Reviewing the Literature on Social Media Sharing Biases

The next ECREA 2024 session is also on polarisation, and I’m chairing as well as blogging it. We start with Petra de Place Bak, whose interest is in the cognitive preferences that make specific types of online content more salient and shareable. One aspect of this might be sentiment- and emotion-based biases.

Petra’s focus is on social media communication, which has to address the twin challenges of information abundance and attention scarcity; this is affected both by platform algorithms and users’ own cognitive preferences. Negative content biases can play a role especially in the latter, as can biases (both positive and negative) in favour of emotive stimuli. This is true both for question of attention and (re)transmission – but in transmission the social context also moderates the degree of negativity biases.

Such biases can be assessed at scale in large-data studies, coupled also with natural language processing tools; however, existing research points in very opposite directions: confoundingly, early studies pointed to both positivity biases and anger biases; later work pointed variously to negativity and positivity biases.

Petra has conducted a systematic literature review to better unravel these contradictions. Using keywords like ‘spread’, ‘emotion’, ‘social media’, etc., she identified some 55 relevant articles and systematically analysed their approaches and findings. These predominantly addressed Twitter, with some less focus on Weibo, Facebook, and TikTok; most gathered data based on keywords and hashtags. They also studied widely varying timeframes, from 24 hours to 14 days.

Analysis of such data used manual coding or natural language processing tools for binary or overlapping category coding; it variously found positivity and (more strongly) negativity biases) in content sharing decisions, and pointed especially to emotions like anger, fear, sadness, and disgust, as well as happiness, joy, anticipation, and trust.

Politics, health, and misinformation were the principal domains, and in politics there was a stronger negativity bias, while in health there was a greater balance between positivity and negativity; the few studies that examined a more general range of topics found greater positivity bias. It appears, then, that sharing decision-making biases are largely moderated by the topic domains in question.