Managing Cognitive Bias in Reddit
Any complex recommender system is guaranteed to contain some level of bias. This is sometimes due to developer oversight but also because recommender systems are inherently approximate. One simple, yet effective way to combat this is to provide the user alternative avenues towards searching and discovering information. Granting obvious alternatives in support of the “top-n recommended results” aids in the overall integrity of information both learned and provided. In other words, the real-time strength of searching capability is in support of a fluid and filtered dynamic database, and vice-versa.
Screenshot from shivarweb
One easy example of this would be from Reddit, where upvotes and downvotes (binary positive and negative ratings) are used to sort links and their respective discussion comments according time-weighted, decreasing-order values. High-value comments and high-value links are, in a default setting, the most accessible and therefore most consumed information on the site. Given user time-constraints, the overwhelming majority of information will not be reviewed in a recommender-only state, and it is for this reason that large-scale forums are subject to problems of balance, selection bias, and other forms of cognitive bias. This ultimately leads to ideas becoming popularized due to reasons not principally related to value of content, but often because of time, manipulation, or “bandwagon effect”.
Reddit link sorting function; n= vote ratio, t= time
Filtering is the predominant corrective action available for normalizing user interaction. Reddit accomplishes this through granting alternative methods of navigation via sorting through recent, archived, scores ranging from best to worst, popular or “hot”, controversial topics, et al. Furthermore, the ability to customize the front-page with more obscure forums or “sub-reddits”, that typically might reflect one’s unique interests, is also very effective at diving into special topics. Reddit also employs methods to weight and time-enhance (see above image) information in an effort to limit bias, but ultimately a strong filtering interface is crucial to balance a robust system. Fundamentally, we find yet another very complex hybrid model, but the complexity is not the point in and of itself. The point is made through maintaining user engagement and not through spoon-feeding media. While this may work for some sites, it is not a winning model.
Sources
https://www.shivarweb.com/3778/use-reddit-seo-content-marketing/