Hoax and Posture Detection, Reply Weight Estimation in Social-Media using Cognitive Contents with Hyper Meta Network Approach

Rumor (Hoax) and stance (Posture) detection, and reply weight estimation in social media were becoming important due to developments in the social network platforms, guaranteeing authenticity in posts, and increasing the scale of social media data. In previous work, hoax detection was implemented using a multi-task learning mechanism (MTLM) for performance improvement and neglected three aspects such as focusing on the textual contents only, limited posture information used, and neglecting posture information at stance labels. This work designed a hyper meta-network approach for reply, hoax, and stance detection in social media by considering textual and visual content. A meta-knowledge-sharing scheme is used to solve problems raised in hoax and stance detection. The attention mechanism is used in estimating the reply weight of each one for better utilization of posture information hidden in the posture labels. On implementation of the proposed approach with different Twitter-based datasets demonstrates that the method achieved in updating models.