Alex Austin

Introduction

Amazon tracks its users’ buying behavior, stores that information, and offers it back in the form of suggested purchases. “Frequently purchased together” will appear with a list of relevant products below an item a user is viewing. This visualization focuses on political books sold through Amazon and illustrates which books are frequently purchased together. The resulting image reveals the political beliefs that underpin book purchasing habits.

Materials

I sourced a dataset from CASOS, used OpenRefine and Google Sheets to convert the file types and clean the data, and I created the visualization in Gephi.

Methods

The dataset started as an XML file. It contained information that centered around a series of “nodes” and “edges,” as is standard for network datasets. The nodes (what is represented by the dots in the visualization) represent books that Amazon users have purchased, and the edges (the lines connecting the dots) represent frequent co-purchasing of the books.

I imported the XML file into OpenRefine and converted it into a CSV file. Then I imported the CSV into Google Sheets and created one spreadsheet of nodes and one spreadsheet of edges. I removed the excess columns of irrelevant information, organized the order of the columns, and added a column specifying that the type of edge data was “undirected,” meaning that all connections between nodes were reciprocal.

I imported the CSV files into Gephi, and started to play with the software. I adjusted the layout to ForceAtlas2 and ran a series of functions including modularity. The modularity function revealed that the data fit into four groups of frequently co-purchased groups. I researched the titles listed in each group and looked for a unifying theme or perspective for each group, and chose a color that held a corresponding meaning. The two largest groups on the outermost sides of the network represented books that were written from democrat and republic perspectives, respectively. Those I colored blue and red. The two groups in the center were less polarized and not affiliated with a political group of their own, so I chose to color them with yellow and beige, which hold no strong political meaning in American culture.

While creating this visualization, I reflected back on an animated visualization we had examined together in class. It was an animated network published in Business Insider that illustrated the 70 year evolution of political polarization in American politics. This visualization presented a clear narrative through simple coloring of red and blue dots, and an emphasis on the location of the nodes. 

The books in the yellow group tended to focus on the failures of governmental agencies and institutions, especially in regards to international politics and war. The books in the beige group also sometimes focused on the governments’ failures, though some of them carried an overtone of hope for how the government’s institutions can have a positive impact when operating well.

Results and Interpretation

It was not surprising to me that the books that were critical of government agencies and their part in war lay in the center of the visualization. In a post-Bush age, a wariness of engagement in foreign wars is still a fairly centrist inclination.

The edges in this visualization represent frequent co-purchasing of books by the same buyers. The edges in this graph are not measured by degree, which indicates that the edge values are binary (books are either bought together FREQUENTLY or they are NOT). What numeric threshold Amazon uses to determine whether books are frequently co-purchased is unclear. However, the threshold certainly is accurate. When I ran a modularity function through Gephi to determine the grouping of nodes by similarity, it became clear that the books represented by the nodes had strong grouping with fairly distinct boundaries between the groups. This means that readers often buy books with similar political points of view, and rarely venture out of these siloed buying patterns.

Since Amazon uses its own “frequently bought together” data to suggest purchases to buyers as they browse specific items, one has to question the effect that this has on buyer behavior. Were “frequently purchased together” data not used to suggest further purchases to buyers as they browse, would the siloing of co-purchases remain the same? As our culture has become aware of the polarizing political effects of social media algorithms, so too must we question whether Amazon’s purchasing suggestion algorithm has had similar effect on readers’ buying habits and resulting political beliefs. Without having a control group dataset of political book purchasing in physical bookstores for comparison, it is difficult to answer this question. One must also acknowledge that the development of political beliefs that are expressed through Amazon book purchases does not exist independently from the development of political beliefs that are expressed through social media. 

Reflection

I chose this dataset because I anticipated that it would yield fairly distinct groups of nodes. I tend to feel very uneasy when using new software, so I figured if the dataset was one with a fairly intuitive meaning, I would be able to check whether I was using the software correctly. This proved to be a good strategy as I had enormously difficult time working with OpenRefine, and a fairly difficult time working with Gephi. It’s a steep learning curve, but I’m coming along.

An important aside: Often in discussions of political polarization, “polarization” is characterized as a negative phenomenon. However, many may agree that it is more nuanced than that. In this area of discussion, I find it hard not to allow my personal beliefs to show, or more accurately, I don’t care to hide them. While the rise of the far-right is highly concerning, the shared beliefs at the center of bipartisan politics 50 years ago are also concerning. Do we really wish to return to a time when both major political parties supported a system in which women could not open their own bank accounts and gay marriage was decades aways from consideration? Bipartisan has a tendency to look like center-right. That the left has moved away from this center is more hopeful than concerning. That the right has moved farther right from this center is exceptionally concerning; though in my more cynical moments I wonder whether the right is moving father right or simply becoming more transparent with its ideologies.