Text classification method and system

Abstract

The embodiment of the invention provides a text classification method and system. The method comprises the following steps: obtaining a to-be-classified text; inputting the to-be-classified text intoa text classification model, and obtaining the category of the to-be-classified text according to the output result of the text classification model; wherein the text classification model is formed based on a plurality of convolutional neural networks and a plurality of recurrent neural networks, and is obtained by training based on a sample text and a predetermined text category label. The embodiment of the invention provides a method and system. A CNN and an RNN are fused to form a text classification model; texts are classified through the text classification model; The method overcomes thedefect that the CNN is not suitable for extracting long sequence semantic information and the defect that the RNN has an extraction effect on long sequence global semantic information, and can be suitable for extracting local semantic information and long sequence global semantic information, so that the text classification accuracy is greatly improved.

Description

technical field [0001] Embodiments of the present invention relate to the technical field of natural language processing for artificial intelligence deep learning, and in particular, to a text classification method and system. Background technique [0002] Deep learning is the best way to achieve text classification tasks, among which Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) are the most widely used feature extraction methods in deep learning. [0003] CNN can be traced back to the proposal of the back propagation (BP) algorithm in 1987, and then Yann LeCun used it in multi-layer neural networks in 1989, until Yann LeCun proposed the LeNet-5 model in 1998, CNN gradually revealed prototype. CNN can be divided into three parts: input layer, hidden layer and output layer. In natural language processing, the input information of the input layer is represented as an input sequence by a pre-trained word embedding model or word vector. Commonly used...

AI Technical Summary

Problems solved by technology

Recurrent neural networks are slow to train because the hidden layers are recurrent

Moreover, in long sequences, as time goes by, the features extracted by the hidden layer are more biased towards the later input information, which will affect the extraction effect of RNN on global semantic information, while the features of each part of the input information extracted in CNN are equal. or no preference

[0005] In summary, CNN is only good at extracting local semantic information, and is not suitable for extracting long sequence semantic information. RNN is suitable for analyzing long sequence global semantic information, but as time goes by, the features extracted by the hidden layer are more inclined to the later input information, which affects The effect of extracting global semantic information

Images