Deep theme model-based large-scale text classification method

Abstract

The invention discloses a deep theme model-based large-scale text classification method, and mainly solves the problems that the gradient cannot be solved and the gradient update step length of each layer cannot be adjusted in the prior art. The method mainly comprises the steps of 1, establishing a training set and a test set; 2, setting a Poisson-gamma belief network and network parameters, and performing gradient update training on the network by using the training set in the step 1; 3, storing the trained network global parameters; 4, initializing global parameters of a test network by using the global parameter values stored in the step 3, and training the test network by using the test set in the step 1; 5, storing network parameters of the test network; and 6, classifying texts by utilizing the test network parameters obtained in the step 5, and outputting and predicting text types and text classification accuracy. According to the method, multilayer information of text information can be extracted; the problems that the gradient is difficultly solved and the step length of each layer cannot be adjusted are solved; and the method can be used for text information extraction and text classification.

Description

technical field [0001] The invention belongs to the technical field of text processing, and in particular relates to a large-scale text classification method of a deep topic model based on adaptive step size stochastic gradient training, which can be used for text data mining and classification. Background technique [0002] At present, the topic models used in the field of natural language processing are not only more and more, but also more and more mature, but there are differences in training methods and practicability. Although some shallow models are more practical, But it is not as good as the deep model in performance. For example, the commonly used LDA topic model is one of them. Although the performance of the deep topic model is improved compared with the shallow topic model, because of its complicated training, the training time is too long, especially When the data set is particularly large, it is lacking in practicability, such as the Poisson-Gamma Belief Netwo...

AI Technical Summary

Problems solved by technology

At present, the stochastic gradient method is mostly applied to the shallow model. For the deep model, the stochastic gradient method is rarely applied, not only because there is a large correlation between the layers of the deep model, the gradient calculation is very complicated; and in During the training process of the deep topic model trained by the stochastic gradient method, the gradient update step size between layers cannot be adjusted

Images