Computational Approaches for Consumer Memory and Decision Making

How does memory influence decisions? While the importance of memory has been previously highlighted by researchers, until recently, little was known about how people use memory when making everyday decisions. In my prior work, I developed a novel experimental task and a cognitive model to compare the memory processes in standard memory and decision tasks. In the first chapter of this dissertation, I extend my published results into a formal theoretical model of memory-based decision making. Overcoming multiple technical challenges with natural language processing and machine learning approaches, I present a joint model of memory and decision making that can be applied to rich naturalistic choice items (or any arbitrary item that a decision maker recalls when faced with everyday decisions) and make quantitative predictions for out-of-domain memory-based consideration sets. In the second chapter, I move beyond the study of individual-level memory-based consideration sets and examine why some words are more memorable than others. Here, I rely on semantic representations from large-scale digital data, and I develop a computational model that can accurately predict word recall and recognition memorability. Furthermore, I identify psychological constructs that are more likely to be remembered, and I test theoretical hypotheses to enhance our understanding of human memory. In the third chapter, I use similar computational and methodological tools. I present a data-driven approach that uses text explanations on healthcare websites in combination with large-scale survey data to predict, interpret, and better understand lay health perceptions and decision making. Altogether, this dissertation attempts to make methodological and theoretical contributions in multiple fields, including marketing, psychology, and cognitive science. The models in these chapters are not only accurate (compared to benchmark models and/or objectively) in making predictions for out-of-sample data; they are also cost-effective, scalable, interpretable, and a priori blind to any psychological theories. Thus, using these computational methods, I am excited to answer important questions from academia and industry concerning human decision making and memory.