Finding, Scoring and Explaining Arguments in Bayesian Networks

TL;DR

This paper introduces a novel method for explaining Bayesian Networks by defining probabilistic arguments, extracting relevant independent arguments, and translating them into natural language explanations.

Contribution

It presents a new definition of probabilistic arguments, an algorithm for extracting relevant independent arguments, and a natural language explanation scheme for Bayesian Networks.

Findings

The algorithm effectively identifies relevant independent arguments.

Arguments can be used to approximate message passing.

Natural language explanations enhance interpretability.

Abstract

We propose a new approach to explain Bayesian Networks. The approach revolves around a new definition of a probabilistic argument and the evidence it provides. We define a notion of independent arguments, and propose an algorithm to extract a list of relevant, independent arguments given a Bayesian Network, a target node and a set of observations. To demonstrate the relevance of the arguments, we show how we can use the extracted arguments to approximate message passing. Finally, we show a simple scheme to explain the arguments in natural language.