New arXiv paper shows how LLMs can enhance social network analysis
Researchers can now use LLMs to decode meaning and narratives in social ties.
This preprint from arXiv (July 2026) presents a framework for using Large Language Models in qualitative and mixed-methods social network analysis. The author, Moses Boudourides, argues that LLMs should enhance—not replace—human researchers. The paper starts by outlining the core principles of qualitative SNA, such as understanding the meaning of ties, narratives, and relational identities. It then details how LLMs can assist with data collection (e.g., extracting context from texts), coding (e.g., thematic or narrative coding), and theory-building through abductive reasoning. The emphasis is on deepening rigor while keeping human interpretation central.
The paper does not shy away from limitations. It highlights risks like algorithmic bias, hallucinated relationships, and the need for constant researcher reflexivity. Ethical challenges include ensuring diverse perspectives aren't flattened by dominant training data. The author closes with concrete research designs and recommendations—such as using LLMs for exploratory phases and validating outputs against human analysis. For researchers, this is a practical guide to integrating LLMs thoughtfully into qualitative SNA without losing the human touch.
- LLMs are positioned as augmentation tools, not replacements, for qualitative social network analysis.
- The paper covers use cases like data collection, coding, theory-building, and abductive reasoning.
- Ethical challenges (bias, hallucination, reflexivity) are addressed with actionable recommendations.
Why It Matters
This gives researchers a responsible blueprint for using LLMs to deepen qualitative network insights without sacrificing rigor.