Cognitive Heuristics and Collective Opinions in Peer Recommendation

The world presents far more information than people have the capacity to examine. As a result, humans have evolved to use cognitive heuristics to decide quickly what information to pay attention to. For example, people pay more attention to items near the top of a list than those below [Payne 1951]. A consequence of this cognitive heuristic, called position bias, is the strong effect the presentation order — item ranking — has on individual choices. It affects which items in a list of search results users click on [Buscher et al. 2009], and the answer they select in response to multiple choice questions [Blunch 1984]. Another common cognitive heuristic is social influence: people pay attention to the choices of others. Social influence affects most daily decisions, such as what to buy and who to vote for. Studies showed that social influence biases individual judgements [Salganik et al. 2006; Muchnik et al. 2013], creating an “irrational herding” effect that can obscure the underlying quality of choices. Cognitive heuristics also play an important role online, where rapid proliferation of user-generated content makes it difficult to identify high-quality items. Since people often do not have time or energy to evaluate all available choices, they may rely on the opinions of others. Crowdsourcing, peer recommendation, and markets are some of the mechanisms for aggregating individual decisions into a collective opinion, which can help individuals identify high-quality items. The choices content providers make about how and what information to display to users has a profound effect on collective behavior. For instance, the choice of how to rank items shown to users can significantly affect outcomes of peer recommendation, since people will pay more attention to items near the top of the list. Similarly, displaying a social signal, which shows people how many others have liked an item, will also affect how much attention it receives. These effects combine with item quality to determine collective outcomes of peer recommendation. We disentangle some of these effects through controlled web-based experiments using Amazon Mechanical Turk. Our studies quantify how position bias and social influence bias affect individual choices and collective outcomes of peer recommendation, and clarify results of earlier studies. For example, we demonstrate that large variation in popularity can arise in the absence of social influence simply due the attention items receive due to their position [Lerman and Hogg 2014]. In addition, we measure how social influence signals affect popularity after controlling for item quality and position. We find that social influence causes people to rely on others, rather than personal judgement, to determine whether the item is interesting. Although this contributes to the “irrational herding effect” [Salganik et al. 2006; Muchnik et al. 2013], social influence has a benefit that was not appreciated previously: by reducing the effort required to evaluate items, it increases the efficiency of peer recommendation. We also demonstrate that we can leverage people’s innate cognitive biases to more effectively aggregate collective opinions in peer recommendation. We show that simply by changing item ranking, we can distribute collective attention more homogeneously over higher quality items, reducing the unpredictability and variance of peer recommendation outcomes.