The impact of social design on user contributions to online communities.

Abstract

The World Wide Web has become increasingly participatory through the widespread adoption of interfaces that facilitate user-generated content. These interfaces can be made more social by allowing users to view and respond to the actions of others. For example, Flickr (http://flickr.com) encourages photo sharing by allowing users to view and comment on others' photos, and Amazon (http://amazon.com) encourages purchases through the use of book reviews, discussion forums, and recommendations. In this thesis, we explore the utility of social designs for improving the quality and quantity of user contributions to online communities. We investigate the use of social design at several levels. First, in a series of online field experiments in MovieLens (http://movielens.org), we examine the potential for increasing the quantity of user contributions through the display of personalized, social information. Second, in a comparative, controlled field study across a variety of popular question and answer (Q&A) sites, we compare different models of participation and their impact on the quality of user contributions. Finally, in a study of hand-coded questions from several Q&A sites, we use machine learning techniques to understand the characteristics of users' requests that are predictive of informational quality. We find evidence that appropriate use of social information can increase the quantity of user contributions: social comparisons led MovieLens members to rate more movies, and persuasive messages to visit the MovieLens discussion forum were most effective when they compared the user viewing the message to another member. We also find evidence that the unstructured participation models characteristic of Web 2.0 sites increase the quantity, diversity, and responsiveness of user contributions, with no apparent overall cost to information quality. However, we find that unstructured participation leads many users to treat these sites as purely social resources. Therefore, to better support the utility of Q&A sites as informational resources, we contribute a computational framework that can reliably characterize user interactions as informational or conversational.