The spread of true and false news online
Transcript of The spread of true and false news online
![Page 1: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/1.jpg)
Enis Ceyhun Alp - 230420211
The spread of true and false news onlineSoroush Vosoughi, Deb Roy, Sinan Aral
![Page 2: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/2.jpg)
2
![Page 3: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/3.jpg)
3
![Page 4: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/4.jpg)
4
![Page 5: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/5.jpg)
5
![Page 6: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/6.jpg)
5
![Page 7: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/7.jpg)
5
![Page 8: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/8.jpg)
Problem definition‣ Social media technologies are very prominent in guiding our access to
information and news
‣ Little known about their contribution to falsity online
6
![Page 9: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/9.jpg)
Problem definition‣ Social media technologies are very prominent in guiding our access to
information and news
‣ Little known about their contribution to falsity online
6
How and why do truth and falsity diffuse differently? Are there categorical differences that influence diffusion?
What factors of human judgement explain these differences?
![Page 10: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/10.jpg)
Some terminology
7
![Page 11: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/11.jpg)
Some terminology‣ “True” and “false” news
- Any story or claim with an assertion in it
- Any asserted claim made on Twitter
7
![Page 12: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/12.jpg)
Some terminology‣ “True” and “false” news
- Any story or claim with an assertion in it
- Any asserted claim made on Twitter
‣ Rumor
- Social phenomena of a news story/claim spreading through Twitter
- Involves sharing of claims between people
7
![Page 13: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/13.jpg)
Some terminology‣ “True” and “false” news
- Any story or claim with an assertion in it
- Any asserted claim made on Twitter
‣ Rumor
- Social phenomena of a news story/claim spreading through Twitter
- Involves sharing of claims between people
‣ Rumor cascade
- Starts with a user making an assertion on Twitter (text, photos, links, etc.)
- Unbroken retweet (RT) chain with a common & singular origin
- Rumor diffusion through one or more cascades
7
![Page 14: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/14.jpg)
Overview‣ Investigating the differential diffusion of true, false, and mixed news stories
‣ Dataset of ~126,000 rumor cascades on Twitter
- Spread by ~3M people more than ~4.5M times
- Between 2006-2017. Access to all tweets ever posted.
- Investigated by six independent fact-checking organizations (95-98% agreement on classifications)
8
![Page 15: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/15.jpg)
Data collection
9
![Page 16: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/16.jpg)
Data collection‣ Collect all English tweets with link to the organizations’ websites (500k)
9
![Page 17: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/17.jpg)
Data collection‣ Collect all English tweets with link to the organizations’ websites (500k)
‣ Consider only the replies. For each reply tweet, extract the original tweet and all of its RTs (= RT cascades).
9
![Page 18: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/18.jpg)
Data collection‣ Collect all English tweets with link to the organizations’ websites (500k)
‣ Consider only the replies. For each reply tweet, extract the original tweet and all of its RTs (= RT cascades).
‣ Each RT cascade = rumor propagating on Twitter
- Veracity of the RT cascade - through the reply with link
9
![Page 19: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/19.jpg)
Data collection‣ Collect all English tweets with link to the organizations’ websites (500k)
‣ Consider only the replies. For each reply tweet, extract the original tweet and all of its RTs (= RT cascades).
‣ Each RT cascade = rumor propagating on Twitter
- Veracity of the RT cascade - through the reply with link
‣ Manually and automatically ensure that replies address the original tweet
- Only considered replies that target the original tweet (no replies to replies)
- Compared the headline of the linked article to that of the original tweet
• Cosine similarity
9
![Page 20: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/20.jpg)
Rumor cascades
10
![Page 21: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/21.jpg)
Rumor cascades
‣ Depth: number of RT hops (by a unique user) from the origin tweet over time
10
![Page 22: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/22.jpg)
Rumor cascades
‣ Depth: number of RT hops (by a unique user) from the origin tweet over time
‣ Size: number of users involved in the cascade over time
10
![Page 23: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/23.jpg)
Rumor cascades
‣ Depth: number of RT hops (by a unique user) from the origin tweet over time
‣ Size: number of users involved in the cascade over time
‣ Max. breadth: max number of users in the cascade at any depth
10
![Page 24: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/24.jpg)
Rumor cascades
‣ Depth: number of RT hops (by a unique user) from the origin tweet over time
‣ Size: number of users involved in the cascade over time
‣ Max. breadth: max number of users in the cascade at any depth
‣ Structural virality: average distance between all pairs of nodes in the cascade
10
![Page 25: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/25.jpg)
Rumor cascades
11
![Page 26: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/26.jpg)
Rumor cascades
11
![Page 27: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/27.jpg)
Rumor cascades
12
![Page 28: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/28.jpg)
Rumor cascades
13
![Page 29: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/29.jpg)
Rumor cascades
14
![Page 30: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/30.jpg)
Rumor cascades
15
![Page 31: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/31.jpg)
Diffusion dynamics of rumors
16
![Page 32: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/32.jpg)
Diffusion dynamics of rumors
16
![Page 33: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/33.jpg)
Diffusion dynamics of rumors
17
![Page 34: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/34.jpg)
Diffusion dynamics of rumors
18
![Page 35: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/35.jpg)
Diffusion dynamics of rumors
19
![Page 36: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/36.jpg)
20
Diffusion dynamics of rumors
![Page 37: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/37.jpg)
21
Diffusion dynamics of rumors
![Page 38: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/38.jpg)
Falsehood diffused significantly farther, faster, deeper and more broadly than the truth in all
categories of information.
