The invention provides a method for automatically extracting text information, which can continuously improve the accuracy of automatically extracting marker content and label. The method comprises the following steps: acquiring a text file uploaded by a user, and converting the text file into a document format capable of character-by-character analysis by a computer; preprocessing the text content of the converted document to form a multi-level text unit which is easy to parse by using natural language processing technology; capturing text segments selected by a user, determining markup contents corresponding to the text segments based on the formed multi-level text units, and recommending labels for each markup contents; based on the determined markup content and the recommended tags foreach markup content, an online learning training idea being adopted to train the automatic text extraction model, so as to realize the automatic extraction of markup content and tags. The invention is suitable for automatic text information extraction operation.