around topics.
- Network structure: Figure 1 presents the distributions of node degree and community size in the Relational Instagram Network.The analysis focuses on how these structural distributions characterize the network.
- Network structure: The network exhibits a scale-free node-degree distribution and a broad distribution of community size.These characteristics suggest growth might follow preferential attachment and communities might arise through self-organization around topics.
- Community structure: Communities separate close clusters of individuals from clusters of isolated individuals in the network visualization.Nodes in the same community share colors, while edge hues transition between source and target communities.
3.2 Content production and consumption
Instagram content production and consumption show distinct, broadly distributed dynamics: media output varies across users, while likes follow a power law and comments exhibit two regimes.
- Content production is analyzed through the amount of media uploaded by each Instagram user.The supplied passage introduces the user-level distribution but does not report its complete shape.
- Content consumption measures users’ likes or comments on media produced by others.This defines consumption as an interaction with another user’s media rather than redistribution for information diffusion.
- Likes follow a power law with exponent γ = 1.391 (xmin = 3, σ = 0.001).
- Comments show two regimes, x ⪅250 and x ⪆250, and no significant power law fit was found.The likes distribution also shows a finite-system-size tail cutoff.
- Media popularity measured by likes grows through preferential attachment, whereas likes and comments may follow different dynamics.
3.3 Social tagging dynamics
Instagram tagging combines concentrated individual usage with broad global tag popularity, while user entropy reveals variation from focused to highly heterogeneous tagging behavior.
- Tagging analysis covers global tag popularity, tags per media, user tag totals and vocabularies, and individual tag-use diversity.
- Tag popularity follows a power law with exponent γ = 1.865 (xmin = 2, σ = 0.002), while tags per media fit exponential decay.
- Most media use only a few tags, and larger tag sets become increasingly unlikely, consistent with a least-effort tagging pattern.
- Users’ total and distinct tag distributions are fat-tailed, but their distinct-tag vocabularies span over one order of magnitude less than total usage.The authors suggest users adopt only a few tags because they cannot track all tags emerging on the platform.
- Tag-adoption entropy peaks between 5 and 6; about 50% of users fall between 4 ⪅x ⪅7, while others are focused below 4 or extremely heterogeneous above 7.Entropy is used as a proxy for how concentrated or spread users’ attention is across tagging contexts.
3.4 Topical clusters of interest
Users are clustered from tag-frequency vectors into five topical groups, whose characteristic tags distinguish application use, geographical and subject themes, and attention-seeking or microcommunity behavior.
- Users are represented as term-frequency vectors over media tags and clustered with Bisecting k-Means.The clustering uses CLUTO’s globally optimized implementation for high-dimensional, large datasets.
- Figure 8 displays a five-way clustering using selected descriptive and discriminating tags, reordered through hierarchical clustering.Cluster-column width is proportional to the logarithm of cluster size.
- The five clusters are quite well-balanced.
- The first two clusters are characterized by application hashtags such as #VSCOCam and #latergram, with #latergram also linked to #tbt and geographical hashtags.
- The fifth cluster centers on feeling- and nature-related tags, while the third and fourth involve attention-seeking or microcommunity-focused tags.
3.5 User popularity and topicality
User popularity is positively associated with topical entropy, but popular users also show unusually specialized or broad topical interests. The analysis models interests from media tags and compares popularity across topical-entropy groups.
- Topic modeling: The study models users’ topical interests from media tags using a 10-topic Latent Semantic Indexing model.Tags occurring only once were filtered out, representing roughly 20% of the corpus; results were consistent with larger topic counts.
- Popularity and topical entropy: As user popularity increases, topical entropy also increases, indicating broader interests among more popular users.Very popular users had median topical entropy around 0.1 bits higher than unpopular users, with t=3.674 and p=0.0005.
- Popularity and topical entropy: Popular users include outliers with either unusually low or unusually high topical entropy.Thus, popularity is associated with both highly specialized and very broad topical interests.
- Popularity and topical entropy: Unpopular users tend to have more focused interests than popular users.The reported relationship is mild rather than deterministic, because popular users span extreme topical-entropy values.
4. DISCUSSION
The discussion summarizes Instagram’s network, content, tagging, and topicality findings. It links popularity to broader topical interests on average while emphasizing that popular users can be either highly specialized or broadly interested.
- Network and community structure: Instagram’s interaction network has scale-free node degree and broadly distributed community sizes, consistent with preferential attachment and topical self-organization.The discussion interprets these structural patterns as possible explanations for network growth and topic-oriented communities.
- Content production and consumption: Already engaged users may produce more content, while users favor likes over comments in line with a least-effort consumption principle.The production and consumption mechanisms are described as heterogeneous and effort-sensitive, respectively.
- Social tagging: Most media use only a few tags, producing a power-law distribution of tagging activity.The discussion relates this pattern to limited attention and competition among tags.
- Topical clusters of interest: Tags reveal user clusters associated with content-production practices, approval-seeking, or membership in microcommunities.These clusters are detected through the tags users adopt to label their content.
- Popularity and topicality: Higher topical entropy is associated with higher popularity, yet popular users exhibit both highly specialized and broadly varied interests.The discussion states that general-interest and specialized users have the same chance of becoming popular.
5. RELATED WORK
Prior work uses social media and online communities to study behavior, interaction, influence, trends, and attention at scale. Instagram-related studies provide context on network evolution and interactions, while this work additionally incorporates content information.
- Online social systems: Social media research has examined human behavior, interaction, influence, protests, political interests, trend diffusion, and attention.These studies motivate using online communities as proxies for large-scale social processes.
- Related platform studies: Flickr studies analyzed network topology, picture popularity, and information propagation over time without incorporating tags or comments.This work contrasts that focus with an analysis that uses content and interaction signals together.
- Related platform studies: Facebook research found that interaction activity is concentrated on a small portion of users’ social links, producing an interaction graph distinct from the social graph.The Instagram network likewise treats likes and comments as interaction signals.
6. CONCLUSIONS
The paper presents a broad analysis of Instagram’s combined social, tagging, and media-sharing ecosystem. It finds relationships between network structure, content behavior, tagging, topical interests, and popularity, while noting that network structure’s role in popularity remains unresolved.
- Overall contribution: The framework combines social relationships, user interactions, social tags, and media sharing to analyze users, media, and their interrelations.The analysis spans network-, semantic-, and topical-based research questions.
- Network and communities: Topical interests may shape user interconnectivity and interactions, forming communities that can be explained by self-organization.The paper presents this as an observed pattern and possible explanation rather than a definitive causal mechanism.
- Topicality and popularity: Users with broader interests tend to be more popular, but popular users can produce either highly specific or broadly appealing content.This conclusion describes a mild relationship alongside extreme topical behavior among popular users.
- Open question: Further work is needed to determine what role network structure plays in determining content and user popularity.The paper identifies this as an unresolved direction for online ecosystems based on social connectivity and content sharing.