A Bootstrapping Approach to Assessing the Bias of Facebook’s Graph API
Justin Chun-ting Ho
Changes in Graph API
- On 7 November 2017,
Facebook introduced a new limitation
- The API will only return a maximum of
“600 ranked, published posts per year”
- How these posts are selected is currently unknown
Is it biased?
Spoiler Alert: It is.
- Scottish National Party’s official Facebook page
- All posts between 1 January - 31 December 2016
- Two Rounds of Data Collection:
7 May 2017 (1031 posts) &
6 April 2018 (598 posts)
- Extracted using Netvizz
Top Terms Analysis
- Kendall's Tau
- Top terms (n) between 10 and 1500 in steps of 10
- Compare: Full Data vs Bootstrap Samples &
(Morstatter et al. 2013)
Full Data vs Partial Samples
Reverse Engineering the Ranking Method
Sentiment and Emotion Scores
Is it the end of the world?
- If you research on information diffusion...
- If you analyise user engagements...
Misrepresentation of user engagements
- If you analyise post texts...
Bias in top terms
- If you research on public opinion...
Over-representation of sentimental terms