Papers-
[1] Echo chambers online?: Politically motivated selective exposure among Internet news users – Kelly et. al.
[2] Bursting your (filter) bubble: strategies for promoting diverse exposure – Resnick et. al.
Summaries-
[1] discusses how user’s political leanings affect how they interact with news articles. It collects data from hundreds of users of news sites, and conducts a behavioral tracking experiment to see whether users prefer to interact with content they agree with or content they disagree with. It finds that users are less likely to interact with information they disagree with, but they do not actively avoid it. It constructs five hypotheses, considering whether users look at information which supports or detracts from their own viewpoints, and how long they spend looking at these articles.
[2] is a very short survey type paper, which, after quickly defining the need to design tools to provide diverse exposure and discourse on the Internet, goes on to discuss some implementations try to address this problem by helping users understand the biases of the content they consume, or to consider/explore alternate perspectives, or engage in discourse with a wide variety of viewpoints.
Reflection-
[1] is an interesting read, and makes some fascinating claims, but it has a few flaws. Firstly, it was published around 2009, which was right at the dawn of the age of machine learning for recommender systems. This meant that most websites did not have user specific curated content at that time. The hypothesis discussed by the article, which suggests that the internet may not create echo chambers, since users are not particularly averse to looking at views which go against their own, is not as valid in today’s world. Due to automatic recommender systems, users do not have a choice in this matter anymore, and may be continually exposed to partisan information simply because of their prior information usage patterns.
Secondly, the paper admits that the selection of candidates for the study was not exactly a good representation of the entire nation. The users who signed up for this study already had strong political views, since they were active on either a left leaning or a right leaning website from before. Moreover, more than half of them had a college degree, and the ethnicities of the participants were heavily skewed.
Interestingly, [1] mentions that while only 20,000 people saw the recruitment statement on the left-leaning website (AlterNet), 100,000 people saw the statement on the right-leaning website (WorldNetDaily). However, both sites were almost equally represented in the final selection of candidates, despite the recruitment statement being seen by 5 times as many people on WorldNetDaily. This could hint at an inherent “willingness to participate” of left-leaning people, or might simply be because the readers of the left-leaning site had a lower income on average (as claimed by the paper), and thus desired the participation prize more.
[1] also makes a claim that opinion challenging posts would also lead to an increase in the duration for which the user engages with the content, which is later backed by the data. However, users would probably be less inclined to immediately close articles which they disagreed while interacting with a new unfamiliar software interface, when they know they were taking part in a monitored survey, as they would be when browsing privately.
It is interesting to see that fears of rising political polarization catalyzed by Internet technologies were prevalent not just in 2009, but also as early as 2001, as indicated by the citations made by [1]. It is almost eerie to see these insights become relevant again, more than a decade later.
Many of the systems discussed in [2] would also have a tendency to become biased, depending on the beliefs of the majority share of the users of the systems. For example, if more liberals used Reflect or Opinion Space, then those comments would be more prevalent, and would receive more positive reviews from other liberals.
Opinion Space in [2] reminded me of the abstract interfaces mentioned in the Chat Circles paper, as it creates a space for users to navigate with, where users interact with different types of comments. It also changes the physical characteristics of the comments based on how the users interact with them.