![Page 39: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/39.jpg)
Diffusion dynamics of false political news
23
![Page 40: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/40.jpg)
Diffusion dynamics of false political news
24
![Page 41: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/41.jpg)
Diffusion dynamics of false political news
25
![Page 42: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/42.jpg)
False political news diffused farther, faster, deeper and more broadly than any other type of false
news.
![Page 43: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/43.jpg)
Network structure‣ What if the structural elements of the network or individual characteristics of
the users explain why falsehood travels faster than the truth?
27
![Page 44: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/44.jpg)
Network structure‣ What if the structural elements of the network or individual characteristics of
the users explain why falsehood travels faster than the truth?
27
![Page 45: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/45.jpg)
Network structure‣ What if the structural elements of the network or individual characteristics of
the users explain why falsehood travels faster than the truth?
27
Falsehood diffused farther and faster than the truth despite the differences in the network structure, not because of them.
![Page 46: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/46.jpg)
Novelty of information
28
![Page 47: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/47.jpg)
Novelty of information‣ Novel information:
- Attracts human attention
- Contributes to decision-making
- Encourages information sharing
- More valuable both from an information theoretic perspective and from a social perspective
28
![Page 48: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/48.jpg)
Novelty of information‣ Novel information:
- Attracts human attention
- Contributes to decision-making
- Encourages information sharing
- More valuable both from an information theoretic perspective and from a social perspective
‣ Q1: Was falsity more novel than the truth?
28
![Page 49: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/49.jpg)
Novelty of information‣ Novel information:
- Attracts human attention
- Contributes to decision-making
- Encourages information sharing
- More valuable both from an information theoretic perspective and from a social perspective
‣ Q1: Was falsity more novel than the truth?
‣ Q2: Were Twitter users more likely to RT information that was more novel?
28
![Page 50: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/50.jpg)
Novelty of information
29
![Page 51: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/51.jpg)
Novelty of information‣ Randomly selected ~5000 users
29
![Page 52: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/52.jpg)
Novelty of information‣ Randomly selected ~5000 users
‣ Extracted a random sample of ~25,000 tweets they were exposed to in the 60 days prior to their RT of a rumor
29
![Page 53: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/53.jpg)
Novelty of information‣ Randomly selected ~5000 users
‣ Extracted a random sample of ~25,000 tweets they were exposed to in the 60 days prior to their RT of a rumor
‣ Latent Dirichlet Allocation Topic model with 200 topics and trained on 10M English tweets
- Calculate information distance between the rumor tweets and exposed tweets
29
![Page 54: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/54.jpg)
Novelty of information‣ Randomly selected ~5000 users
‣ Extracted a random sample of ~25,000 tweets they were exposed to in the 60 days prior to their RT of a rumor
‣ Latent Dirichlet Allocation Topic model with 200 topics and trained on 10M English tweets
- Calculate information distance between the rumor tweets and exposed tweets
‣ Compare topic distributions of the rumor tweets and exposed tweets
29
![Page 55: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/55.jpg)
Novelty of information‣ Randomly selected ~5000 users
‣ Extracted a random sample of ~25,000 tweets they were exposed to in the 60 days prior to their RT of a rumor
‣ Latent Dirichlet Allocation Topic model with 200 topics and trained on 10M English tweets
- Calculate information distance between the rumor tweets and exposed tweets
‣ Compare topic distributions of the rumor tweets and exposed tweets
29
False rumors were significantly more novel than the truth.
