Abstract: Innovation is the center of the enterprise development, which is the main driving force of enterprises' competitiveness. To evaluate the enterprises' innovation ability, we firstly establish an index system which aims at analyzing the enterprise innovation ability. We also retrieve the data of the listed companies in Wind database from 2015 to 2019 and label them using factor analysis method. Then, a new deep learning classificational framework with attention mechanism and LSTM is established. The results show that when attention mechanism and LSTM are added into the convolutional neural network(CNN), the model's prediction performance is better improved, and the accuracy, recall, precision and F-score are 0.914, 0.914, 0.916 and 0.915, respectively. This indicates the strong generalization ability of our new model. Finally, we also find that patents and R&D expenditures are the most important factor affecting the corporate innovation ability through SHapley Additive exPlanations(SHAP). Companies with more patents and R&D expenditures are generally considered to be more innovative.

Key words: Corporate innovation ability, convolutional neural network, Attention mechanism, long short-term memory