Leveraging Computational Tools to Understand the Dissemination of STEM Content on Social Media

Core Concepts

Computational analysis of social media engagement data can provide insights into how audiences interact with and disseminate STEM-related content, enabling content creators to optimize their strategies for broader reach and impact.

Abstract

The authors used open-source machine learning methods such as correlation analysis, clustering, regression, and sentiment analysis to investigate the dissemination of STEM content on social media platforms, particularly TikTok. They analyzed over 1,000 videos and associated metrics from 6 STEM-focused social media creators to gain insights into how audience engagement signals (likes, comments, bookmarks, shares) correlate with video views, and how the algorithms may treat content from newer creators differently.
The key findings include:

Likes had the strongest correlation with views, but all engagement signals were important, suggesting users utilize multiple signals to interact with content.
Clustering analysis revealed that a small subset of videos (7.4%) with very high view counts (>1 million) were predominantly from accounts with large follower bases (>500k), suggesting the creator's follower count may play a significant role in content dissemination.
For newer creators, likes were the only strong predictor of views, while other engagement signals had much weaker correlations, highlighting the importance of this single metric for these creators.
Analysis of internal creator data, such as video production time and viewer engagement metrics, showed that high-performing videos did not necessarily require significant time investment, and that viewers tend to watch videos for less than 27 seconds on average.
Sentiment analysis of user comments provided insights into the concepts and emotions that resonated most with the audience, with positive and neutral sentiments dominating.
The authors provide recommendations for fellow academics on how to effectively leverage social media to share STEM content, emphasizing the accessibility and potential impact of this approach, as well as the value of computational tools for understanding audience engagement.

Stats

"A single video has the potential to reach tens of millions of viewers, many videos only receive hundreds or thousands of views."
"Likes had the strongest correlation with views (r=0.95), demonstrating that the number of likes is the most predictive of views."
"Clusters 2-4, accounted for 7.4% of all analyzed videos, and were composed of videos with very high view counts (>1 million)."
"Cluster 1, which accounted for 92.6% of videos, only likes had a strong correlation (r2=0.86) with views as measured by linear regression."
"Bookmarks and shares are used ~10x less than likes, while comments are comparably rare and used ~100x less than likes."
"The average user will not spend longer than 27 seconds on a video, and that few videos are ever watched in full, especially as they become longer."

Quotes

"Our analysis allowed us to identify and visually represent any correlation between likes, comments, bookmarks, and shares with total views."
"Effectively, this helps new creators because they can best understand the impact of their work by looking at a single interest signal."
"While we all seek to have videos that reach millions of views, many, many videos are still seen hundreds to thousands of times, presumably by different individuals. When put in the perspective of seminar or classroom attendance, there are few rooms on any academic campus that could hold that many people. Hence, every video can matter and make a difference, especially to individuals that don't have access to academic campuses."

Key Insights Distilled From

Investigating the dissemination of STEM content on social media with computational tools

by Oluwamayokun... at arxiv.org 05-01-2024

https://arxiv.org/pdf/2404.18944.pdf

Deeper Inquiries

How can the insights from this analysis be applied to other social media platforms beyond TikTok to optimize the dissemination of STEM content?

The insights gained from the analysis of STEM content dissemination on TikTok can be applied to other social media platforms to optimize the reach and visibility of similar content. One key aspect is understanding the importance of various public engagement signals such as likes, comments, shares, and bookmarks. Content creators can use this knowledge to tailor their content strategy on platforms like Instagram, YouTube, and Twitter by focusing on generating high levels of engagement through these signals. By analyzing the correlation between these signals and views, creators can prioritize creating content that is likely to resonate with their audience and drive higher viewership.
Additionally, the clustering and regression analysis conducted in the study can be applied to other platforms to understand how videos are treated by the algorithms. By grouping videos based on engagement metrics and analyzing the relationship between these metrics and views, creators can gain insights into how the platform's algorithm prioritizes content. This information can help creators tailor their content strategy to align with the platform's algorithms, increasing the likelihood of their content being recommended to a wider audience.
Furthermore, the sentiment analysis approach used in the study can be extended to other platforms to better understand audience reactions. By analyzing the sentiment of comments and feedback, content creators can gain valuable insights into the emotional responses evoked by their content. This information can be used to refine content strategy, create more engaging and relatable content, and foster a deeper connection with the audience across different social media platforms.

What other factors, beyond the public engagement signals analyzed, might the social media algorithms consider in determining the reach and visibility of STEM-focused content?

In addition to public engagement signals like likes, comments, shares, and bookmarks, social media algorithms may consider several other factors when determining the reach and visibility of STEM-focused content. Some of these factors include:

Relevance and Quality of Content: Algorithms may prioritize content that is relevant to the user's interests, preferences, and search history. High-quality, informative, and engaging content is more likely to be recommended to a broader audience.

Consistency and Frequency of Posting: Platforms may favor creators who post content consistently and frequently. Regularly updating content can signal to the algorithm that the creator is active and engaged with their audience.

Audience Engagement and Interaction: Algorithms may take into account the level of audience engagement and interaction with the content, such as the duration of views, click-through rates, and shares. Content that generates meaningful interactions is more likely to be promoted.

Creator's Profile and Authority: The credibility, expertise, and authority of the content creator may influence how the algorithm ranks and recommends their content. Established creators with a strong following and engagement history may receive preferential treatment.

Trending Topics and Virality: Social media algorithms often prioritize content related to trending topics, current events, and viral trends. Content that aligns with popular themes and discussions may receive higher visibility.

By considering these additional factors, content creators can optimize their content strategy to align with the algorithms' preferences and increase the reach and visibility of their STEM-focused content on social media platforms.

How can the sentiment analysis approach be further developed to better understand the nuances of audience reactions and provide more actionable feedback for content creators?

To enhance the sentiment analysis approach for a deeper understanding of audience reactions and to provide more actionable feedback for content creators, several strategies can be implemented:

Contextual Analysis: Incorporate contextual analysis to understand the underlying meaning and context of comments. Natural Language Processing (NLP) techniques can be used to analyze sentiment in relation to specific topics, themes, or events discussed in the content.

Emotion Detection: Expand sentiment analysis to include emotion detection to capture a broader range of audience reactions. By identifying emotions such as joy, sadness, anger, or surprise, creators can gain insights into the emotional impact of their content.

Sentiment Trends: Track sentiment trends over time to identify patterns and fluctuations in audience reactions. By analyzing changes in sentiment, creators can adapt their content strategy to better resonate with their audience.

Audience Segmentation: Segment the audience based on sentiment analysis to understand how different groups of viewers respond to the content. This can help creators tailor their content to specific audience preferences and interests.

Feedback Integration: Integrate sentiment analysis feedback into content creation processes to iteratively improve content quality. By using audience sentiment as a guide, creators can refine their messaging, storytelling, and engagement strategies to better connect with their audience.

By implementing these advanced techniques and strategies, sentiment analysis can evolve into a powerful tool for content creators to gain deeper insights into audience reactions, enhance engagement, and optimize the dissemination of STEM content on social media platforms.