報告題目：Perceptive users for online social systems: the Netflix case
摘要：Identifying perceptive users and user perceptibility is of importance for understanding the user collective behavior for user-object bipartite networks. Investigating the Netflix data set (containing 37,755,925 ratings delivered by 218,319 users on 7,803 movies during 2,241 days), we track the ratings given to the 13 objects which are nominated for the Oscar awards before and after the award-nomination time. The distribution of the time difference between the rating and the award-nomination time show that there exists a few users concern the award-nominated movies before the award-nomination time. In this paper, we present a parameter-free method to identify the user perceptibility, which is defined as the capability that a user can identify high-quality objects before they actually be widely approved (award- nominated). Besides the empirical results that high perceptibility users have larger degree, stronger correlation of rating series and higher reputation, we investigate the behavior patterns of the perceptive users from the burstiness and memory of rating durations, as well as user preference. The experimental results indicate that high perceptibility users prefer to rate less popular objects and the rating durations of high perceptibility users show lower burstiness and higher memory effects. Furthermore, the results of predicting high perceptibility users by means of machine learning algorithms show that the burstiness and memory coefficients along with user preference can improve the prediction performance in identifying the high perceptibility users based on user behavior patterns. This work provides a further understanding on the collective behavior patterns and perceptive users.