End-to-End Beam Retrieval for Multi-Hop QA
What it is
This is the official code repository for the NAACL 2024 paper 'End-to-End Beam Retrieval for Multi-Hop Question Answering'. It provides training and evaluation scripts for a beam retrieval model using DeBERTa as the backbone, along with downstream reader components for multi-hop QA datasets like MuSiQue, HotpotQA, and 2WikiMultihopQA.
The problem it solves
Standard retrieval-augmented generation pipelines often treat retrieval and reading as separate stages, leading to suboptimal performance on multi-hop questions where the retrieval strategy needs to be adaptive to the question structure.
What you could use it for
- Training dense retrieval models for multi-hop question answering
- Evaluating retrieval performance on MuSiQue, HotpotQA, and 2WikiMultihopQA benchmarks
- Researching end-to-end optimization of retrieval and reading components
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.