Online Reputation Model Using Moving Window

Users are increasingly dependent on decision tools to facilitate their transactions on the internet. Reputation models offer a solution to the users in supporting their purchase decisions. The reputation model takes product ratings as input and produces product quality as score. Most existing reputation models use naïve average method or weighted average method to aggregate ratings. Naïve average method is unstable when there exist a clear trend in the ratings sequence. Also, the weighted methods are influenced by unfair and malicious ratings. This paper introduces a new simple reputation model that aggregates ratings based on the concept of moving window. This approach enables us to study variability of ratings over time which allows us to investigate the trend of ratings and account for sudden changes in ratings trend. The window size can be defined by either number of ratings or duration. The proposed model has been validated against stat-ofart reputation models using Mean Absolute Error and Kendall tau