Maybe you’ve noticed the odd little circle diagrams below some news articles on The Washington Post’s web site? (I hadn’t, but a more observant co-worker pointed them out). The Post has partnered with a company called Evri which provides, at the bottom of Post articles, an “Evri widget” that
makes article recommendations based on Evri’s semantic understanding of how the people places and things in the article being read are connected to other people, places and things being discussed on the web.
The widget, in its own words, offers readers the chance to “understand more about” (via links) some of the people and places mentioned in the article. My co-worker gave it a try after reading the (rather bleak) Post article today, “Homelessness: The Family Portrait” (about how “the face of homelessness in America” may be changing as, in the current economic climate, “more two-parent families [are] seeking shelter.”) In the article, readers meet six-year-old Jake, whose family is temporarily housed in a motel until space opens at a county homeless shelter.
Readers who then take Evri’s opportunity at the bottom of the article page to “understand more about” Jake are linked, for example, to a three-day-old article from a British tabloid about “Heather Locklear’s Melrose Return” (which includes a reference to Melrose Place’s “heart-breaker Jake”) and a 2003 Post review of the Lindsay Lohan movie Freaky Friday (including a reference to the character “Jake, played by swoon-inducing Chad Michael Murray”).
I’m not seeing how this might be any better than many of those seemingly random and not-very-useful in-article links that already populate online newspaper articles. The ones that presume, for example, that when you’re reading a New York Times article about a man who got a perm in hopes of raising money for a sick friend and you’re told that his post-perm hair “suggested a little bit of Michael Bolton, a little bit of Weird Al Yankovic and, once subjected to a hair dryer, a whole lot of the Guns N’ Roses guitarist, Slash,” you might also want to revisit everything the Times has ever reported about Guns N’ Roses — though not Michael Bolton or Weird Al.
Liz,
thanks for taking the time to look at, and review -- even critically -- what we are doing here at Evri. It is true that we don't currently do well on all articles uniformly, but we do provide relevant content on most articles viewed on the Post site. A better example than above would be what is one of the most read articles today: "Politically, Stimulus Battle Has Just Begun" http://tinyurl.com/brlj3r.
We do best with articles that have well-known entities -- like most of the articles published by the Post. In the article I mention, both the names at the top, and the article recommendations, are highly relevant and not what is given by any of the other options on the page.
Articles that have little additional information relating to the people, places and things in the article can still stump our otherwise sophisticated software -- but this gets better all of the time. But, using the linguistic structure and content of the sentences in the article can and does achieve much better than random results.
You might want to even try this on some of your own posts. I took your post on Alexandra Pelosi and got articles from Salon, the Wash Post, USA Today, and Gawker that were all relevant and not easily achievable by other means. (You can try it with any content here: http://www.evri.com/about/demo.html?widget=jirafa)
So, give us another try, and while YMMV of course, I believe (and hope) you will see the value here. Regardless, thanks for the feedback.
#1 Posted by Neil Roseman, CJR on Mon 16 Feb 2009 at 07:44 PM