The invention discloses a question labelling method based on deep reinforcement learning for an online question and answer platform. The method is based on a deep neural network and a reinforcement learning technology. According to the technical scheme, the method comprises the following steps: firstly, establishing a model, innovatively adding an index for measuring the diversity of problem labels while ensuring the labeling accuracy when designing model rewards, and meanwhile, considering a tail label effect, namely, long labels which are described in a complex and detailed manner during labeling, so that the labels can describe the problems deeply and more detailedly. According to the method, the accuracy and diversity of labels are comprehensively considered, the influence of the taillabel effect on problem labeling is reduced, the training efficiency and accuracy of the reinforcement learning model are improved by introducing the deep neural network, and the matching error rangecan be ensured under a certain confidence degree. According to the scheme provided by the invention, mass questions and labels in the question and answer platform can be accurately and diversely matched.