An Evidence Reasoning Approach to Solve the Plan Recognition Problem

Abstract

Plan recognition has wide applications in the real world. In this paper, we present a novel probabilistic plan recognition method based on an evidence reasoning approach. We analyze the theories of intentional action, and put forward our plan recognition method on the basis of the agent model and the environment. By extracting the characteristics of the agent actions from the history data, we can make a brief model of the agent. During the online plan recognition process, we first initialize the probability for each goal according to the environment model. Then we use the agent model, treat the observed action as evidence and use a variant of Dempster-Shafer theory, weight assignment with a forgetting factor, to combine the newly observed action with the former estimation. We illustrate our approach with an example, then make further discussions and conclusions.