Abstract
Studying social media influence in the Global South is often hampered by a lack of high-quality metadata regarding account identities. This paper introduces DISMISS (Database of Indian Social Media Influencers on Twitter), a manually curated dataset of 11,580 influential Indian accounts. Categorized into 7 broad sectors and 24 subcategories, DISMISS provides a verified ground truth for researchers analyzing narrative propagation, political polarization, and public health discourse within the unique Indian social media ecosystem.
Introduction
While social media has become a primary arena for public discourse, identifying the specific actors driving "influence" remains a significant challenge. Standard data collection via the Twitter API provides vast quantities of text and engagement metrics but lacks critical metadata: it does not distinguish between a government official, a professional journalist, or a "digital-first" meme account.
Existing automated methods for identifying user attributes, such as Machine Learning classifiers, frequently suffer from "overoptimistic accuracy." These tools often struggle with the nuances of local contexts, particularly in the Global South, where influence is fragmented by language, region, and specific professional hierarchies. DISMISS was created to bridge this gap, providing a human-annotated "who's who" of the Indian Twitter landscape.
Methodology
The researchers utilized an iterative, "human-in-the-loop" approach to construct the database, ensuring higher fidelity than automated classification alone.
- Seed Set Selection: The process began with a known list of Indian political accounts. This was based on the logic that politicians typically follow individuals of public importance, such as journalists, activists, and celebrities.
- Snowballing: The list was expanded by examining the "friends of friends"—specifically, the accounts followed by the initial seed set.
- Filtering and Relevance: To maintain focus on the Indian context, non-Indian accounts were manually removed. The team also applied an "in-group connectivity" filter, retaining only accounts followed by at least three users from the seed set to eliminate personal acquaintances and non-influential connections.
- Manual Annotation: This core phase involved human verification of each account. Researchers assigned a Category (e.g., Media, Government, Business) and a Subcategory (e.g., Journalist, Police, Brand) to every entry. They also distinguished between Individual and Entity accounts.
The DISMISS Dataset
The resulting DISMISS dataset contains 11,580 unique records. Each record includes the unique Twitter ID (id_str), handle, name, bio, follower count, and verified status, alongside the manually assigned categorical metadata.
Unlike previous datasets that focus narrowly on specific groups like politicians or celebrities, DISMISS captures a broad spectrum of actors. This includes "Digital First" accounts—entities like meme pages or informational bots that derive their influence entirely from online activity rather than traditional offline fame.
Key Insights and Behavioral Patterns
Analysis of the dataset reveals distinct behavioral signatures across different sectors of Indian society:
- The Dominance of the Fourth Estate: The "Media" category is the largest segment of the dataset, comprising over 5,000 accounts. Within this, "Journalist" is the most common subcategory, highlighting the central role traditional media practitioners still play in anchoring the Indian Twitter ecosystem.
- Broadcasters vs. Networkers: Different industries exhibit unique networking patterns. Business accounts typically function as "broadcasters," following very few people (often fewer than 100) while maintaining massive follower counts. Conversely, Civil Society accounts (academics, NGOs) tend to have a narrower follower range ($10^3$ to $10^5$), suggesting they operate within more specialized, niche networks.
- The Rise of "Digital First" Influence: The dataset identifies 684 "Digital First" accounts. These accounts represent a unique challenge for researchers; they often have massive reach but unverifiable offline identities, making them critical players in both organic viral trends and coordinated manipulation efforts.
Results and Validation
The dataset's utility is evidenced by its scale and categorization:
- Media: 5,079 accounts
- Creative: 2,698 accounts (entertainers, writers)
- Civil Society: 1,184 accounts (academics, specialists, NGOs)
- Government: 977 accounts (bureaucrats, police, official entities)
DISMISS has already been utilized in peer-reviewed research to analyze polarization during political crises and the spread of "dangerous speech" during the Bangalore riots, proving its value as a foundational resource for social science.
Limitations and Future Directions
While DISMISS is a robust resource, it has limitations. The initial sampling process showed a linguistic bias toward Hindi-speaking states, which dominate the national narrative. Although the authors attempted to mitigate this by adding regional journalists and politicians, some imbalance remains.
Furthermore, Twitter is a dynamic platform where handles change and accounts are suspended. The authors recommend that researchers use the provided id_str to refresh metadata via the API before use. Future work will focus on deepening specific subcategories, such as "Business," to better characterize sectoral interactions on the platform.
Summary updated