Public defence, Computer Science, MSc Hasti Nariman Zadeh

Title of the thesis: Inferring Latent Structures in Social Systems: From Network Patterns to Subjective Judgments

Thesis defender: Hasti Nariman Zadeh
Opponent: Professor Matteo Magnani, Uppsala University, Sweden
Custos: Associate Professor Mikko Kivelä, Aalto University School of Science

Many socially important characteristics cannot be observed directly. Scientific disciplines, political positions, social divisions, and individual perceptions often have to be inferred from their observable consequences. Digital high-dimensional data provide increasingly detailed traces of social activity, but these traces are incomplete and shaped by the systems through which they are produced. 

This doctoral thesis examines how hidden social structures and characteristics can be inferred from such indirect evidence. It develops computational approaches for studying social systems through three types of observable data: networks, online communication, and human judgments. 

The first part investigates what can be learned from network structure. One study uses scientific citation networks to identify disciplinary organization while separating it from known sources of structure, such as the historical development of science. Another develops a method for estimating the climate-policy positions of actors in an online community from patterns of interaction combined with a small number of expert assessments. 

The second part examines how social divisions become visible in language. By studying online discussions about climate change and COVID-19, the research shows how divisions between groups can be expressed through hostile and uncivil communication, and how antagonistic patterns associated with one political issue can spill over into another domain. 

Finally, the thesis turns to subjective human judgment. Concepts such as toxicity are not perceived in the same way by everyone, making them difficult to measure reliably. The thesis investigates comparison-based evaluation as an alternative to assessing items independently and shows that comparisons can reduce measurement error and some forms of systematic judgment bias. 

Together, the studies contribute methods for recovering socially meaningful information that is not directly observable but leaves traces in patterns of connection, communication, and evaluation. The results are relevant to research on scientific communities, political and social polarization, online interaction, and subjective measurement. More broadly, the thesis shows how computational methods can help researchers move from observable behavior towards a more systematic understanding of the hidden structures and processes that shape social systems.

Keywords: Networks, Community detection, Network inference, Social networks

Thesis available for public display 7 days prior to the defence at Aalto University's public display page.