Algorithmic or Human Source? Examining Relative Hostile Media Effect With a Transformer-Based Framework

The relative hostile media effect suggests that partisans tend to perceive the bias of slanted news differently depending on whether is in favor or against their sides. To explore an algorithmic vs. human source perceptions, this study conducts a 3 (author attribution: human, algorithm, human-assisted algorithm) x (news attitude: pro-issue, neutral, anti-issue) mixed factorial design online experiment (<em>N</em> = 511). This uses transformer-based adversarial network auto-generate comparable headlines. framework was trained with dataset 364,986 stories from 22 mainstream outlets. results show occurs when people read headlines attributed all types authors. News sole perceived as more credible than two algorithm-related sources. For anti-Trump headlines, there exists interaction between author attribution and issue partisanship while controlling for people’s prior belief machine heuristics. difference perceptions partisan groups relatively larger compared pro-Trump