ESTIMATING REPUTATION SCORES IN REPUTATION SYSTEMS

摘要

The disclosed embodiments provide a reputation system. The reputation system includes a statistical model associated with a set of items and a set of dimensions of the items in the reputation system, wherein the statistical model is trained using a positive class and a negative class. The reputation system also includes a scoring apparatus that applies the statistical model to a set of features for each of the items to estimate a set of reputation scores for the items. Finally, the reputation system includes a ranking apparatus that enables use of the set of reputation scores in the reputation system.

主权项

1. A computer-implemented method for facilitating use of a reputation system, comprising:
obtaining a statistical model associated with a set of users and a set of skills of the users in the reputation system, wherein the statistical model is trained using a positive class and a negative class; applying the statistical model to a set of features for each of the users to estimate a set of reputation scores for the users; populating a matrix of the users and the skills with the reputation scores and unknown values for user-skill pairs that do not have explicit reputation scores; calculating a factorization of the matrix to produce at least two factor matrices, wherein the at least two factor matrices are multiplied to generate one or more inferred values that replace the unknown values in the matrix; and obtaining a ranking of the users by one or more skills specified by a query, wherein the ranking is based on a linear combination of reputation scores associated with the one or more skills; wherein obtaining the ranking of the users includes:
obtaining an inverted index of the one or more skills,using the inverted index to identify a subset of the users with reputation scores in the one or more skills, andranking the subset of the users by the reputation scores.