Graph Attention with Hierarchies for Multi-hop Question Answering

TL;DR

This paper enhances a hierarchical graph neural network model for multi-hop question answering by adding new edges and a novel attention mechanism, improving performance on the HotpotQA benchmark.

Contribution

It introduces two key extensions to the Hierarchical Graph Attention Network, improving its ability to perform multi-hop reasoning in question answering.

Findings

Enhanced model with new edges improves reasoning accuracy.

Sequential node update via hierarchy boosts performance.

Proposed modifications demonstrate efficiency on HotpotQA.

Abstract

Multi-hop QA (Question Answering) is the task of finding the answer to a question across multiple documents. In recent years, a number of Deep Learning-based approaches have been proposed to tackle this complex task, as well as a few standard benchmarks to assess models Multi-hop QA capabilities. In this paper, we focus on the well-established HotpotQA benchmark dataset, which requires models to perform answer span extraction as well as support sentence prediction. We present two extensions to the SOTA Graph Neural Network (GNN) based model for HotpotQA, Hierarchical Graph Network (HGN): (i) we complete the original hierarchical structure by introducing new edges between the query and context sentence nodes; (ii) in the graph propagation step, we propose a novel extension to Hierarchical Graph Attention Network GATH (Graph ATtention with Hierarchies) that makes use of the graph hierarchy to update the node representations in a sequential fashion. Experiments on HotpotQA demonstrate the efficiency of the proposed modifications and support our assumptions about the effects of model related variables.