Online discussions, user reviews and comments onthe Social Web are valuable sources of informationabout products, services, or shared contents. The rapidlygrowing popularity and activity of Web communitiesraises novel questions of appropriate aggregation anddiversification of such social contents. In many cases,users are interested in gaining an extensive overviewover pros and cons of a particular track of contributions.We address the problem of social content diversificationby combining latent semantic analysis with featurecentricsentiment analysis. Our FREuD approach providesa representative overview of sub-topics and aspectsof discussions, characteristic user sentiments underdifferent aspects, and reasons expressed by differentopponents. In experiments with real world productreviews we compare FREuD to the typical implementationof ranking reviews by the usefulness rating providedby users as well as a naive sentiment diversificationalgorithms based on star ratings. To this end we hadhuman users provide a fine-grained gold standard aboutthe coverage of features and sentiments in reviews forseveral products in three categories. We observed thatFREuD clearly outperforms the baseline algorithms ingenerating a sentiment-diversified set of user reviewsfor a given product.