Reimagining Historical Credibility : A Semi-Supervised Computational Analysis of User Comments on TikTok

Sammanfattning

As short-form video platforms become central to how audiences encounter historical information, the credibility frameworks developed in earlier digital media contexts may no longer fully account for how users evaluate such content. This thesis investigates how credibility heuristics can be theoretically adapted to TikTok’s historical content, addressing two questions: how heuristic cues are empirically expressed through observable signals in user comments on TikTok history videos, and how they differ from those identified in pre-algorithmic credibility frameworks. Using a mixed-methods design that combines a SetFit-based semi-supervised classification with close reading, the study analyses a corpus of comments on TikTok historical videos within an eight-category heuristic framework. Three main findings emerge. First, observable credibility evaluation is largely rebuttal-driven, while implicit acceptance operates through opaque cognitive processes. Second, algorithmic curation systematically amplifies commenter-level credibility performance, emotional engagement, and identity-aligned responses over rational evaluation of the video’s historical claims. Third, the presentation form and scenarios of the video content shape whether credibility-evaluative work is activated. In sum, the stable cue-credibility mapping assumed by pre-algorithmic frameworks no longer holds in this context, which raises challenges for the development of an adaptive framework. By examining credibility heuristics in a cultural-information context, this thesis moves beyond the political bias and medical misinformation domains that have dominated prior research, and by characterizing how algorithmic mediation reshapes the reception of historical content, the thesis contributes empirical and methodological insights that can guide future research on credibility evaluation in algorithmically curated environments.