Design of Intelligent Document Categorization System for Office Software Combined with Neural Networks

Abstract

Office software documents as a carrier of effective information; the complexity of its text determines the efficiency of information extraction. The current research hot spot involves accurately categorizing the document text. In this paper, the document data are processed by document cut, text segmentation, de-duplication, text feature extraction, word frequency statistics and other data processing, and the Word2Vec model is used to represent the text of office software documents. The improved CNNSVM text classification model was constructed by replacing the classifier and fused the attention mechanism module based on the convolutional neural network. In this way, a document categorization system based on a neural network is designed. On the CR dataset, the model in this paper was trained for 25-90 rounds later, and the loss value, recall rate, accuracy rate, and F1 value converged to about 0.1, 0.91, 0.85, and 0.88 in turn. And the accuracy rate of this dataset has been improved by 18.71% when compared to the worst comparison model. The attention module can be used to display different text sequence weights in the model in this paper to correctly classify text. The above experimental results fully demonstrate the superior performance of the model in this paper and its high matching on multiple datasets.