Multi-step Entity-centric Information Retrieval for Multi-Hop Question Answering
This work addresses the problem of multi-hop evidence retrieval for question answering, which is crucial for improving QA systems, though it appears incremental as it builds on existing IR and QA methods.
The paper tackled the challenge of retrieving multiple supporting evidence for multi-hop question answering by introducing an IR technique that uses entity information from initially retrieved evidence to hop to other relevant evidence. The approach significantly boosted retrieval performance on over 5 million Wikipedia paragraphs and increased an existing QA model's F1 score by 10.59 on the Hotpot benchmark.
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.