The paper analyzes the conditions of research on the current scientific literature text feature extraction methods, applies the TF-IDF method based on background knowledge in the medical text feature extraction, and conducts experimental comparison in four medical fields. The result indicates that this method can obviously improve the extraction effect when there are few vocabularies to be extracted, and is obviously superior to the IDF based TF-IDF method in the aspects of filtration of commonly-used words in the text set and identification of important feature words.