An Evolutionary Prediction Model for Enterprise Basic Research Based on Knowledge Graph

Abstract

Currently, China's enterprise basic research faces problems due to a need for more systematic guidance and dispersed themes. The construction of an enterprise basic research knowledge map is of great practical significance for tracking cutting-edge technologies, tapping the primitive scientific and technological innovation ability of enterprises, and playing the leading role in enterprise innovation. By mining the intrinsic correlation between the data, a multilayer CNN-LSTM-based correlation prediction model for enterprise basic research is proposed, using the number of published papers of enterprises as the experimental dataset, reasoning and completing the knowledge graph of enterprise basic research, and predicting the future direction of enterprise basic research; and constructing a probability calculation model of enterprise basic research with a multi-attention mechanism, and probabilistically calculating the future hotspot direction of enterprise basic research. The experimental results show that compared with the existing classical model, the recall and F1 value of the multilayer CNN-LSTM evolutionary prediction model are significantly improved. It can more accurately capture the cutting-edge research topics in several types of basic research fields, which provides a new perspective for the related personnel to predict the trend of basic research fields.