A Filtering-Based General Approach to Learning Rational Constraints of Epistemic Graphs

Abstract

Epistemic graphs are a generalization of the epistemic approach to probabilistic argumentation. Hunter proposed a 2-way generalization framework to learn epistemic constraints from crowd-sourcing data. However, the learnt epistemic constraints only reflect users’ beliefs from data, without considering the rationality encoded in epistemic graphs. Meanwhile, the current framework can only generate epistemic constraints that reflect whether an agent believes an argument, but not the degree to which it believes in it. The major challenge to achieving this effect is that the time performance will become unacceptable when the number of restricted values increase. To address these problems, we propose a filtering-based approach using a multiple-way generalization step to generate a set of rational rules which are consistent with their epistemic graphs from a dataset. This approach is able to learn a wider variety of rational rules that reflect information in both the domain model and the users model, and therefore more suitable to be applied to some situations, e.g. automated persuasion system, where the statistical information about the beliefs of a group of users is exploited to predict the behaviours of a specific user. Moreover, to improve computational efficiency, we introduce a new function to exclude meaningless rules. The empirical results show that our approach significantly outperforms the existing framework when expanding the variety of rules.