Effect of the initial configuration for user-object reputation systems

Identifying the user reputation accurately is significant for the online social systems. For different fair rating parameter q, by changing the parameter values α and β of the beta probability distribution (RBPD) for ranking online user reputation, we investigate the effect of the initial configuration of the RBPD method for the online user ranking performance. Experimental results for the Netflix and MovieLens data sets show that when the parameter q equals to 0.8 and 0.9, the accuracy value AUC would increase about 4.5% and 3.5% for the Netflix data set, while the AUC value increases about 1.5% for the MovieLens data set when the parameter q is 0.9. Furthermore, we investigate the evolution characteristics of the AUC value for different α and β, and find that as the rating records increase, the AUC value increases about 0.2 and 0.16 for the Netflix and MovieLens data sets, indicating that online users' reputations will increase as they rate more and more objects.