United Airlines Bankruptcy and Newsbots

By on

There’s a very interesting story on Hot Hardware (of all places) about how two news robots interacted to help kickoff the (false!) rumor that UAL was going bankrupt, leading to a huge selloff in the stock.  Apparently the news bot for a newspaper moved the story to the front page because someone read it during a “low news time” — during which even a single read was a lot.  Google News saw the story on the page, and picked the date up from the top of the page, since there was no date on the article.  This date was the current date, not the six years ago date of the original story.  An analyst saw the story on Google News, and a billion $ later, the rest was history.

As a recommender researcher, the scariest part of the story is the inference drawn from that single view of the article, which looked at the time — in the wee hours — like a statistically significant indicator that the article was becoming interesting.  This problem reoccurs in recommenders all the time: if we’re looking to make recommendations of items that are not popular, we’ll often be recommending based on not very much data. 

John

SpinSpotter

By on

Read/WriteWeb has an article about a fascinating new browser plugin called SpinSpotter. The idea is that readers of journalism on the Web can mark it up as “spin”, and propose alternative edits to the text that was in the original article.  For instance, if a journalist writes “Most observers believe that …”, the reader might change it to “Some observers believe that …”.  (If the reporter doesn’t have a study to cite, it’s spin for him to suggest that he knows it’s “most”.  He might be right, but journalism demands that he either cite a source or leave the opinion out.)  Other readers can then choose to see the original text, the original text with symbols marking the “spin”, or the edited text.

Two studies that would be fascinating to do: (1) Does the presence of the spin markers change the messages that a reader takes away from an article?  Are readers able to judge articles more carefully because of the markup?  (2) Do different users agree on the appropriate spin markers?  If not, should a recommender system be used to show users spin marks that they agree with?  What effects does that have on the messages readers take away from articles?

John