The Research And Implementation On Text Classification Algorithm Applied For Aerospace Intelligence

In recent years,it has made remarkable achievements for China's aerospace industry.It would not develop rapidly without the support of aerospace intelligence for China's aerospace industry.The foreign aerospace intelligence information has important significance for the development of China's aerospace industry.However,the traditional classification of aerospace intelligence uses manual processing method,its processing efficiency is extremely low and it has been unable to adapt to current aerospace intelligence work.The modern intelligence processing has formed an information-centric processing model which is able to classify data automatically,model standardized and study scientifically.Therefore,how to classify,store,analyze aerospace intelligence information and standardize the work flow of the entire intelligence tracking research,and realize the automated classification and management of aerospace intelligence data has become an important issue of the current aerospace intelligence processing.In this thesis,based on a brief introduction to the status quo of knowledge management and text classification technology in the field of aerospace intelligence,the traditional text classification technology which includes the K-nearest neighbor algorithm and the support vector machine algorithm is first described,and then the deep neural network based approach is introduced.The deep learning text classification technology is introduced,which includes the text classification technology based on the recurrent neural network and the text classification technology based on the convolutional neural network,finally we draw on previous research results and summarize the advantages and disadvantages of different text classification methods.A Text RCNN-A text classification algorithm based on the Attention mechanism is proposed and designed.The features of this algorithm are as follows:(1)For word representation learning,this thesis proposes a convolutional neural network model based on a bidirectional recurrent structure.Compared with the conventional neural network model using only a fixed window,this method can accurately capture contextual information,and can also better eliminate the ambiguity of the word;(2)For text representation learning,this thesis designs an attention pooling-based convolutional layer.It could effectively reduce information loss.The attention mechanism can reasonably allocate emotion weights of contextual words and further improve the classification accuracy of the model.The time complexity of this pooling layer is O(n).The overall model is a cascade of recurrent structure and the pooled layer.Therefore,the time complexity of our model is still O(n).Text RCNN-A text classification algorithm not only has the advantage of recurrent neural network and convolutional neural network,but also has new features which is reasonable about emotional weight distribution for contextual words.By comparing and analyzing the experiments result of text classification with other models on five public data sets,the Text RCNN-A text classification method designed in this thesis has significantly improved classification accuracy compared with other methods.Finally,the algorithm is applied to the field of aerospace intelligence,and a aerospace intelligence classification system based on Text RCNN-A algorithm is designed and implemented.The system has improved the accuracy of aerospace intelligence information classification,improved the efficiency of aerospace intelligence work and reduced the workload of aerospace intelligence staff.