Composite text multi-classification method and system based on deep learning model

Abstract

The invention belongs to the technical field of natural language processing, and particularly relates to a composite text multi-classification method and system based on a deep learning model, and the method comprises the steps of converting a composite text into word granularity level text representation, preprocessing the converted word granularity level text representation, and representing the word granularity level text representation as word vectors through a word embedding method; taking word vectors as deep learning model input, extracting text features through a model CNN convolution layer, selecting vectors after convolution, reserving global partial sequence correlation information, adding weights to the text feature vectors through a model self-attention layer, performing equal-length vector sequence splicing, and extracting text feature vectors through a model LSTM circulation layer; and carrying out averaging operation on the text feature vector through a model pooling layer, obtaining an input text category probability by utilizing a softmax classifier, and obtaining a text classification result according to the probability. According to the invention, the precise multi-classification problem of the composite text can be solved, and the practical application of natural language text multi-classification prediction recognition is met.

Description

Technical field [0001] The present invention belongs to the technical field of natural language processing, and in particular, the multi-class method and system based on the depth learning model, multi-class prediction of the text through the composite depth learning model CNN, LSTM, and Self-Attension. Background technique [0002] Text Classification is one of the key and foundation tasks in natural language processing. Its common ways have traditional machine learning classification models such as simple Bayes, support vector machines, logic regression, etc., and evolving to a series of deep learning based on deep learning. Classification model, mainly including CNN, LSTM, Attention, etc. [0003] The Text-CNN is simple to support parallel, and its main features are local perception and weight sharing. Local Perception Make the model only requires only a small partial area, and reduces the number of participating capabilities while reducing the number of participation. Weight ...

AI Technical Summary

Problems solved by technology

In the practical application of text classification, the CNN model is simple, the training speed is fast, and the effect is considerable, but the interpretability is not strong. When tuning the model, it is difficult to adjust specific features according to the training results, and it is not easy to adjust the importance of each feature. Evaluation; the Attention mechanism can ignore the examples between words and directly calculate the dependency relationship, and can learn the internal structure of the sentence, which is simple and can be calculated in parallel. However, it is necessary to calculate the correlation between each feature vector when performing weight calculation. When the feature vector is relatively The amount of calculation is high for many times; LSTM can effectively process sequence information, but it is not intuitive enough and lacks interpretability

Images