The growing polarization of opinions in America on a wide range of topics has attracted attention in both public discourse and academia. Widespread adoption of personalized recommender systems has transformed the way that opinions are formed, expressed, changed and disseminated. As a consequence, opinions tend to be reinforced by exposure to cultural and ideological grouping even when individuals within those niches are not explicitly tied by social relations. This project will use data from Amazon book reviews and purchases to construct a book co-purchase and co-reviewer network with which to study opinion clustering and polarization of options. Particular attention will be directed to four key questions: Are there dense clusters of books, with few ties between the clusters? What genres of books produce clustering? Are there sorts of books that bridge between the clusters, and do these books then cluster into a smaller number of topics that correspond to crosscutting cultural dimensions? Do books that bridge between clusters enjoy greater sales, given the broader market, fewer sales, given their deviation from the core in each cluster, or similar sales but with a higher standard deviation?
The proposed project will make use of advanced computational methods to collect and process book co-purchasing data and time-stamped book reviews from Amazon.com. Books will be hand-coded into categories and network analysis will be used to locate the structural positions of those topics in the co-purchase and co-review. This dissertation makes three principal contributions to research on opinion dynamics: 1) a better understanding of how online recommender and review systems both reflect and promote opinion polarization, 2) the extension of theories and models of homophily to clustering within bipartite networks of people and cultural objects; 3) the development of novel research methods that use digital traces of consumer purchases and reviews to study opinion dynamics and polarization.
This research has practical implications for understanding and addressing social divisions in America. It provides a unique but complementary perspective on the causes and dynamics of polarization and may yield practical suggestions for ways to bridge divisions. This project is also one of the first to use "big data" to study polarization, and will provide the research community with new datasets and tools. In addition, the results may be helpful for authors who want to target their books to a more diverse audience. The project will also advance the training of undergraduate and graduate students who work on the project.
