Text classification model and method based on reinforcement learning and capsule network

Abstract

The present invention relates to the technical field of natural language processing, text classification, and more particularly, to a text classification model and method based on reinforcement learning and capsule network. The invention adopts reinforcement learning Actor-Critic, capsule network CapsNet as the basic framework, the characteristics of the text information are extracted through thecapsule network, reinforcement learning is carried out to identify the relationship between the capsule layers. The invention introduces the reinforcement learning to learn the routing relationship between the capsule network layers and the capsule network to solve the task of multi-label classification in the text classification model. Using the advantages of capsule network in multi-label classification task, it can be applied to multi-label text classification task, so as to achieve better results; Using the mechanism of reinforcement learning error adjustment, a better way to link the routes is learned.

Description

technical field [0001] The present invention relates to the technical fields of natural language processing and text classification, and more specifically, to a text classification model and method based on reinforcement learning and capsule network. Background technique [0002] Feature learning is a basic problem in the field of artificial intelligence, especially in natural language processing, feature extraction is even more important. In natural language processing, text classification is a very basic and common process, which is very dependent on the learning process of features. Unlike the image field, the semantic logic of text is more difficult to capture and express, which makes the classification task on text more difficult. The basis for the realization of general artificial intelligence in natural language processing is that the machine must understand human language, that is, understand the semantic information of the text, so as to be able to perform the pres...

AI Technical Summary

Problems solved by technology

1. The bag-of-words feature model is a text feature representation method that does not consider the order of the words in the sentence. It encodes the words in the sentence. The length of the feature vector of the words in the sentence is the size of the bag of words. For example, Mohit et al. proposed The DAN model, which splits and marks the words in a sentence, and then passes them into the neural network structure, these marked words do not retain the original position information; the fasttext model proposed by Joulin et al., which directly passes the words through a lookup table, and add a neural network model, without taking into account the order information of words

2. The sequence representation model is a type of model that considers the order of words, such as Convolutional Neural Network, Recurrent Neural Network, etc., but one of its shortcomings is that it does not use the information of sentence structure

The model based on the attention mechanism has a relatively large advantage in obtaining the corresponding relationship between input and output, but the text model based on the attention mechanism does not take into account the information of word order order, which is very bad for natural language, because the text word order contains great information

Images