Multi-step Entity-centric Information Retrieval for Multi-Hop Question Answering

TL;DR

This paper presents a novel entity-centric multi-hop information retrieval method that leverages entity information to improve evidence retrieval for multi-hop question answering, significantly boosting performance on large-scale datasets.

Contribution

Introduces an entity-aware IR technique that learns to hop between evidence pieces, enhancing multi-hop QA retrieval effectiveness without additional training.

Findings

Achieves significant retrieval performance improvements on large-scale Wikipedia data.

Increases QA model F1 score by 10.59 on Hotpot benchmark.

Demonstrates effectiveness of entity-based hopping in multi-hop IR.

Abstract

Multi-hop question answering (QA) requires an information retrieval (IR) system that can find \emph{multiple} supporting evidence needed to answer the question, making the retrieval process very challenging. This paper introduces an IR technique that uses information of entities present in the initially retrieved evidence to learn to `\emph{hop}' to other relevant evidence. In a setting, with more than \textbf{5 million} Wikipedia paragraphs, our approach leads to significant boost in retrieval performance. The retrieved evidence also increased the performance of an existing QA model (without any training) on the \hotpot benchmark by \textbf{10.59} F1.