50,000,00 Twitter fans can't be wrong

50,000,00 Twitter fans can’t be wrong ...right?

Marie Boran

Wednesday 9 October 13

Measuring User Influence in Twitter: The Million Follower Fallacy

Cha et al., 2010ICWSM 2010

Cited by 678 (Google Scholar), 498 readers (Mendeley),


What is influence?

• Traditional communication theory - target the influentials (Rogers 1962)

• Influence spreads through opinion leaders (Katz and Lazarsfeld 1955), innovators (Rogers), hubs/connectors/mavens (Gladwell 2002)

• Doesn’t take into account the ordinary users

• Influentials are neither vital nor sufficient for all diffusions (Watts and Dodds 2007)

• Anyone can spark a revolution as long as the mood is right! (Watts 2007)


Influence on TwitterRT retweets

@mentions

indegree


Influence on Twittervalue of one’s

tweets

user’s name value

popularity


An empirical analysis of influence patterns

• Treated Twitter as a news spreading medium

• Studied types and degrees of influence within the network

• Focused on three “interpersonal” Twitter activities

• Used collected data to analyse characteristics of top users


• Used Twitter API to gather tweets and social links for user IDs 0-80 million (Back in ’09 when Twitter API was more accessible!)

• Gathered 55m in-use accounts & 1.75bn tweets

• Filtering: ignored private accounts, those not connected to anyone, users <10 tweets, invalid usernames

• Left with 6m active, connected users - computed 3 influence values for each and compared


Methodology

• Used Twitter API to gather tweets and social links for user IDs 0-80 million (Back in ’09 when Twitter API was more accessible!)

• Gathered 55m in-use accounts & 1.75bn tweets

• Filtering: ignored private accounts, those not connected to anyone, users <10 tweets, invalid usernames

• Left with 6m active, connected users - computed 3 influence values for each and compared


FindingsBased on top 20 users for each measure

Most followed users (unsurprisingly) were public figures and news sources

Most retweeted were content aggregation services, businesspeople, news sites

Most mentioned users were mostly celebrities: people like to mention them without necessarily retweeting their content

Marginal overlap between categories. Two users made the top 20 in all three *cough* Ashton Kutcher and Puff Daddy *cough* < they are entrepreneurs as well as celebs after all!

Mr Fry has 6.2m followers


Insights

• RTs are content driven (92% contain URL), mentions are identity driven ( >30% contain URL)

• RT activity reinforces theory that probability of adopting an innovation increases when not one but a group of users repeat the same message (Watts and Dodds 2007)

• Strong correlation between retweet influence and mention influence

• Indegree was not related to the other measures thus providing evidence for the million follower fallacy so it’s not the follower count that matters but how you use it!


Is influence topic-dependent?

Top news trends in 2009: Michael Jackson’s death, Iranian elections, swine flu.

Authors searched Twitter dataset for related keywords

<2% (13,219) of Twitter users mentioned these topics discussed all three

These users were: well connected, average of 2k followers, tweeted about many topics - perfect group to study user influence across varied topic genres

Power-law: top influentials were RT’d or @’d disproportionately more times than majority of users


Message to marketers: tapping into these top influentials has great potential payoff


Maintaining engagement

Authors measure influence over time in two ways:1.Track popularity of top users over long term2. Look at users who increased influence in specific topic over short time period

Remember Figure 1 overlap? - we look at these all-time influentials and their scores over a 9 month period (had to normalise for Twitter growth spurt; more users, more tweets). FYI Google does this when analysing search trends

All three groups (top 10, top 100, top 233) increased their influence over time but interesting stuff happening with top 10; their popularity fell over time. These were mostly media sources so while users RT breaking news as the follower count grows it becomes difficult for top 10 to engage with audience


Group 2 (celebs) get mentioned more than RT’d due to their name value

Group 3 (evangelists) increased influence by conversing with others (they’re driven by desire to promote themselves!)

Note: Authors say overall slight increase due to limited number of tweets per day. “Broadcasting too many tweets puts even popular users at risk of being classified as spammers”.


What about us ordinary folk?

Back to the news topics: top 20 users (based on follower count) for each topic, referred to as the topical influentials

Included previously unheard of users & figures like Kevin Rose (Digg) who increased popularity after mentioning these news topics

If we look at influence (both RT and mentions) of those talking about Iranian elections we see it peaks in June/July ’09 when elections were ongoing

Those who talked about swine flu and Jackson had bumps in mentions but this soon faded as the news grew stale

Authors found (by manual inspection) that users who stick to a single topic gained the largest increase in influence scores


Conclusions

• Indegree represents popularity but is not related to other kinds of influence such as engaging the audience

• Retweets are content driven, mentions are personal brand driven

• There are three distinct kinds of influential users on Twitter

• Top Twitter users have disproportionate amount of influence

• News orgs good at getting RTs, celebs consistently get high no. of mentions

• Influence isn’t spontaneous or accidental, takes time and effort


Critique

• Conclusions apply to general news topics. Authors don’t explain difference between niche topics or users, could have done this with their dataset, identified communities of influence perhaps, might find different results in e.g. tech, science, sports, politics.

Romero et al., 2001. Differences in the Mechanics of Information Diffusion Across Topics: Idioms, Political Hashtags, and ComplexContagion on Twitter

Information diffusion across various topics by using hashtags

On premise that “widespread intuitive sense that different kinds of information spread differently on-line”

Use concept of “social contagion” to explain spread of topics

Look at “stickiness,” the probability of adoption based on oneor more exposures, but also to a quantity that could be viewed as akind of “persistence”—the relative extent to which repeated exposuresto a hashtag continue to have significant marginal effects


Thank you very much ...questions?


50,000,00 Twitter fans can't be wrong

Technology

twitter api

twitter

inuence