A Deep Learning Framework for News Readers' Emotion Prediction Based on Features From News Article and Pseudo...

Area of Science:

• Natural Language Processing
• Computational Social Science
• Artificial Intelligence

Background:

• Online readers increasingly share emotions about news events, making emotion prediction crucial for social media applications like sentiment analysis and opinion summarization.
• News articles often lack explicit emotion words, posing a significant challenge for accurately predicting reader emotions solely from text.
• Reader comments are rich in emotional content, offering a valuable resource for understanding and predicting emotions associated with news.

Purpose of the Study:

• To develop a deep learning framework for predicting reader emotions by integrating news article content with reader comments.
• To address the challenge of implicit emotions in news articles by proposing a novel encoding model.
• To enhance the prediction of emotions for unpublished news by creating a pseudo comment representation.

Main Methods:

• A deep learning framework that merges information from news articles and reader comments to predict emotions.
• Development of a pseudo comment representation technique to infer emotions for news without existing comments.
• Introduction of the Block Emotion Attention Network (BEAN) with an emotion attention mechanism and hierarchical structure to effectively encode implicit emotions in news articles.

Main Results:

• The proposed framework, incorporating emotional comments and BEAN, significantly improves the accuracy of reader emotion prediction.
• BEAN achieves state-of-the-art performance on three public datasets, demonstrating its effectiveness in encoding implicit emotions.
• Experiments on four self-collected datasets confirm the benefits of using emotional comments and the BEAN model.

Conclusions:

• Integrating reader comments with news articles is a viable strategy for enhancing emotion prediction accuracy.
• The Block Emotion Attention Network (BEAN) offers a superior method for encoding implicit emotions within news articles.
• The developed framework and model provide a robust solution for predicting reader emotions in the context of online news consumption.