If sheep could tweet

In Connie Willis' book, Bellwether, two researchers acquired a herd of sheep (they were studying fads). However, no sheep agreed to start a new fashion of pressing a button for food. What they needed was a bellwether, a fads-starting sheep.

Cha et al. searched for bellwethers ('influentials') on Twitter. They sampled more than six million active users ('active' means 'more than ten tweets'). They used three measures of influence: followers (indegrees), retweets and mentions. The number of followers indicated the size of the user's audience, retweets indicated the value of a tweet's content, and mentions indicated the user's ability to engage in conversation with other users.

The most followed users were news sources, politicians (Obama) and celebrities in general. However, the most retweeted users were the Mashable blog, Twittertips and TweetMeme, as well as businessmen (Guy Kawasaki) and news sites (they include The Onion under this category, which amused me greatly). Retweets are influential due to their ability to pass and reinforce a message to users way beyond the followers of the tweet's creator. The authors consider retweets citations of users' content.

Mentions - celebrities were often at the top of the 'most-mentioned' list. Since less than 30% of the 'mention' tweets contained URLs, the authors concluded that mentions are more about a person than about content.

The number of tweets and number of people a user follows (outdegrees) weren't significant influence indicators, simply because those were spammers.

Even ordinary users can rise to fame (mostly of the 15-minutes kind) if they have interesting content. Users like iranbaan, oxfordgirl and TM_Outbreak became immensely popular during the Iranian elections. Unlike those users, the Swine flu bellwethers, in the absence of catastrophic flu outbreaks, remained relatively stable in influence and popularity.

In conclusion

• The number of followers doesn't necessarily make one a bellwether.
• Retweets are mostly content-driven
• Mentions are mostly user-driven
• News sites do better at retweets, while celebrities get more mentions.
• Influence on Twitter takes supplying plenty of content.

Cha, M., Haddadi, H., Benevenuto, F., & Gummadi, K. P. (2010). Measuring User Influence in Twitter: The Million Follower Fallacy ICWSM '10: Proceedings of international AAAI Conference on Weblogs and Social Media.

Tweeting emotions: sentiment in Twitter events

Thelwall*, Buckley and Paltoglou (it's a preprint of a paper which is going to be published in JASIST soon) studied the emotional responses in 34, 770, 790 Twitter messages, tweeted from February 9 to March 9, 2010. The main two events in the media that month were the Oscars and the Winter Olympics.

They identified the 30 most popular events (biggest spikes of interest) and tried to classify the sentiment strength expressed in the tweets using an algorithm called SentiStrength. SentiStrength was developed for analysis of MySpace comments, which means it is most suitable for analysing, shall we say, English that didn't come straight out of the Oxford dictionary. SentiStrength measures emotions with two 1-5 scales, one for positive sentiment and one for negative.

As far as emotions on Twitter go, important events mean mainly an increase in negativity, with heavier traffic hours having stronger negativity sentiment. Also, hours after the traffic's peak have stronger sentiment than the hours before. There was also a correlation between higher traffic hours and stronger positive sentiment, in comparison with the hours before, but it wasn't as strong as the correlations found for negative sentiment.

The authors conclude that "important events in Twitter are associated with increases in average negative sentiment strength".

Increase in negative sentiment strength doesn't always say a decrease in positive sentiment, because people can see an event from various view points (is Sandra Bullock winning the Oscar good or bad? Depends on the person asked).

Unfortunately, the authors weren't able to identify the importance of an event from the strength of the expressed emotions. While statistically significant, the changes in emotion strength related to popular events were only around 1% and "were far from universal". The authors' conclusion was that tweeting about events is less about emotional "gut" reactions than about "affording posters opportunities to satisfy personal goals" (say, making a joke).

Thelwall, M., Buckley, K., & Paltoglou, G (2010). Sentiment in Twitter events JASIST

*same disclosure as last post.

Authorities and hubs in Twitter conference feeds

"Understanding how Twitter is used to spread scientific messages" is another conference paper studying the scientific uses of Twitter.

Letierce, Passant, Breslin and Decker (2010) analysed Twitter feeds from the International Semantic Web Conference (#iswc2009), the Online Information Conference 2009 (#online09) and the European Semantic Technology Conference (#estc2009). First, they checked the distribution of tweets per user, then the distribution of tweets that were directed to individuals (@user messages). They found that both were Power Law distributions.

After that, they used the HITS (Hyperlink-Induced Topic Search) algorithm to determine the hubs and authorities of the conference feeds. Users who addressed many @user messages were considered hubs, while the users who received many @user tweets were considered authorities. Letierce et al.'s not-very-surprising conclusion was that users with both high hub and authorities scores were often the organizers of the events studied in the research. Also, users with real-world authority (their example was @timberners_lee) also had t-authority. Of course, Letierce and her colleagues couldn't determine if there are more real-world authorities that don't use Twitter (perhaps a content analysis of the tweets can determine if there are talks about people who aren't Twitter users, but that's very time-consuming and not the point of the research here).

In addition, the paper includes a small survey (61 participants) conducted by the authors, but I chose not to discuss it here, because of the small sample.

Letierce, J., Passant, A., Breslin, J., & Decker, S. (2010). Understanding how Twitter is used to spread scientific messages Proceedings of the WebSci10: Extending the Frontiers of Society On-Line, April 26-27th, Raleigh, NC: US.

Scientometrics 2.0, part I

Today we're going to discuss this paper, by Priem and Hemminger (2010), dealing with Scientometrics (a general name for Bibliometrics, Webometrics, Influmetrics, and all sorts of other metrics regarding scientific activity). But, before we start, I want to protest First Monday's terrible references format. Sure, it looks fine in-text, with the last name and the year, but at the reference list the references start with the first letter of the author's first name! How am I supposed to know the author's first name? Only Darwin knows.

Anyway, Priem and Hemminger offer a state-of-the-art review of Web 2.0 tools that can be mined for scholarly data. They suggest seven categories of tools, of which I'm only going to mention four in this post.

1. Microblogging - Microblogging means Twitter. Twitter is used by scientists, among others, to discuss papers and conferences. From my experience, many people tweet from conferences under the conference hashtag.

2. Social Bookmarking - Social bookmarking services like Delicious, Connotea and CiteULike can be mined and show scientific trends by pointing out the popular papers bookmarked and popular tags.

3. Wikipedia - Wikipedia is popular among students and faculty as a starting point for basic knowledge. For other web users, Wikipedia is often their only knowledge source. So, papers cited in Wikipedia are more likely to have public impact. Indeed, the JCR and Wikipedia citations correlate well.

4. Blogging - By now, blogs are well- established in the web culture. Many scientists, Fields Medalists included, maintain scholarly blogs. When discussing academic papers, those scientists often cite their sources in traditional manner. Excellent examples are posts aggregated by the Research Blogging service.

That's it for today. The next part will discuss other categories of Web 2.0 services with possible scholarly use.