Effects of the Inclusion of Non-newsworthy Messages in Credibility Assessment

Social media has become influential and affects large public perception. Anyone can post and share messages on social networking sites. However, not all posts are trustworthy. Many online messages contain misleading or false information. There has been an extensive research to assess the credibility of social media data. Previous studies evaluate all online messages, which may be inappropriate due to a large amount of such data that can result in ineffectiveness of the system. This paper studies and presents the effects of the inclusion of such data—namely, non-newsworthy messages—in credibility assessment. Our findings affirm a negative effect of training a model with non-newsworthy data. The degree of performance degradation is also shown to have a strong connection to a degree of non-newsworthiness in training data.