When you perform a search 🔎, your query is matched against a collection of documents. Hopefully, the result will be the most relevant documents.
Instead, a Question Answering system expects a question asked in a natural language and returns an answer.
QA systems can use different technical approaches: they can be rule-based, machine learning-based, or hybrid.
The knowledge used by Question Answering systems can be stored in different forms: SQL databases, search platforms, vector databases, knowledge graphs... In some machine learning QA systems, the knowledge is included in the machine learning model.
Closed-domain QA systems focus on a specific domain and can rely on domain-specific knowledge frequently formalized in ontologies. In contrast, open-domain QA systems expect questions about a wide variety of topics, are less specific, and typically deal with larger amounts of data.
Within the ODQA category, we can distinguish between open-book QA and closed-book QA.
In open-book QA 📖, the relevant information is retrieved from a knowledge repository. Then, this information is processed to extract or generate an answer.
In closed-book QA 📕, a text generation model can only rely on its internal knowledge to answer a question.
- An Introduction to Open Domain Question-Answering: Pinecone blogpost with a detailed and practical overview on Question Answering.
- Understanding Semantic Search: Deepset blogpost with a timely overview of the landscape and jargon of semantic search and question answering systems.