Recommending Sellers to Buyers in Virtual Marketplaces Leveraging Social Information

Abstract

Social information such as stated interests or geographic check-ins in social networks has shown to be useful in many recommender tasks recently. Although many successful examples exist, not much attention has been put on exploring the extent to which social impact is useful for the task of recommending sellers to buyers in virtual marketplaces. To contribute to this sparse field of research we collected data of a marketplace and a social network in the virtual world of Second Life and introduced several social features and similarity metrics that we used as input for a user-based k-nearest neighbor collaborative filtering method. As our results reveal, most of the types of social information and features which we used are useful to tackle the problem we defined. Social information such as joined groups or stated interests are more useful, while others such as places users have been checking in, do not help much for recommending sellers to buyers. Furthermore, we find that some of the features significantly vary in their predictive power over time, while others show more stable behaviors. This research is relevant for researchers interested in recommender systems and online marketplace research as well as for engineers interested in feature engineering.

Bartering Books to Beers

“…The effect of incorporating social information into collaborative filtering models has been shown to improve prediction accuracy and alleviate data sparsity (e.g., Ma et al [14], [7]). As discussed in Section 3, users tend to repeatedly trade with a selected subgroup of peers on the observed bartering platforms, suggesting that their choices have a strong social (or simply trust) component.…”