Abstract
This research presents a false comment identification method based on
rolling-type collaborative training. False comments pose a significant challenge in
online platforms, impacting credibility and user experiences. The proposed method
effectively utilizes unlabeled samples to assist model learning and integrates multiple
characteristics, including emotion and text representation, to enhance the identification
performance. The method involves obtaining comment text and determining its content
characteristics, as well as obtaining reviewer information and determining their
behavior characteristics. By combining these characteristics, the method performs false
comment identification and outputs the identification result. Experimental results show
that the proposed method achieves a 3.5% improvement in accuracy compared to
traditional methods. The rolling-type collaborative training approach demonstrates the
potential to enhance the reliability of comment evaluation systems and combat the
spread of false information in online platforms.