Using Visibility to Control Collective Attention in Crowdsourcing

Online crowdsourcing provides new opportunities for ordinary people to create original content. This has led to a rapidly growing volume of user-generated content, and consequently a challenge to readily identify high quality items. Due to people’s limited attention, the presentation of content strongly affects how people allocate effort to the available content. We evaluate this effect experimentally using Amazon Mechanical Turk and show that it is possible to manipulate attention to accomplish desired goals. Peer production systems allow ordinary people to contribute original content in the form of photos (e.g., Flickr, Instagram), videos (e.g., YouTube, Vimeo), news and text (e.g., Twitter, blogs), reviews (e.g., Amazon, Yelp), and much more. While the quantity of user-generated content has skyrocketed, its quality varies dramatically: on YouTube, for example, one can find home videos of a four year old’s first violin recital, as well as virtuoso performances by accomplished violinists. Content providers have innovative methods to identify high quality content based on crowdsourcing or peer recommendation. Social news aggregator Digg, Flickr and Yelp for example, ask their users to recommend interesting news stories, photos and restaurants respectively, and prominently feature most recommended items. Content providers face a dual goal: identify quality content while keeping users engaged by showing them quality content. These goals often conflict. To understand why, consider a simple content evaluation strategy in which the provider shows users a random selection of items and asks them to evaluate them, for example, by recommending interesting items. After enough people examine an item, its quality, or how interesting it is to people, should be reflected in the number of recommendations it receives (its popularity) (Salganik, Dodds, and Watts 2006). However, if there are only a few quality items, the selections seen by most people may not contain a single interesting item. When users continue to see low quality content, they may not return to the site. The provider could instead present the highest quality items that it knows about, but if users do not inspect all items, some high quality items may be missed.