A New Trust Model Based on Time Series Prediction and Markov Model

Abstract

In this paper, we propose a new statistical predictive model of Trust based on the well-known methodologies of the Markov model and Local Learning technique. Repeatedly appearing similar subsequences in the trust time series constructed from history of direct interactions or recommended trust values collected from intermediaries over a sequence of time slots are clustered into regime. Each regime is learnt by a local model called as local expert. The time series is then modeled as a coarse-grain transition network of regimes by using a Markov process and value of the trust at any future time is predicted by selecting the local expert with the help of the Markov matrix. © Springer-Verlag Berlin Heidelberg 2010.