A Deep Architecture for Sentiment Analysis of News Articles

In this paper, we present a deep architecture to perform aspect-level sentiment analysis for news articles. We combine some neural networks models proposed in various deep learning approaches, aiming at tackling specific issues commonly occurring for news articles. In this paper, we explain why our architecture can handle typically-long and content-specific news articles, which often cause overfitting when trained with neural networks. Moreover, the proposed architecture can also effectively process the case when the subject to be analyzed sentimentally is not the main topic of the concerned article, which is also a common issue when performing aspect-level sentiment processing. Experimental results with real dataset demonstrated advantages of our approach as compared to the existing approaches.