Collaborative Filtering Using Multidimensional Psychometrics Model

Abstract

In this paper, the psychometrics model, i.e. the rating scale model, is extended from one dimension to multiple dimension. Then, based on this, a novel collaborative filtering algorithm is proposed. In this algorithm, user’s interest and item’s quality are represented by vectors. User’s rating for an item is a weighted summation of the user’s latent ratings for the item in all dimensions, in which the weights are user-specific. Moreover, user’s latent rating in each dimension is assumed to follow a multinomial distribution that is determined by the user’s interest value, the item’s quality value in this dimension, and the thresholds between two consecutive ratings. The parameters are estimated by minimizing the loss function using the stochastic gradient descent method. Experimental results on the benchmark data sets show the superiority of our algorithm.