Online Reputation Model Using Multiple Quality Factors

Abstract

Users on the internet are looking for ways to minimize their experiences in performing online transactions. Reputation systems as a decision support tool are trying to facilitate online transactions. However, many reputation systems use Naïve methods to compute the reputation of an item. These methods are unstable when there is sparsity in the ratings. Also, they cannot discover trends emerging from recent ratings. Other methods, which use weighted average or probabilistic model, usually focus on one aspect of the reviewer ratings. Even though models that combine multiple factors often accomplish that through an arbitrary set of weights. This research study looks at various aspects of reviewers’ ratings and proposes a new reputation model that attempts to assess the reviewer reputation by combining four factors through a Fuzzy model. These weights are then involved in computing the item reputation. The proposed reputation model has been validated against state-of-art reputation models and presented significant accuracy regarding Mean Absolute Errors (MAE) and Kendall correlation. The proposed reputation model also works well with the sparse and dense dataset.