Discrimination of News Political Bias Based on Heterogeneous Graph Neural Network

The polarization of western political ideology is becoming more and more serious, and it is difficult for news articles to keep objective justice. Their contents are often biased towards a particular political party to guide the trend of public opinion. Therefore, judging the political inclination of news texts is of great significance to national election prediction and public opinion control. The existing modeling methods based on news content mostly rely on the semantic information of news. The combination of various element features and structural information in the news is insufficient. This paper proposes a political bias discrimination method of news based on a heterogeneous neural network, with multiple information related to prejudice in the news as the nodes of a heterogeneous network. By enriching the representation of nodes through a heterogeneous graph neural network and using the fused node features to distinguish the political news bias. The experimental results show that our model can achieve 84.30% accuracy and 83.34% Macro F1 value in news political bias classification. Compared with the baseline model Bert +CNN with the best experimental results, the accuracy and Macro F1 are improved by 1.92% and 0.4%, respectively.