Twitter : Social Network or News Media

This year's WWW conference has an interesting paper by four researchers from KAIST, KOREA titled "What is Twitter, a social network or news media? The paper has lots of really interesting insights.

• Structure of Twitter is not similar to typical human social networks having a 'non-power law' follower distribution, short effective diameter and low reciprocity.


• The top ranked users calculated using the number of retweets is different from the users with the largest the number of followers.


• Any retweeted tweet spreads very quickly irrespective of the number of followers of the original tweet.


• Subset of users with reciprocated links display the characterstics of a social network including homophily.


• Majority (85%) of trending topics are headlines or persistent news in nature.

So it seems there are a subset of users in Twitter who use it for communicating with people in their network while a larger portion of users use it just as a source of news. And similar to sites like Digg collective intelligence (in the form of retweets) decides which information is the most important.

However Twitter is very unlike a traditional news media in that it's main utility is as a distribution medium. Unlike the traditional media or blogs Twitter cannot be the main source of original content.

Google's SETI

Google has developed a new large scale machine learning system called SETI (Since it's searching for signals in a large space). In the article Simon Tong from Google research talks about how they sacrificed accuracy by a bit to achieve ease of use and system reliability. He says
"Seti is typically used in places where a machine learning system will provide a significant improvement in accuracy over the existing system. The gains are usually large enough that most teams do not care about the small differences in accuracy between different flavors of algorithms. And, in practice, the small differences are often washed out by other effects such as better data filtering, adding another useful feature, parameter tuning, etc. "

The last point has been true in our own work of developing classifiers for Kindle Content Ingestion. Data cleaning, feature selection, parameter tuning and post processing have yielded the largest improvements of accuracy.