To make computers obtain a deeper thinking, we propose a question semantic representation model based on concept knowledge tree, which integrates concepts and knowledge to express questions' semantics. Combined the question semantic representation model and related knowledge, we introduce a knowledge-driven semantic reasoning algorithm to recursively obtain new knowledge for deepening the comprehending of questions. We evaluate the performance of our system on a collected dataset, and the top-1 accuracy reaches to 62.1%. Experimental results demonstrate our proposed methods improve the depth of question comprehending.