The digital age has revolutionized science communication, providing unprecedented reach through platforms like YouTube. However, the competitive nature of social media necessitates understanding how to effectively engage the public. Traditional peer-to-peer science communication differs significantly from social media communication. Social media users tend towards superficial content consumption, highlighting the importance of surface-level aspects like language. Furthermore, indirect sharing via recommender systems amplifies engaging content, creating positive feedback loops. This study focuses on affect as a key element influencing engagement, building upon previous research showing a correlation between affect-rich language and increased sharing or citations in other media. TED Talks, widely viewed as a modern form of science communication with a lay audience, provide a valuable dataset for analyzing the link between affect and engagement on YouTube, a gap in current literature. The research questions are: How is affect used in TED Talks compared to other science communication media? And, is affect associated with audience engagement on YouTube?

Prior research indicates that affect-rich language in the New York Times correlates with higher email sharing rates. Similarly, studies suggest that affectively described scientific findings are more likely to be shared and cited. However, the systematic investigation of the affect-engagement-dissemination link in social media-based science communication has been lacking. Existing research on YouTube science communication has focused on presenter characteristics (gender, background, authenticity) and viewer psychological processes (eye movements, comment analysis), but not the affective features of the scientific content itself. This study addresses this gap.

The study analyzed 2962 TED Talks published on YouTube between early 2007 and the end of 2020. Data included transcripts from TED.com and engagement metrics (views, likes, dislikes, comments) from the YouTube API. Talks were matched based on title, using exact and approximate string matching. To identify scientific content, a semantic network analysis of co-occurring tags was used to group talks into seven topics (Mind, Entertainment, Tech, Health, Cosmos, Environment, Society). A science index was calculated for each topic based on the presence of science-related tags or words. Sentiment analysis was performed using the SentiWordNet dictionary, calculating valence (overall positivity/negativity) and density (frequency of affect-laden words). Principal component analysis was used to reduce the four engagement variables into two components: popularity (views and likes) and polarity (dislikes and comments). Regression analyses were conducted to assess the relationship between valence, density, and engagement, controlling for topic, publishing date, video duration, and readability (Flesch Reading Ease score). Moderation analyses were conducted to investigate the influence of talk topic and science index on the affect-engagement relationship.

The study found that TED Talks exhibit higher affective valence and density compared to text-based media but lower than video-based media. Valence and density significantly predicted engagement. Higher valence was associated with increased popularity and decreased polarity. Higher density was associated with increased popularity across almost all topics. The effects of affect on engagement were substantially moderated by talk topic, but not significantly by the inclusion of scientific content. Specifically, the relationship between density and popularity was attenuated for Environment talks (especially those with “Green” or “Sustainability” tags) but enhanced for Mind talks (especially those with “Decision-Making” or “Mental Health” tags). For Society talks, valence was more strongly related to popularity (especially those tagged with “Immigration” or “Refugees”). Conversely, in Health-related talks (especially “Medicine” or “DNA”), valence showed a weak negative relationship with popularity. Regarding polarity, density showed a positive relationship in Tech (“AI”, “Machine Learning”) and Environment (“Green”, “Sustainability”) talks. For Society (“Refugees”, “Criminal Justice”), density showed a weak negative relationship with polarity. Moderation by scientific content was minimal.

The findings suggest that affect significantly influences public engagement with scientific content on social media. Two potential explanations are proposed: affect influences audience emotional state (mood, arousal) and signals strong opinions, thus increasing engagement (positive or negative). Moderation by topic highlights the importance of considering content context; the effectiveness of increasing valence and density may vary depending on audience expectations and topic sensitivity. The minimal moderation effect of the science index suggests that this link between affect and engagement holds regardless of whether the content focuses on science.

This research demonstrates a significant association between affective content in TED Talks and engagement on YouTube. While a correlational design limits causal inference, it suggests that science communicators can leverage affect to enhance engagement. Increasing density and positive valence can generally improve popularity, but this effect is moderated by topic. Future experimental work is needed to confirm these findings and investigate causal mechanisms. The study also highlights the need for scientists to become fluent in multiple forms of science communication beyond traditional publications to effectively reach wider audiences.

The study's correlational design limits causal inferences. The findings may not generalize to all forms of public science communication (e.g., academic social media posts, press releases). The dataset is limited to TED Talks, which have a unique format and audience. Finally, the absence of “shares” data in the engagement metrics may have influenced the results.