Research on Topic Fusion Graph Convolution Network News Text Classification Algorithm

Abstract

In the face of a large amount of news text information, how to make a reasonable classification of news text is a hot issue of modern scholars. To solve the problem that only word co-occurrence was considered in the Text Graph Convolutional Network (Text-GCN) method to build a graph model, a news text classification algorithm which fuses themes and is based on Graph Convolution Network, is presented. Firstly, the LDA topic model is used to process the corpus to obtain the distribution of themes of the corpus. Secondly, a graph model is built to construct a global map by using the related topic words and their subject distribution in each article. Finally, the text graph is input into the Graph Convolution Network layers to compute the learning representation of combining feature in order to complete the text classification task. The experimental results show that this method can effectively realize the word level interaction of information in text. In the experiment on Chinese and English datasets, adding theme information improves the accuracy by 1% compared with the Text-GCN method.

Research of news text classification method based on hierarchical semantics and prior correction

“…To do this, the label predictions are sorted in descending order based on their probabilities. For simplicity, the sorted probabilities are denoted as shown in the equation: (21) where n represents the number of labels, and p i represents the predicted score for the i-th label.…”

“…Depending on the different model embeddings, character level [17], word level [18] and sentence level [19] have been proposed. As research progresses, researchers have started exploring how to parse semantics using graph structures, including TextING [20] and [21]. With the development of pre-training, pre-trained models such as ERNIE [22,23] have gradually gained attention.…”