Resilient Ambassadors of Change trend-tracking. idea-generation. progress.

Looking at dating websites is all in a day’s work.

logo pulled from Junebug's website

We've talked about search engines customizing your results according to what you seem to usually like at work (and here on the blog), so when I heard a very brief mention of a "new" dating website called Junebug this morning on NPR, I thought I should mention it. 

It was a really quick story - quick enough that they didn't even list it on their website as something on the show!  The story explained Junebug as a dating site that works with a similar algorithm that google and amazon use.  They get you go answer some questions, just like every other dating site, but they also track the profiles you look at, extract the "most important" pieces of information from that profile, and find others with the same characteristics to suggest to you. 

I'm no expert in online dating sites - not  since I found my boyfriend on okcupid two years ago, anyway - but this seems more sophisticated than the usual, "fill these questions out and we'll find people who answered those questions similarly and match you" kind of thing.  It's also more robust than simply asking, "Do you think opposites attract?" and then adjusting the results to find your mortal enemies online and pair you with them.

Naturally, there are disclaimers all over the about section about how it takes a while to get a good feel for who Junebug thinks you'd like, but I think that just like every other recommender system it will become very good at figuring out what you find attractive in people.

I'd tell you to to ask the lady in the story who met her husband on the website, but I couldn't find it anywhere on NPR's site.  Reference fail!

Mammographs Eyeballed by Computers

I read in the New England Journal of Medicine that in many European countries, it is standard for two people to
read mammogram films as opposed to having just one person read them as it is
practiced in the United States.  Having two people read the films obviously
increases the chance that something that should be looked at will be caught.
However, this can be a time and resource-consuming process.  In the New England
Journal of Medicine, I saw an article from a group of scientists who wondered
if there might be a way to use computers to read the films as a second set of
“eyes” instead of another person. If so, can it be as accurate as
two sets of human eyeballs?

This study found that the cancer-detection rate (people with
definite cancer) was about the same with one person reading with
computer aid and with two people reading the films.  However, the recall rate
(people with suspicious masses that could be cancer) was significantly
higher in the single person with computer aid group.  I see an up-side and a
down-side to these findings.  The up-side:  Using one person as opposed to two
frees up time and resources to read more mammogram films.  The down-side:  The
recall rate is higher, meaning that many people who have benign tumors may be
being called back for further testing.  This probably uses more time and
resources than having two people initially read the scans. 

It’s very easy to connect this research with
computer-assisted projects like our own item parsing pilot, Lucy, or with our
interest in the field of computer simulations in general.  Even when results
like these seem to break even, it’s important that people are doing this
kind of research.  We’ll never figure out what technology works best
without people experimenting in this field.

For those who are interested in the science-y numbers, check out the article in the New England Journal of Medicine 2008 Oct 16;359(16):1675-84. Epub 2008 Oct 1.

Google’s new browser

Google released its new web browser "Chrome" today--a project that has been in the making for more than a year.  Chrome is designed to be fast and simple to use.  Apparently, another design goal was to assure that Google would have control over how well its web-based applications perform, by creating a browser that runs today's complex web applications and future web applications with ease.  I don't really have a deep understanding of the technology behind Chrome, for that you'll have to rely on the comic book developed by comic artist and writer Scott McCloud to introduce the general public to Chrome's inner workings. 

I downloaded Chrome and started using it in seconds--not minutes.  The last time I downloaded a new version of Internet Explorer, it got lost and I had to reboot and download it all over again.  Then I had to reboot again. So, I'm off to a good start.  It was really simple to import my bookmarks, and explanations for some of the features I hadn't experienced in a browser before were pretty straightforward, although some of the video links were broken (that's why it's beta!).   

After browsing for a bit, each time I opened a new tab, it was populated with my most recent searches and bookmarks..  One really neat feature was the dynamic tabs.  You just click and drag a tab if you want to view the site in its own window.  There is also an option to open an incognito tab, and none of the history, cookies, or downloads from that session will be recorded.  When you download a file, the file seems to float to the bottom of the page, where you can then drag it wherever you want, instead of searching all over the place for your downloaded file.  Another feature treats some web applications like computer applications.  For example, Gmail can be launched from a desktop shortcut in its own window, without the address bar and other tools it does not need.   

Download it, give it a try.

A mental model for mashed potatoes?

In the most recent edition of Edge, Chris Anderson suggests that perhaps the "data deluge" will make the scientific method obsolete.  That is, in an age where data comes by the petabyte, perhaps mental models and hypotheses can be abandoned and analyzing information for correlations is enough.  I'm not a strict follower of the scientific method, so I don't feel that I can make a strong point either way.  However, I am  reminded of the mashup craze.  In creating a mashup, people don't necessarily have a strong model going in; in many cases they just throw data together and see what happens.  This "on the fly" experimentation is possible because of computers and applications that enable it.  Intuitively, I would say that reality falls somewhere in the middle.  We might create a mental model (e.g., a recipe) for the perfect mashed potatoes, or we might throw a bunch of potatoes and spices in a pot and see what happens.  And both approaches are probably important.

Who lives and who dies; who succeeds and who does not

A new computer system can predict which death row prisoners will live and which will be executed - with chilling accuracy. The US executes only a small proportion of the people it sentences to death; 53 of 3228 in 2006, and until now, no clear patterns were evident to researchers about which inmates end up paying with their lives and which do not.  By inputting data about the fates of 1000 death row inmates from 1973 through 2002, criminologists and computer scientists were able to train an artificial neural network (ANN) to predict the fates of an additional 300 inmates with greater than 90% accuracy. From this experiment, they learned that gender was the most significant factor linked to executions, as women are rarely executed.  Race, although apparently a huge factor in death sentence rates, is not an important factor when it comes to actual executions.  The most significant factor is educational level, particularly the number of years spent in high school.  According to an article in New Scientist, "This finding confirms that being executed is not about what you've done, but more about your ability to defend yourself."

I'm thinking that I would not want to be on death row surrounded by college graduates.  Researchers have also used ANNs to predict the likelihood that paroled juveniles will reoffend and to identify students most likely to drop out of college.  I'm also thinking that predicting success would certainly be a more cheerful task for researchers.  Wouldn't it be great to know whether or not you are likely to succeed in your chosen career path before you invest your energy, your heart, and your money?  Would dire predictions of an ANN be enough to stop some people from pursuing their dreams?  What effect would knowing the outcome have on the journey?