The Problem with Google+
Google+ is notoriously known for its failure to catch on as a top social platform despite the fact that it was widely known. In fact, many Gmail users didn’t have to sign up for Google+; their Google account was also their Google+ account. Despite this, Google+ shut down in 2019 due to low usage.
I wanted to look at how Google+ social networks among users were structured in order to potentially gain insight into why people didn’t use the platform despite its accessibility and recognition. Specifically, I’m interested in Damon Centola’s theory that Google+ failed because of its combination of mass awareness and limited use. Centola claims in his 2021 book, Change: How to Make Big Things Happen, that combining widespread awareness with a lack of use leads people to view a product as unused and assume that something is wrong with it. As a result, people make a decision to not use the popularly unused product.
Because Google+ created automatic accounts for users yet many didn’t use the platform, I wanted to examine a network graph to see what the typical numbers of connections looked like. Based on the platform’s ultimate failure, I would assume that users tended to have very little connections/centrality, and influencers would not be prevalent. Without strong communities of users, people may not have been enticed to continue using the platform. In my network graph, I test this idea.
The data I am using to answer the question of whether a lack of centrality of users is a potential reason for social platform failure is from the Data for Good Lab. It is found in a CSV file where Google+ users are represented by ID numbers. In the left column is a list of IDs, and in the right column is the ID of someone who the person in the left column is connected to on the platform. This data allows me to determine characteristics of connections such as centrality, thus helping me to better understand whether the failure is in part due to people not interacting enough/having enough centrality on this platform.
In this data, a node represents a Google+ user (identified by a unique ID) and edges represent connections between Google+ users (meaning that in the data, these two user IDs are in a row together). Because Google+ members interact with connections by adding them to circles, we can assume that if a node has an edge to another node, it means the first user has added the second user to their Google+ circle.
‘Importance’ in my graph can be defined by centrality, as in being central to other nodes. This means that a node has a significant amount of influence. I calculated the degree centrality, a representation of the number of edges connected to a node. The node with the most degree centrality is node 136198 with a degree centrality of 0.008476. We see a quick drop-off of values within even the top 20 values, with 20th being 0.004172. The average degree centrality is much lower with a value of 5.1274337333260864e-05.
I also calculated the eigenvector centrality of our data. Eigenvector centrality differs slightly from degree centrality, measuring the influence of a node in a network by accounting for centrality to neighbors as well as the number of connections. The most important node in terms of eigenvector centrality is node ID 116002 with a centrality of 0.101318. However, eigenvector centrality drops off quickly even within the top 20 nodes, with the 20th most central (based on this measure of centrality) having a centrality of 0.069070. The average eigenvector centrality for this data is 0.00018; again, much lower than our top values.
As for closeness centrality, the measure of which nodes have the shortest paths, I used a subgroup of every 100th node because my computer couldn’t handle processing the full dataset. Within this subset, the most important node in terms of closeness centrality is ID 116002 with a closeness centrality of 0.009252. Meanwhile, the average closeness centrality within this subset is 0.0001, once again indicating a wide spread in values.
I cleaned up this data while graphing by using every 100th node. When I tried to include every node in my graph, my computer became overwhelmed and unable to properly plot points. With so many data points, plotting every 100th node still provides an accurate visual representation of what the Google+ network looks like as a map. I imported this selected data to Gephi and used the ForceAtlas2 layout.
Looking at this map, it is clear that nodes tend to not have edges outside of their small clusters. Yes, we divided the data by 100 so clusters are expected to be smaller and we can expect to see less edges, but it is still noticeable that most nodes have no edges and the central, more-connected nodes don’t have connections to other groups. This indicates a lack of communication between users on Google+, suggesting that although users automatically had accounts, they were not connecting with each other, potentially due to a widespread assumption that no one else wanted to use the platform to communicate.
My analysis is missing a measure of users’ activity levels. Because Google+ accounts were automatically created for many, it would be interesting to see how many of these users simply added a few connections and then ceased to use the platform. This could be measured by determining the average minutes per week a user used the platform and including these values in the dataset. As our data is, I am using centrality as an indicator for how much a user used the platform (assuming a user with more centrality is more active) but we have no way to know if the user was actually actively using the platform at the time this data was taken. Personally, when I was about eight years old, I added a bunch of family and friends to my Google+ ‘circle’ at once and then never used the platform again because I realized none of my new ‘connections’ were active users. We have no way to determine which of these users are in a similar situation, and this information would be necessary to determine the impact that centrality has on platform use. For now, we simply assume that centrality is positively associated with platform use.
I would also like to acquire data on successful social media platforms, such as Instagram, Facebook, and TikTok, and compare the centrality metrics of these platforms with that of Google+. I would assume that Google+ would have much lower average centrality measures than these other platforms. If we had this data, then we would be able to better understand just how different Google+ average centrality of users is from that of a successful platform, and therefore better understand how much centrality of the average user is correlated with platform use and ultimately platform success.
Overall, we can conclude that Google+ users tended to not have very high centrality, whether that be degree centrality, eigenvector centrality, or closeness centrality. Perhaps this is a reason why use of Google+ dwindled despite widespread public awareness of the platform; people simply weren’t connected enough. It seems as though Centola’s hypothesis that people didn’t want to use the platform because they were very aware that others were not using it may have been correct; if one doesn’t believe their real-life connections will use the platform, then why would they add them to their online ‘circle’? This phenomenon may have led people to not add each other as ‘connections’, therefore hindering centrality. With further analysis, we could compare this data to that of other platforms to see if centrality metrics were really the sole difference in use (and therefore the reason for failure), and determine how much centrality corresponds to activity, which would provide evidence on why this platform failed.
Below is a link to my GitHub repository which contains the code I used in the file assign2.ipynb
