Topic-sentiment evolution over time: a manifold learning-based model for online news

Topic and sentiment detection has been considered as an effective method to reveal the facts and sentiments in a massive volume of information. Existing works mainly focus on separate topic and sentiment extraction or static topic-sentiment associations, neglecting topic-sentiment dynamics and missing the opportunity to provide a in-depth analysis of online news. Actually, sentiment orientations are highly dependent on topic content and thus detecting topic-sentiment associations and their evolution over time is very important. This paper proposes a manifold learning-based model to explore the topic-sentiment associations and their evolution over time in the online news domain. The proposed model can visualize the hidden sentiment dynamics of topics in a low-dimensional space. Extensive experiments are conducted on online news crawled from the American Cable News Networks (CNN) website. The experimental results show that the proposed model outperforms the KL distance-based and the Similarity-based methods and improves the accuracy of topic classification by 12 % .