SmallWorlds: Visualizing Social Recommendations

Abstract

We present SmallWorlds, a visual interactive graph-based interface that allows users to specify, refine and build item-preference profiles in a variety of domains. The interface facilitates expressions of taste through simple graph interactions and these preferences are used to compute personalized, fully transparent item recommendations for a target user. Predictions are based on a collaborative analysis of preference data from a user's direct peer group on a social network. We find that in addition to receiving transparent and accurate item recommendations, users also learn a wealth of information about the preferences of their peers through interaction with our visualization. Such information is not easily discoverable in traditional text based interfaces. A detailed analysis of our design choices for visual layout, interaction and prediction techniques is presented. Our evaluations discuss results from a user study in which SmallWorlds was deployed as an interactive recommender system on Facebook.

Inspectability and control in social recommenders

“…In the realm of social recommenders, Groh et al [17] present a study that outlines the "extensive need" for explanation, and Gretarsson et al [16] present a small-scale study of an explanation interface, finding that the explanation process has a positive effect on satisfaction with recommendations.…”

“…More recent work [16,35,37] discusses interactive graphical representations of the recommendation process, to enable control over both item-and user-level preferences in collaborative recommender systems.…”

“…Although this reasoning may seem intuitive, little research has been done to establish the effect of inspectability and control on the users' experience with social recommender systems ( [16,17] are notable exceptions). This paper describes the results of an online user experiment (N=267) with a Facebook music recommender in which we independently manipulated the level of inspectability and control.…”

User-controllable personalization: A case study with SetFusion

“…PeerChooser (O'Donovan, Smyth, Gretarsson, Bostandjiev, & Höllerer, 2008), and SmallWorlds (Gretarsson, O'Donovan, Bostandjiev, Hall, & Höllerer, 2010) are examples of interactive visual interfaces that represent a collaborative filtering paradigm, where users increased their satisfaction under the visual interactive interface compared to a more static condition. More recent work has focused on providing users control over the recommendation interface by allowing users to sort the recommendation list based on different item features in an energy-saving application (B. P. Knijnenburg, Reijmer, & Willemsen, 2011), by letting users indicate their preferences at different levels of granularity in a music recommender (Hijikata, Kai, & Nishida, 2012), or by permitting them to combine several recommendation sources using sliders in a music and a job recommender (Bostandjiev, O'Donovan, & Höllerer, 2012.…”

Explorative Analysis of Recommendations Through Interactive Visualization

“…One of the first notable proposals in this area is PeerChooser [20], whose graph-based interface is supposed to provide a visual explanation of the collaborative filtering process and to enable the users to interactively manipulate their neighborhood. Also using a graph-based interface, SmallWorlds [21] allows to express item preference profiles with which personalized item recommendations are generated in a transparent manner. This enables the users both to explore their relationships to their peers and to discover new relevant items.…”