Academic News Text Classification Model Based on Attention Mechanism and RCNN

With the expeditious development of Internet technology and academic social media, massive academic news generated by academic social media have provided rich information for scholars to communicate and learn about the latest academic trends. How to effectively classify academic news data and obtain valuable information have become one of the important research directions of information science. Traditional classification methods have the problems of high dimensions, high sparseness and weak feature expression ability, etc. Deep neural network models such as CNN and RNN are also often affected by their own parameters. In this paper we present a deep neural network model based on attention mechanism and RCNN (ARCNN). We capture the context of each word and generate the word vectors with deep bidirectional LSTM layers after preprocessing. Then we use attention mechanism to calculate the attention probability distribution of news titles and contents, effectively highlighting key information. In our experiments, we use news data of academic social network SCHOLAT and Fudan University document classification set to evaluate our model and achieve better results than other widely used text classification algorithms.