Exploring Abstract Argumentation-Based Approaches to Tackle Inconsistent Observations in Inductive Logic Programming

Abstract

Noisy (uncertain, missing, or inconsistent) information, typical of many real-world domains, may dramatically affect the performance of logic-based Machine Learning. Multistrategy Learning approaches have been tried to solve this problem by coupling Inductive Logic Programming with other kinds of inference. While uncertainty has been tackled using probabilistic approaches, and abduction has been used to deal with missing data, inconsistency is still an open problem. In the Multistrategy Learning perspective, this paper proposes to attack this latter kind of noise using (abstract) Argumentation, an inferential strategy aimed at handling conflicting information. More specifically, it defines a pre-processing operator based on abstract argumentation that can detect and remove noisy atoms from the observations before running the learning system on the polished data. Quantitative and qualitative experiments point out some strengths and weaknesses of the proposed approach, and suggest lines for future research on this topic.