Jointly Trained Convolutional Neural Networks for Online News Emotion Analysis

• Published in: Web Information Systems and Applications Vol. 11242; pp. 170 - 181
• Main Authors: Zhao, Xue, Zhang, Ying, Guo, Wenya, Yuan, Xiaojie
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2018; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: CNN; Emotion analysis; Joint training
• ISBN: 3030029336; 9783030029333
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-02934-0_16

Emotion analysis, as a sub topic of sentiment analysis, crosses many fields so as philosophy, education, and psychology. Grasping the possible emotions of the public can help government develop their policies and help many businesses build their developing strategies properly. Online news services have attracted millions of web users to explicitly discuss their opinions and express their feelings towards the news. Most of the existing works are based on emotion lexicons. However, same word may trigger different emotions under different context, which makes lexicon-based methods less effective. Some works focus on predefined features for classification, which can be very labor intensive. In this paper, we build a convolutional neural network (CNN) based model to extract features that can represent both local and global information automatically. Additionally, due to the fact that most of online news share the similar word distributions and similar emotion categories, we train the neural networks on two data sets simultaneously so that the model can learn the knowledge from both dataset and benefit the classification on both data sets. In this paper, we elaborate our jointly trained CNN based model and prove its effectiveness by comparing with strong baselines.