Relation extraction method based on prompt learning in low-resource scene

The invention discloses a relation extraction method based on prompt learning in a low-resource scene, and the method comprises the following steps: 1), decomposing a complex relation into joint representation of a plurality of perspectives, the perspectives including a figure, a place, an action, and an active and passive relation; 2) using virtual words to represent each view angle relation obtained by splitting; 2.1) representing the representation relation of each view angle; 2.2) sampling the representation relation of each view angle; 2.3) using a mask pre-training model, creating m different relation virtual words for each relation r in a word list of the mask pre-training model, each relation virtual word paying attention to different view angles of one relation; and 3) according to the representation of each view angle relationship and the generation probability of each view angle relationship representation, obtaining a final complex relationship representation. According to the method, a relation e