The invention provides an automatic abstract generation method based on a neural Turing machine. The automatic abstract generation method comprises the following steps: step 1, constructing a news Chinese corpus; step 2, carrying out abstract generation model training; and step 3, decoding by adopting a neural Turing machine, and after the encoder reads the whole source text, forming a word sequence of the text abstract by the decoder part. According to the method, the main content of the news is extracted in an automatic abstract mode, short abstract content easy to understand is formed, manual semantic feature extraction is not needed, and dependence on features extracted by natural language processing tools such as part-of-speech tagging and semantic analysis is avoided.