Methods

The Computational Communication Science Lab (CCL) focuses on computational methods of different kinds. Here you can find an overview of our most commonly employed methodology.

Exponential random graph models

Exponential random graph models (ERGM, or p* model) allow examining underlying generative mechanisms that produce network ties (called edges) between a set of actors (called vertices) based on stochastic micro-level social processes. Unlike traditional parametric statistical methods, ERGMs can accommodate a great range of network-endogenous processes. This opens the door for new questions and theories on interactions between actors, which long has been the core interests of social sciences. Because of this, ERGMs has emerged as one of the most appropriate types of inferential techniques for relational data (including social and communication networks) that has seen increased usage in recent years.

Agent-based modeling

Agent-based modeling (ABM) is a computational model for understanding complex social systems by modeling adaptive behaviors of social actors, or “agents” within given environment. Agents are autonomous actors that behave in accordance to certain pre-defined rules, and they continuously interact with other agents and the surrounding environment. This approach allows us to examine complex, large scale system-level consequences that arise from micro-processes among many agents, which is not otherwise amenable to robust inferences using traditional statistical methods commonly employed in Communication science.

Automated Content Analysis

Text mining refers to the process of extracting meaningful information from unstructured natural language text. Text mining is an umbrella term for an assortment of computational techniques that include classification, natural language processing, and topic modeling, among others.

Vast amounts of available digital and digitized data become increasingly hard (if not impossible) to analyze relying solely on traditional content analysis methods. Automated techniques of text analysis allow researchers to work with and draw inferences from large scale corpora. Ongoing work focuses on the application of hierarchical dictionary-based approaches across multiple languages.

Machine learning for text classification

Automated text analysis increasingly relies on the application of various machine learning methods. Both supervised (e.g., dictionary-based) and unsupervised (e.g., clustering and topic modeling) classification methods are being extensively applied by the CCL. Current research focuses on developing semi-supervised machine learning algorithms that are trained on human-coded data, typically obtained from crowdsourcing procedures.