![Page 56: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/56.jpg)
Perceived novelty
30
![Page 57: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/57.jpg)
Perceived novelty‣ False rumors are measurably more
novel than true rumors
30
![Page 58: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/58.jpg)
Perceived novelty‣ False rumors are measurably more
novel than true rumors
‣ But is this how users perceive them as well?
30
![Page 59: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/59.jpg)
Perceived novelty‣ False rumors are measurably more
novel than true rumors
‣ But is this how users perceive them as well?
‣ Compared the emotional content of replies to true and false rumors
- ~140,000 English words & their associations with eight emotions
- ~32,000 Twitter hashtags & their weighted associations with the same emotions
30
![Page 60: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/60.jpg)
Perceived novelty‣ False rumors are measurably more
novel than true rumors
‣ But is this how users perceive them as well?
‣ Compared the emotional content of replies to true and false rumors
- ~140,000 English words & their associations with eight emotions
- ~32,000 Twitter hashtags & their weighted associations with the same emotions
30
![Page 61: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/61.jpg)
Validation of findings (1)
31
![Page 62: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/62.jpg)
Validation of findings (1)‣ Multiple cascades for every true and false rumor
31
![Page 63: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/63.jpg)
Validation of findings (1)‣ Multiple cascades for every true and false rumor
‣ Variance and error terms associated with cascades corresponding to the same rumor will be correlated
31
![Page 64: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/64.jpg)
Validation of findings (1)‣ Multiple cascades for every true and false rumor
‣ Variance and error terms associated with cascades corresponding to the same rumor will be correlated
‣ Solution:
- Specified cluster-robust standard errors and calculated variance at rumor-cluster level
- Clustering reduced precision but the significance of the results did not change
31
![Page 65: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/65.jpg)
Validation of findings (2)
32
![Page 66: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/66.jpg)
Validation of findings (2)‣ Potential selection bias
- Restricting sample to tweets that are fact-checked by the six organizations
- Fact checking may select certain types of rumors or draw additional attention to them
32
![Page 67: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/67.jpg)
Validation of findings (2)‣ Potential selection bias
- Restricting sample to tweets that are fact-checked by the six organizations
- Fact checking may select certain types of rumors or draw additional attention to them
‣ Solution:
- Independently verified another sample of rumor cascades that was not verified by any fact-checking organizations
- Manually fact-checked by three undergrads (agreed on the veracity of 90% of the 13,240 cascades)
- Diffusion dynamics of true & false rumors are nearly identical to the main data set
32
![Page 68: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/68.jpg)
Validation of findings (3)
33
![Page 69: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/69.jpg)
Validation of findings (3)‣ Potential presence of bots that can bias conclusions about human
judgement
33
![Page 70: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/70.jpg)
Validation of findings (3)‣ Potential presence of bots that can bias conclusions about human
judgement
‣ Solution:
- Bot-detection algorithm to identify and remove all bots before the analysis
- Adding back bot traffic into the analysis did not change the main conclusions
- Inclusion of bots accelerated the spread of both true and false news and affected their spread equally
33
![Page 71: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/71.jpg)
Conclusion
34
![Page 72: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/72.jpg)
Conclusion‣ False news spread more pervasively than the truth online
- Faster, farther, deeper and more broadly than the truth in all categories of information
34
![Page 73: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/73.jpg)
Conclusion‣ False news spread more pervasively than the truth online
- Faster, farther, deeper and more broadly than the truth in all categories of information
‣ Effects are more pronounced for political news
34
![Page 74: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/74.jpg)
Conclusion‣ False news spread more pervasively than the truth online
- Faster, farther, deeper and more broadly than the truth in all categories of information
‣ Effects are more pronounced for political news
‣ Network structure and individual characteristics of spreaders do not favor promoting false news
34
![Page 75: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/75.jpg)
Conclusion‣ False news spread more pervasively than the truth online
- Faster, farther, deeper and more broadly than the truth in all categories of information
‣ Effects are more pronounced for political news
‣ Network structure and individual characteristics of spreaders do not favor promoting false news
‣ False news are more novel than true news - greater likelihood of being shared with others
34
![Page 76: The spread of true and false news online](https://reader030.fdocuments.us/reader030/viewer/2022013009/61cdc10549c68f22a501ac80/html5/thumbnails/76.jpg)
Conclusion‣ False news spread more pervasively than the truth online
- Faster, farther, deeper and more broadly than the truth in all categories of information
‣ Effects are more pronounced for political news
‣ Network structure and individual characteristics of spreaders do not favor promoting false news
‣ False news are more novel than true news - greater likelihood of being shared with others
‣ Bots accelerated the spread of true and false news at the same rate
34