AMAE: Adversarial multimodal auto-encoder for crisis-related tweet analysis

Social media platforms have grown in importance as sources of information and as a complement to traditional media. If informative and relevant tweets on social media platforms can be effectively detected and analyzed during crisis events, it will help humanitarian organizations with situational awareness and planning relief activities. In this work, we propose an adversarial multimodal auto-encoder model for detecting and analyzing crisis-related tweets, which analyzes the complex multimodal content of tweets and integrates adversarial strategies to generate a joint representation containing information from multiple sources. Through extensive experiments on real datasets, we demonstrate the superior performance of the proposed model.