About Me

Diya Saraf dsaraf2 [at] jhu [dot] edu

👋 I’m a Ph.D. student in Computer Science at Johns Hopkins University, where I’m part of the Social Computing group advised by Prof. Tiziano Piccardi. My research sits at the intersection of computational social science, algorithmic auditing, and human-AI interaction, focusing on the gap between users’ stated and revealed preferences in social media feeds — what people say they want to see versus what algorithms learn they engage with.

🔍 Research Interests

Computational social science · algorithmic auditing & accountability · recommender systems & personalization · LLMs & NLP for modeling user behavior · human-AI interaction · network science · trust & safety (misinformation, coordinated manipulation, harmful content amplification) · responsible & human-centered AI · data-driven experimentation (field studies, browser instrumentation, controlled audits)

💼 I’m broadly interested in industry roles at the intersection of ML and social systems — research internships in responsible AI, integrity/trust & safety, recommender systems, and computational social science teams.

🧪 Current Research

• 🎯 Preference Modeling: comparing LLM-verbalized preference summaries against embedding-based approaches for predicting user engagement with social media posts.
• 📖 Content Auditing at Scale: auditing knowledge platforms (e.g., Grokipedia vs. Wikipedia) for systematic differences in coverage and framing.
• 🧰 Instrumented Data Collection: building browser-based tooling to collect real feed data from consenting study participants.

📚 Prior Research

I completed my M.S. in Computer Science at the University of Southern California, where I was part of the HUMANS Lab at the Information Sciences Institute (ISI), working with Emilio Ferrara and Luca Luceri on:

• 📱 TikTok VISTA Project: studying algorithmic amplification of nicotine-related content.
• 🕸️ Telegram Coordination Graphs: analyzing multilingual propaganda networks (npj Complexity, 2025).
• 🎭 Audit Puppets: using controlled personas to audit YouTube and TikTok recommendations.
• 🤖 Generative Influence: examining how large language models drive or detect online persuasion.

🎓 Background

• B.S. Computer Science & Engineering, minor in Math, Santa Clara University – Clare Boothe Luce Scholar, advised by Yuhong Liu – Research on online trust and emoji-based engagement (IEEE DPSH 2024).
• Software Engineering Intern, Dell Technologies – Worked on CI/CD automation and encryption infrastructure.

✨ Interests

Data storytelling, media history, and sociotechnical systems.

Fun facts:

• 🎶 My music taste is an unpredictable mix — retro groove one minute, current pop the next.
• 🧩 I’m a daily word-puzzle person — you’ll usually find me mid-NYT Strands or Wordle before I’ve had breakfast.
• ⛰️ I find my peace in the mountains — nothing resets me like crisp air and a good trail.