- Collected 4,644 notes and 12,761 ratings on Birdwatch from January 23 to March 2021
- Used text mining to obtain note contents
- Used Twitter API to get the account name, followers, followees, account age, and verified status for tweets referenced on Birdwatch

Community-Based Fact-Checking on Twitter's Birdwatch Platform - Data Collection

- Misleading – if a tweet was seen as misleading from Birdwatch notes (1 = yes; 0 = otherwise)
- Text explanation – why a tweet is misleading or not (280 characters max)
- Trustworthy sources – “Did you link the sources you believe most people would consider trustworthy” answered by user writing the Birdwatch note (1 = yes; 0 = otherwise)
- HVotes – number of users who find the Birdwatch note helpful
- Votes – number of ratings on a Birdwatch note (could be voted either helpful or unhelpful)

Community-Based Fact-Checking on Twitter's Birdwatch Platform - Key Variables

- Sentiment – used NRC emotion lexicon that categorizes words into positive or negative emotions and the difference between the positive and negative emotion scores is the final sentiment score 
- Text complexity – used Gunning-Fog index that estimates how many years of schooling is required for a user to understand the explanation, with a larger number indicating higher complexity
- Word count – length of Birdwatch note explanations determined by word count 

Community-Based Fact-Checking on Twitter's Birdwatch Platform - Content Characteristics

- Data collection
- Key variables
- Content characteristics

University of Michigan - Ann Arbor

Researchers studied how Twitter users engage with the new feature Birdwatch that allows users to address misinformation. 
- RQ 1: What are specific reasons due to which Birdwatch users report tweets?
- RQ 2: How do Birdwatch notes for tweets categorized as being misleading vs. not misleading differ in terms of their content characteristics? (e.g. sentiment, length)
- RQ 3: Are tweets from Twitter accounts with certain characteristics (e.g. politicians, accounts with many followers) more likely to be fact-checked on Birdwatch?
- RQ 4: Which characteristics of Birdwatch notes are associated with greater helpfulness for other users? 
- RQ 5: Does the level of consensus between users vary depending on the level of social influence of the source tweet? 

Community-Based Fact-Checking on Twitter's Birdwatch Platform

Pröllochs, N. (2021). Community-Based Fact-Checking on Twitter's Birdwatch Platform. arXiv preprint arXiv:2104.07175. https://arxiv.org/abs/2104.07175 

Birdwatch first started on January 15, 2021 and is run by the Twitter community where any user can identify misinformation, write notes to explain why those tweets are misleading, and rate the suggestions made by other users. Currently, Birdwatch is in the testing phase and only available through the Birdwatch website. 

Community-Based Fact-Checking on Twitter's Birdwatch Platform - Birdwatch

- Birdwatch notes (RQ 1 & RQ 2)
- Source tweets (RQ 3)
- Ratings (RQ 4)
- Helpfulness (RQ 4 & RQ 5)

Community-Based Fact-Checking on Twitter's Birdwatch Platform - Analysis

- A lot more notes for misleading than non-misleading notes, primarily due to factual mistakes or unclear contexts (RQ 1)
- Negative sentiments and longer explanations typically associated with misleading tweets (RQ 2)
- A lot more discussion on Birdwatch regarding tweets made by politicians (RQ 3)
- Misleading tweets receive more votes and attention on Birdwatch, and users find such interventions helpful if the overall sentiment is positive and supplementary sources are provided (RQ 4) 
- Tweets by users with more followers usually have more votes but are more likely to be regarded as misleading (RQ 5) 

Community-Based Fact-Checking on Twitter's Birdwatch Platform - Results

- Community fact-checking can be improved if people explain the context of their reasoning and provide reliable sources of reference 
- Note contexts also very important – long and understandable explanations necessary 
- Need to study potential side effects of Birdwatch on Twitter users

Learn Before

Related

Learn After