OpenDomain

Open-domain refers to tasks and systems that operate across a broad set of topics rather than a single, restricted domain. In open-domain question answering (Open-Domain QA), the goal is to answer questions using information drawn from large, heterogeneous corpora such as the web, encyclopedias, or knowledge bases. A typical open-domain QA system executes a retrieval step to fetch relevant passages or documents, followed by a reading or extraction step to produce an answer. Retrieval can be traditional inverted indexing with BM25 or neural dense retrievers such as DPR. Readers can be extractive (select text spans) or generative (produce answer from retrieved content) using large language models; sometimes both in a retrieval augmented generation (RAG) setup. Sources include Wikipedia, web crawls, or curated knowledge bases. Evaluation uses EM and F1 on QA datasets; success depends on coverage, latency, and robustness.

History and development have seen significant milestones, such as DrQA (2017), Dense Passage Retrieval (2019), and RAG (2020), which helped popularize open-domain QA. Challenges facing open-domain systems include hallucination or fabrication of facts, reliability and quality of sources, real-time retrieval requirements, multilingual and cross-domain adaptation, and safety concerns. Applications span search assistants, chatbots, and customer support, where users expect accurate answers drawn from broad information sources. See also open-domain information retrieval and retrieval-augmented generation for related concepts.