Sentiment Bias Detection in Support of News Credibility Judgment

Recently, an increasing number of online news websites have come to provide news browsing and retrieval services. For certain topics, certain news websites may hold sentiment bias, and therefore select and edit information according to their own standpoints before delivering news articles. Lacking conscious awareness of websites' sentiment bias may result in blind obedience to the reported information. We focus on the sentiment aspect of news articles and develop a system which can detect and visualize sentiment tendencies of different websites. Given a topic, the system extracts relevant subtopics and presents sentiment difference between different subtopics. Once a subtopic is specified, sentiment difference between news websites is also provided. The background knowledge of sentiment difference between subtopics and between websites can assist users in judging the news credibility. In particular, the system analyzes four-dimension sentiment, which is more similar to human emotion than conventional positive-negative sentiment. Experimental evaluations show the accuracy of sentiment extraction and subtopic extraction is good, and our observation results show sentiment bias can be detected by the system.