EAR is a model designed for query expansion and reranking in the context of information retrieval and question answering. It implements a method that generates diverse expansions of a given query and then employs a reranker to select those expansions that enhance the retrieval of relevant passages. The approach is detailed in the paper 'Expand, Rerank, and Retrieve: Query Reranking for Open-Domain Question Answering' (Findings of ACL 2023).
The tool is intended for use in tasks such as question answering and information retrieval, particularly where improving the relevance of retrieved passages is important. EAR provides official scorer checkpoints for reranking queries, and it is compatible with the Transformers library, allowing users to integrate the model into their workflows using high-level pipelines or by loading the model directly. It supports English language queries and references the use of models such as BM25 and DeBERTa-v3 in its methodology.
EAR is distributed under the MIT license. The model is available on Hugging Face, where users can find instructions for deploying it with various libraries, inference providers, notebooks, and local applications.
EAR is a Foundation models & chat product. It focuses on improving information retrieval and question answering by expanding and reranking queries for better passage retrieval. EAR is an open-source project aimed at information retrieval researchers and developers. The project is open source (Open Source). EAR is available on the web, API, and the command line.
Behind EAR is voidism, and the product first shipped in 2023. The project is developed in the open on GitHub with 38 stars. Across PulseGate's embedding index, EAR has few near neighbours, marking it as relatively distinct. Among its 5 catalogued features are query expansion, reranking, and question answering.
Latest indexed changes and source events
voidism/EAR discovered by the PulseGate indexer
Other apps tracked under the same category.
