The invention relates to an intelligent question and answer method and device. The method includes: receiving user input information; acquiring the highest semantic similarity between the user input information and all the questions in a question and answer knowledge base, and obtaining a corresponding initial answer when the highest semantic similarity is greater than a preset similarity threshold; performing abstract semantic recommendation processing on the user input information according to an abstract semantic database when the initial answer includes a plurality of clauses, and the plurality of clauses at least correspond to more than two pieces of first pragmatic information, and then obtaining an abstract semantic expression corresponding to the user input information and the type of an abstract semantic, wherein the abstract semantic database includes a plurality of types of abstract semantics, and each type of abstract semantic has a corresponding second pragmatic information; and extracting a clause of the first pragmatic information corresponding to the second pragmatic information from the initial answer as a final result. The method can improve the accuracy rate of reply answers of the intelligent question and answer system, and improve the user experience.