Dissertation: Narrative Generation for Agent-Based Models

Abstract: This dissertation proposes a four-level framework for thinking about having agent-based models (ABM) generate narrative describing their behavior, and then provides examples of models that generate narrative at each of those levels. In addition, “interesting” agents are identified in order to direct the attention of researchers to the narratives most likely to be worth spending their time reviewing. The focus is on developing techniques for generating narrative based on agent actions and behavior, on techniques for generating narrative describing aggregate model behavior, and on techniques for identifying “interesting” agents. Examples of each of these techniques are provided in two different ABMs, Zero-Intelligence Traders (Gode & Sunder, 1993, 1997) and Sugarscape (Epstein & Axtell, 1996).

My interests lie in the intersection of economics, networks, and computation. I am currently studying labour dynamics as a process where people flow throughout the economy by moving from one firm to another. I study these flows by looking at detailed data about employment histories of each individual and every firm in entire economies. Using this information, I construct networks of firms in order to map the roads that people take throughout their careers. This allows to study labour markets at an unprecedented fine-grained level of detail. I employ agent-based computing methods to understand how economic shocks and policies alter labour flows, which eventually translate into unemployment and other related problems.