Reputation “Liquid Rank” Algorithm

Liquid Rank is an algorithm for computing reputation that is described in this blog post. The algorithm can take multiple parameters and inputs into account. Primarily, for different input ratings, it accounts for rating values of the ratings, financial values of the respective interactions, reputation ranks of the subjects supplying the ratings, time when the rating was provided, etc.

From one perspective, Liquid Rank can be thought of as an extension of Google's PageRank algorithm that is better suited to a marketplace, as it accounts for financial values. More expensive and more recent payments have more impact on the reputation of a supplier. As in PageRank, the greater the reputation of the rater, the higher the value of the rater’s ratings.

The algorithm may be implemented either in real-time—so that every transaction changes the reputation ranks within the community—or in a stepwise fashion, where community members’ reputations are calculated and updated hourly, daily, weekly, or monthly. From a practical perspective, the incremental version seems to be the most cost-effective and the specific period can be configured as a system parameter.

In extremely oversimplified form, the reputation of agent i at time t is its own reputation for previous time (Ri t-1 ) added to all the new ratings it has received from other agents j over the time period between t-1 and t, multiplied by the reputations of these raters for the previous time, as follows.

Ri t = Ri t-1 + ∑j ( Rj-1 t* Vi j t )

Reputations may be computed specifically for selected domains, such that a supplier with a high reputation in the area of its expertise might have a much lower reputation in some other domain. Within the same domain, different reputation scores may be computed for different traits—timeliness, cost, accuracy, etc. These fine-grained reputations may not be included in the first version of the reputation system.