research

My background is in computational social science with a focus on political communication and social media. Across projects, my through-line is how non-traditional political actors gain credibility and mobilize attention in fragmented media environments — especially in Nepal and other understudied political contexts.

My existing work is primarily observational and descriptive — using network analysis, text classification, and multimodal machine learning to characterize how political attention and discourse are structured online. I am drawn to questions about how alternative and independent media actors shape political attitudes and mobilization, particularly in elections and moments of political crisis. My methods so far have been largely predictive and associative; where I want to go is toward experimental and psychologically-grounded approaches that can get closer to causal claims about media effects on political attitudes. The research questions below reflect that direction of travel — not a summary of completed work, but the questions I am actively thinking through as I prepare PhD applications.

Research questions I want to pursue

Alternative media, source credibility, and political persuasion

Possible research questions include:

• To what extent do source identity cues — alternative versus legacy media, partisan versus independent framing — shape perceived credibility and political attitude formation among audiences with low institutional trust?
• How do visible social consensus signals (likes, comment tone, engagement volume, mutual-friend endorsements) function as persuasion heuristics when audiences assess political content from alternative media actors?
• Under what conditions does sustained exposure to creator-led political content shift perceived majority opinion and anti-establishment sentiment among digitally active youth?

Influencer dynamics, multimodal framing, and political mobilization

Possible research questions include:

• Through what mechanisms does creator endorsement translate into electoral mobilization — and how much of that effect is attributable to the content itself versus the audience’s pre-existing affinity with the source?
• How do multimodal content features (visual identity, linguistic framing, affective tone) of independent creators differ from legacy media in their framing of political candidates, and which features drive differential engagement in low-resource, high-mobile-penetration contexts?
• What structural features of the online information environment — narrative clustering, cross-platform amplification, echo-count dynamics — explain why some counter-narratives reach non-aligned audiences while others do not?

Current and prior research

Beyond the Ballot: TikTok Virality and Political Engagement in Nepal’s 2022 Local Elections. MSc thesis, Hertie School (submitted and completed 2025; supervised by Prof. Simon Munzert). Revised after committee feedback; not yet published as a journal article. [Thesis PDF] [Code]

A multimodal machine-learning study of 28,165 political TikTok videos; a multimodal subsample of 2,964 videos integrates audio (Whisper), video (CLIP), and metadata features (full model AUC 0.800, 5×5 cross-validation). This is primarily a predictive and associative study — the analysis carefully separates description, prediction, and association, but does not establish causal claims, which remains a significant limitation and an open direction.

The central finding — that political visibility is patterned by sender identity and actor capacity more than communication style, with independent candidates (particularly Balen Shah, now Prime Minister of Nepal) driving disproportionate attention — raised a deeper question: not what content goes viral, but how alternative, non-party media actors built outsized credibility and affective resonance in a fragmented political landscape. That question drives the ongoing research interest.

Beyond the ban: co-engagement network signals on Nepali Reddit in the run-up to the Sep 8 Gen Z protest Ongoing independent research. [Poster]

A cross-platform structural analysis of Reddit and Facebook discourse preceding the protest, using the Meta Content Library and ArcticShift API. Focuses on network-level reorganization — contraction, densification, bridging community rotation — as precursors to volume spikes, rather than content classification. Descriptive and structural; accepted as a poster for the How Are You, Democracy? conference, Karlsruhe (June 2026).

Stance Polarization on Bluesky around a Political Assassination of Right-wing Influencer Candidate in the US. Computational group project, SICSS-Saarbrücken 2025 — I led stance classification and author-network analysis. [Slides]

Descriptive classification study of roughly 156K Bluesky posts combining zero-shot stance classification, toxicity scoring (Perspective API), and author-network analysis over a 46K-node network with community detection. Exploratory in scope; co-authored with SICSS project teammates.

Mapping Elite Attention Ecosystems on Instagram: Gen-Z-Salient Political Attention in Nepal. Exploratory network-analysis project (2026). [Slides] [Proposal]

Exploratory mapping of the attention and following network around Nepal’s digital Gen-Z movement on Instagram, focusing on influential activists as seed nodes. A preliminary structural snapshot rather than a causal or longitudinal analysis.

Fact-Checking in the Feed: The Gap Between Output Metrics and Public Reception. Mixed-methods commentary co-authored at logiq.media. Submitted to the HKS Misinformation Review (under review). [Draft]

A conceptual and mixed-methods piece examining why fact-checking output volume diverges from reception in contested feeds. Not an empirical study — closer to a critical commentary drawing on available evidence and practitioner knowledge from the logiq.media context.

For implementation detail and code, see Projects; for the full list of presentations and talks, see Research Outputs.