Abstract: A method for extracting text relevant to a particular topic from a document to generate a narrower, condensed, more specific, and smaller version obtains information as to the text of a large document and searches the text information of the document based on first keywords to extract first pages. The first pages are inputted into a predetermined learning model to extract second keywords from the first pages, and the first keywords and the second keywords are integrated to obtain third keywords. The method further searches the text of the first pages based on the third keywords to extract second pages and a determination is made as to whether the second pages meet a predetermined page standard. If yes, the second pages are integrated and output. An electronic device and a non-transitory storage medium are also disclosed.