The invention relates to a case-oriented keyword extraction method. Firstly, a lexical vector matrix and a lexical feature matrix used for case description of a training model are obtained according to the case description of the training model. Using the keras package from the python library, the word vector matrix used to describe the case of the training model is put into a keras model, the lexical feature matrix used to describe the case of the training model is also put into a keras model. Then the two models are combined into one model by using the merge function in keras, and the modelis trained by using the fit function in keras. Finally, the case description of the keywords to be predicted is inputted into the model, and the keywords are extracted and the required keywords are extracted. The method of the invention can greatly improve the accuracy of keyword extraction and reduce the workload of case determination.