Researching Misinformation
0 likes30 views
The document discusses the state of research on misinformation, focusing on methodologies such as claim extraction, clustering, and analysis using data sets. It highlights the need for data transparency and best practices for governance in the digital age while addressing challenges in studying misinformation on closed messaging systems. The document also touches on the differing goals of academics and practitioners in fact-checking and moderation efforts.
![Image similarity/hashing
(e.g., PDQ, phash)
[[1 1 1 1 1 1 1 0]
[1 1 1 1 1 1 1 0]
[1 1 1 1 1 0 1 0]
[1 1 1 1 1 0 1 0]
[0 1 1 1 0 0 0 0]
[0 0 0 0 1 1 0 0]
[0 0 0 1 1 1 0 0]
[0 0 1 1 1 1 1 0]]
[1111111011111110111110101111101001110000000011000001110000111110]](https://image.slidesharecdn.com/2019-12-googleapacsummit-researchingmisinformation-191230065814/85/Researching-Misinformation-4-320.jpg)
![1 in 4 WhatsApp messages
contained multimedia content
Resende, et al. 2019. In WWW.
Examining Brazilian WhatsApp groups
Despite representing less than 1% of all
images shared, [known misinfo images]
appeared in 44% of the monitored groups
in the period of the election False](https://image.slidesharecdn.com/2019-12-googleapacsummit-researchingmisinformation-191230065814/85/Researching-Misinformation-10-320.jpg)
Researching Misinformation
- 1. Facilitated by Dr Scott A. Hale, Meedan Researching Misinformation: What's the state of research today? What are the opportunities, what is needed and how do we prioritize?
- 2. Not just determining whether content is “truthful” Problem motivation Data collection, analysis Method development Extracting and clustering claims Matching similar images or text segments Social science insights Confirmation bias, role of social information
- 3. Claim extraction and clustering Konstantinovskiy et al. (2018) arXiv:1809.08193
- 4. Image similarity/hashing (e.g., PDQ, phash) [[1 1 1 1 1 1 1 0] [1 1 1 1 1 1 1 0] [1 1 1 1 1 0 1 0] [1 1 1 1 1 0 1 0] [0 1 1 1 0 0 0 0] [0 0 0 0 1 1 0 0] [0 0 0 1 1 1 0 0] [0 0 1 1 1 1 1 0]] [1111111011111110111110101111101001110000000011000001110000111110]
- 6. Datasets
- 8. Building a data trust to ensure data is open for research Partnerships Harvard CyberLaw Clinic -Kendra Albert Wolfe and Hwang -Tim Hwang Digital Public -Sean McDonald Conduct in-depth research focused on usability, scalability, and security Establish best practices for a new type of data governance in the era of closed networks. Image source: Open Data Institute
- 9. Encrypted messaging systems How can we study / counter misinformation on closed systems? Discover and join groups with links on the web? –> Privacy concerns Operate “tip lines”? –> Need technology to scale (e.g., spam, match images and text; distribute work) CheckWhatsApp
- 10. 1 in 4 WhatsApp messages contained multimedia content Resende, et al. 2019. In WWW. Examining Brazilian WhatsApp groups Despite representing less than 1% of all images shared, [known misinfo images] appeared in 44% of the monitored groups in the period of the election False
- 11. Hale et al. “How digital design shapes political participation: A natural experiment with social information” https://doi.org/10.1371/journal.pone.0196068 Social influence
- 12. Toward Automated Fact-Checking: ...Detecting Check-worthy Factual Claims… ...Developing an Annotation Schema...for Claim Detection
- 13. Academic vs. practitioner goals Academic: Build a system to identify claims needing to be fact-checked / moderated Build a system to do fact-checking / moderation
- 14. Academic vs. practitioner goals Academic: Build a system to identify claims needing to be fact-checked / moderated Build a system to do fact-checking / moderation Practitioner: Wants to select claims to fact-check based on organisational context, own knowledge/expertise, etc. Needs assistance prioritising claims (based on reach, impact, domain, type) Needs assistance managing pool of experts, workload, etc.
- 15. Bridging the gap Credibility Coalition (CredCo): https://credibilitycoalition.org/ Working groups on UX, InfoSec, Images & Memes, Health, Literacy,.... Content Moderation Group (Como)