A reputation inference model based on linear hidden markov process

Abstract

Although reputation is a statistical value about the trust, most existing work uses the summation method for reputation aggregation, which is vulnerable to malicious feedbacks and cannot assess the reputation prediction variance. In this paper, we present a novel Linear Hidden Markov (LHM) model for reputation evaluation. LHM model uses the linear autoregressive function to define the reputation evolution, so that the reputation prediction variance can be assessed by a Markov process. Based on Expectation Maximization (EM) calibration method, LHM model aggregates the feedback by using the Kalman filter, which can support further robust inference techniques. Our experiments show that LHM model can effectively capture the reputation value and its prediction variance.