Infrastructure for Tracking Information Flow from Social Media to U.S. TV News - CAPWIC '25

This study examines the intersection between social media and mainstream television (TV) news with an aim to understand how social media content amplifies its impact through TV broadcasts. While many studies emphasize social media as a primary platform for information dissemination, they often underestimate its total influence by focusing solely on interactions within the platform. This research examines instances where social media posts gain prominence on TV broadcasts, reaching new audiences and prompting public discourse. By using TV news closed captions, on-screen text recognition, and social media logo detection, we analyze how social media is referenced in TV news.

1. Infrastructure for Tracking Information Flow from Social Media to U.S. TV News Web Science & Digital Libraries Research Group Old Dominion University, Norfolk VA, USA @WebSciDL , @oducs Presented By: Himarsha R. Jayanetti @HimarshaJ ‹#› CAPWIC 2025, George Washington University in Washington, D.C, USA, March 29, 2025 Advisors: Dr. Michele C. Weigle, Dr. Michael L. Nelson, and Dr. Alexander Nwala

2. TV Provides Reach Beyond Social Media Subscribers ‹#› https://truthsocial.com/@realDonaldTrump/posts/110833185720203438 https://archive.org/details/MSNBCW_20230808_010000_ The_Rachel_Maddow_Show/start/1980/end/2040 @realDonaldTrump post on Truth Social The same post on MSNBC’s The Rachel Maddow Show Truth Social - 2M monthly active users ( as of Sep, 20 23 ) As of Nov 13, 2023, @realDonaldTrump - 6.48M Followers Post - 17k ReTruths and 49.9k Likes MSNBC - 1.22M average primetime viewers ( as of Sep, 2023 ) The Rachel Maddow Show - 3.9M viewers ( as of Aug, 2023 )

3. ‹#› Median Ages of TV Viewers vs. Social Media Users: At Least a Generation Apart Facebook 40-50 yrs Instagram 28-34 yrs LinkedIn 30-49 yrs Twitter (X) 30-40 yrs YouTube 30-40 yrs TikTok 20-25 yrs Snapchat 20-25 yrs Median Age Breakdown of Social Media Users Median Age Breakdown of C able TV Viewers MSNBC 70 yrs Fox News 69 yrs CNN 67 yrs A ccording to Nielsen data, Article by VideoAmp , May 2024 Approximation based on data from Pew Research Center's Social Media Fact Sheet (survey of U.S. adults, May 19–Sept 5, 2023)

4. RQ1: What is the extent (and nature) of social media content shared on local and national TV news? RQ2: How can we quantify the amplification effect resulting from TV news coverage of social media messages? ‹#› Research Questions

5. Internet Archive TV News Offers Access to Clips From 2.61M US Broadcast Shows Dating Back to 2009 ‹#› https://archive.org/details/tv https://archive.org/details/tv?q =“ truth%20social” We have the ability to query the clips using closed captions!

6. Social Media Can Be Mentioned in TV News Across Various Contexts ‹#› Stories ABOUT social media Using social media posts for EVIDENCE/CONTEXT And others, like mentioning social media in passing and interactive campaigns!

7. ‹#› Stories referencing a single or multiple social media POSTS Mentions Could Refer to Specific Posts or the Platform in General Stories referencing a single or multiple social media PLATFORMS

8. ‹#› Verbal Mentions by the Host or Other Speakers on TV On-screen Text Using OCR (Optical Character Recognition) Social Media Can Be Mentioned in TV News in Different Representation s Visual Cues (Logos and Screenshots of Social Media Posts) social media tweeting social media site x Keywords such as platform name and other social media lexicon can be used to detect social media mentions using on-screen text or verbal mentions

9. ‹#› Display of Social Media Post on Screen Without Verbal Commentary From the Anchor Social media screenshot is up on the screen. However, the platform name is not explicitly mentioned by the anchor during the news reporting. Only visual cue, unable to detect this easily just by using closed captions Truth Social Twitter/X

10. ‹#› Presentation of Idealized Social Media Post on Screen, Lacking On-Screen Cues—Identified Solely by Host's Verbal References https://archive.org/details/PRESSTV_20230926_043200_Palestine_Declassified__ADL_takes_on_Elon_Musk/start/780/end/840

11. ‹#› Blue Sky is → Blue Skies is TikTok → tick, tock Errors in Closed Captions May Result in Words Conveying Unintended Meanings

12. ‹#› Logo is a Huge Clue, but Not Always! A TikTok video on screen, but the logo of TikTok is not on the screen Twitter/X screenshot is posted on Instagram, both the Twitter/X logo as well as the Instagram logo is present on the screen

13. ‹#› Platform Logos Evolve Over Time https://www.logome.ai/blogs/twitter-logo-evolution

14. ‹#› Each Platform Has Its Own Official Logo Versions https://about.meta.com/brand/resources/facebook/logo/ https://about.twitter.com/content/dam/about-twitter/en/brand-toolkit/downloads/twitter-external-brand-guidelines-01272021.pdf https://about.x.com/en/who-we-are/brand-toolkit

