Relevant CommonSense Subgraphs for "What if..." Procedural Reasoning

TL;DR

This paper introduces a multi-hop graph reasoning model that extracts relevant commonsense subgraphs to improve causal reasoning for "What if..." questions in procedural texts, achieving state-of-the-art results.

Contribution

It presents a novel approach for efficiently extracting relevant commonsense knowledge and reasoning over it to answer causal questions in procedural contexts.

Findings

Achieved state-of-the-art performance on WIQA benchmark.

Effectively extracts relevant commonsense subgraphs from large knowledge graphs.

Improves causal reasoning accuracy for "What if..." questions.

Abstract

We study the challenge of learning causal reasoning over procedural text to answer "What if..." questions when external commonsense knowledge is required. We propose a novel multi-hop graph reasoning model to 1) efficiently extract a commonsense subgraph with the most relevant information from a large knowledge graph; 2) predict the causal answer by reasoning over the representations obtained from the commonsense subgraph and the contextual interactions between the questions and context. We evaluate our model on WIQA benchmark and achieve state-of-the-art performance compared to the recent models.