Answer extraction is an important part of the field of automatic question answering (QA). The traditional answer extraction method relies heavily on contextual semantics, which makes the extraction process consume a lot of time and manpower. Aiming at the question of answer extraction, a method based on the bidirectional long-term memory network conditional random field (Bi-LSTM-CRF) is proposed. This question is extracted in parts from the answer segment. Extract the entity that contains the answer, and then extract the final answer from it. On a valid data set, the accuracy of the experimental results can reach more than 0.6050.