15. ‹#› People Often Modify and Adapt Logos by Compromising Visual Integrity

16. ‹#› Methodology Overview: Subtasks Breakdown

17. ‹#› Building the Context Classifier to Automatically Detect the Specific Context of a Social Media Reference in TV News

18. ‹#› Metadata of Social Media References of Prime Time TV News on MSNBC and FOX NEWS Date: Monday, September 18, 2023 Time: Prime time (8:00 PM to 11:00 PM EST) TV Station: MSNBC and FOX

19. ‹#› Manually Labeled Dataset: 20 Hours of Prime Time TV News on MSNBC, FOX News, and CNN https://docs.google.com/spreadsheets/d/1XiQFWi1pXyELtR5kymnFtspaDAXyI9iqeJ2A3ZN-je4/edit?usp=sharing

20. ‹#› Social Media Logo Detection (SMLD) model YOLOv7: https://doi.org/10.1109/CVPR52729.2023.00721 Test data: https://github.com/himarshaj/phd_dissertation_dc/tree/main/onscreen_logos/YOLOv7/results/test_data Dataset (140 Images): Train: 98 (70%) Test: 21(15%) Validation: 21(15%) Dataset after Image Augmentation ( 336 Images) : Train: 294 Test: 21 Validation: 21

21. ‹#› Social Media Logo Detection (SMLD) model YOLOv7: https://doi.org/10.1109/CVPR52729.2023.00721 Test data: https://github.com/himarshaj/phd_dissertation_dc/tree/main/onscreen_logos/YOLOv7/results/test_data

22. ‹#› Social Media Logo Detection (SMLD) model YOLOv7: https://doi.org/10.1109/CVPR52729.2023.00721 Test data: https://github.com/himarshaj/phd_dissertation_dc/tree/main/onscreen_logos/YOLOv7/results/test_data

23. ‹#› Social Media Logo Detection (SMLD) model YOLOv7: https://doi.org/10.1109/CVPR52729.2023.00721 https://github.com/himarshaj/phd_dissertation_dc/tree/main/onscreen_logos/YOLOv7/results/test_data mAP - mean Average Precision IoU - Intersection over Union Model performs well when a 50% IOU threshold. Model’s performance slightly decreases when a stricter range of IoU thresholds is applied. Overall, the SMLD model shows a robust ability to detect and accurately identify social media logos in screenshots.

24. ‹#› Metrics Development for Quantifying TV News Amplification of Social Media Content Temporal, Geographical, and Population

25. ‹#› Quantifying Amplification from TV News: Temporal, Geographical, and Population Metrics Metric Description Sources Temporal The total time that a social media reference is covered across all TV news. Social media content detector and segmenter. Geographical The geographical region most likely to have seen the social media reference on TV news. Internet Archive’s TV news archive metadata, FTC, 3DL news, credible sources like Pew and news organizations' websites, and contacting the news source directly when needed. Population The audience size of a TV news station. Population data for counties, cities, towns, and TV news viewership fractions, along with proprietary data from media firms like Nielsen.

26. ‹#› Quantifying Amplification from TV News: Temporal, Geographical, and Population Metrics Metric Description Sources Temporal The total time that a social media reference is covered across all TV news. Social media content detector and segmenter. Geographical The geographical region most likely to have seen the social media reference on TV news. Internet Archive’s TV news archive metadata, FTC, 3DLNews, credible sources like Pew and news organizations' websites, and contacting the news source directly when needed. Population The audience size of a TV news station. Population data for counties, cities, towns, and TV news viewership fractions, along with proprietary data from media firms like Nielsen.

27. ‹#› Quantifying Amplification from TV News: Temporal, Geographical, and Population Metrics Metric Description Sources Temporal The total time that a social media reference is covered across all TV news. Social media content detector and segmenter. Geographical The geographical region most likely to have seen the social media reference on TV news. Internet Archive’s TV news archive metadata, FTC, 3DL news, credible sources like Pew and news organizations' websites, and contacting the news source directly when needed. Population The audience size of a TV news station. Population data for counties, cities, towns, and TV news viewership fractions, along with proprietary data from media firms like Nielsen.

28. Key Takeaways Evaluating social media reach via mainstream media discourse. I nternet Archive’s TV News Archive offers access to clips from 2.61M US broadcast shows dating back to 2009 which can be queries using closed captions. RQ1: What is the extent (and nature) of social media content shared on local and national TV news? Different contexts and representations. RQ2: How can we quantify the amplification effect resulting from TV news coverage of social media messages? Temporal, Geographical, and Population Challenges: Errors in closed captions. Logo detection issues, including deviations from the official logo, logo evolution, and inconsistent logo presence. Using social media posts for EVIDENCE/CONTEXT Internet Archive’s TV News Archive allows querying the clips using closed captions! Stories ABOUT social media