The invention relates to a verb phrase ellipsis resolution method based on depth learning, and belongs to the technical field of computer artificial intelligence. The invention solves the problems oflow accuracy of triggering word judgment and antecedent phrase recognition existing in the existing verb phrase ellipsis resolution method. According to the method, the determined data set 1 and the determined data set 2 are preprocessed; in the process of judging triggering words, the extraction of sentence context features and sentence-level features is added, and the extracted sentence featuresare transformed into vectors to be input into a support vector machine and then the triggering words of input sentences are determined according to output results of the support vector machine; and finally, a correct antecedent phrase is identified from candidate antecedent phrases generated by the triggering words by using a multi-layer perceptron. When extracting sentence features, the invention adds context features and sentence-level features, so that the accuracy rate of triggering word judgment can reach about 90%, and the accuracy rate of antecedent phrase recognition can reach 85% orabove. The invention can be applied to the technical field of computer artificial intelligence.