In view of such flaws as error transmission or loose connection between different subtasks found in the traditional assembly line method of knowledge base Q&A, a new method of knowledge base Q&A system has been proposed, with multi-task learning incorporated into the knowledge base quiz system so as to improve its effectiveness. Allowing multiple subtasks to share a single encoder enables the model to acquire a better underlying representation, thus helping to improve the generalization ability of the model. Experimental results on the CCKS2022-CKBQA task verifies the better performance of the proposed method in this paper.