What social characteristics enhance recommender systems? The effects of network embeddedness and preference heterogeneity

Recommender systems utilize social relationships to improve recommendation performance. This study explores social characteristics and how they affect recommendation performance. We define social characteristics as network embeddedness and preference heterogeneity. Taking rating characteristics as control variables, we build a regression model to explore the impact of two social characteristics on user-level predictive accuracy and the moderating effect of preference heterogeneity on the relationship between network embeddedness and user-level predictive accuracy. The results suggest that network embeddedness positively influences predictive accuracy, whereas preference heterogeneity negatively influences it. Our research reveals that as the preference heterogeneity increases, the positive effect of network embeddedness on predictive accuracy weakens. Preference heterogeneity has a greater impact on user-level predictive accuracy than network embeddedness. Our findings provide management implications for recommender system designers, which is of great significance for improving the accuracy of user-level prediction and reducing user complaints.