The question answering (QA) system is a one hot topic related to IR, IE, NLP, data mining, and so on. A QA system mines a large set of texts to find a short phrase or sentence that answers a user's question with a certain precision. If the source is the Web or, even further, the whole information world, the challenge of the QA system increases dramatically.

There are basically three kinds of QA systems: slot filling, for which the format of the query and answer are similar; limited domain, for which the domains are limited with dictionaries and ontologies; and open domain, for which there is no limitation of domains.

One of the conceptual architectures among various solutions is illustrated in the following image. The input of the QA system is natural language questions; the output of the system, that is, the answer to the input question, is provided in natural language too. The system is composed of three major parts: the user interface, the processing part for questions, and the generation part for answers.

Assuming the question to be a coherent sentence, the questions from the user interface are processed, and the final right query is generated after a question-term identification, such as word segmentation and key word extraction and expansion. The ontology base here serves to support query expansion.

The answer is selected or generated from the FAQ base by the query that is provided in the earlier steps.

One popular question answering system is Yahoo! Answers, where a huge number of questions are asked and answered every day.