Search Your Feelings
Professor Sara Sood’s cyberquest for the heart of the blogosphere.
Story by Lori Kido Lopez '06 / Photo by John Lucas
Often banal and sometimes brutal, the blogosphere does have a heart. Computer
Science Professor Sara Sood finds blogging’s best lies in the emotional stories—tales of breakups and breakdowns, crushing
woes and spectacular joys, stories that resonate with readers on a deep level because we’ve all been there before.
If only they could be found amid the millions upon millions of
entries that populate the blogosphere. Sood’s quest is to help
people find these gems, and maybe, to teach the computer a
thing or two about human emotions.
“What we found was that the stories that rang true to everyone
were things like dreams, confessions, nightmares, fights,
fears, things like that,” says Sood of the research she began six
years ago. Using this information, she began to develop a search
engine that seeks out stories that are strongly motivated by one
of the six basic emotions—happiness, fear, sadness, surprise,
disgust and anger.
This is vastly different from a typical search engine. If you put
the words “happy” and “Obama” into Google, the sites that pop
up include information about Obama’s White House happy hours
and a mix tape called “Obama’s Happy Ending”—neither of
which have distinct emotional content. Sood’s goal is to be able
to search for content about “Obama” but also to be able to
specify that the stories are emotionally “happy”—and actually be
able to come up with a list of articles where the writer is feeling
joyous about the topic of Obama. These stories might include
topics like the euphoria and love surrounding Obama’s family, or
excitement toward his message of change.
Finding value in the messy human emotions expressed in
blogs may seem an unusual project for a computer scientist to
take on, particularly given that Sood only enters the blogosphere
for research. She doesn’t read blogs on her own time, and she is
too afraid of their public nature to ever write one of her own.
“If your work was to read a thousand blogs a day you probably
wouldn’t read them at home,” she admits, somewhat sheepishly.
Nevertheless, her passion for stories and curiosity about the
way they work keeps her coming back to blogs in project
Sood and student researcher Lucy Vasserman ’10 will present
a paper on the “Emotional-State Search Engine” at an international
conference in May and, by the end of summer, Sood
hopes to have the engine up and running for public use.
She first began examining blogs after being assigned to create
a team of digital improv performers during her doctoral program
at Northwestern University—an assignment given to her
by a graduate school advisor who himself was an improv comedian.
The task, which she considered “kind of daunting,” led
her straight to the emotionally rich database of the blogosphere,
where countless stories were just waiting to be discovered.
Hidden amongst the thousands of Web sites such as Blogger,
Livejournal and Xanga are these sprawling networks mapping
the minutiae of one person’s life. Everything from big events,
like landing a new job or getting married, to the smallest of
moments, like the creepy person standing behind you at the
checkout counter or what you cooked for dinner last night, can
all make their way into blog
“My husband and I got into a fight on Saturday night; he was drinking and neglectful, and I was
feeling tired and pregnant and needy. It’s easy to understand how that combination could escalate,
and it ended with hugs and sorries, but now I’m feeling fragile,” reads one of the blogs that
Sood tapped for the project. It’s an intensely personal story socked with a variety of emotions—desperation
and worry, anger and exhaustion—and it’s a prime example of the kind of thing that people are writing about every day for the whole world to see.
As April Wensel ’08, one of Sood’s mentees, has discovered,
one of the most interesting and satisfying things about blogs is
that they provide a way to get deep into people’s heads. “Going
up to people on the street and asking them how they felt when
their grandma died or when they got dumped doesn’t really
work,” she says. “And yet for some reason, people divulge exactly
this kind of emotional and personal information online in
their blogs every day.”
Sood’s quest is to find the most compelling of those emotional
voices and let them be heard.
That goal might seem simple as Sood tells it—one gets the
feeling from talking to her that she could make even the most
complicated programming languages sound easy—but getting
there has been part of a long journey. First, she had to train
the computer to recognize entries that were stories, as opposed
to entries that were mere catalogues or exposition. This work
comprised the bulk of her dissertation project, and involved
putting each entry through a series of filters to see if it had
qualities of a story, such as the right length, focus on just one
topic, and uses phrases common to storytelling, such as “I had
a dream last night.”
After figuring out which entries were stories, she needed to
find out which stories were the most emotional and compelling.
To do this, she gave the computer 100,000 reviews of movies
that had been awarded between one and five stars. Since the
one-star reviews were damning and the five-star reviews were
glowing, this produced a database of positive language and negative
language for the computer to track. If a story contained a
lot of positive words (loved, excited, thrilling) or a lot of negative
words (frustrated, stupid, resentful), it was probably emotionally
charged. If the words were somewhat neutral, the program
would cast the story aside.
Through this process, Sood began to see a glimmer of her
ultimate goal to teach computers how to detect a complex array
of emotions within stories. One of her most involved attempts
to integrate this technology, called Buzz, began with these questions
of how to seek out emotional blog entries, but ended up
as an acclaimed digital theatre art installation. For an entire year,
audience members at the Second City Theater in Chicago could
see four digital avatars telling stories in the theatre’s lobby.
Although Sood’s computer programming background helped her to write the code for the
project, it was her artistic sensibilities that inspired her to take
extra time to make sure that the avatars’ voices didn’t sound like
monotone robots. She fiddled with the pitch, volume, and
speed of the recording until the voices actually conveyed the
emotions within the stories. Additionally, she designed it so that
when one avatar speaks, the others look toward him or her, like
a blog-reading Brady Bunch family. It’s a cute effect, but it conveys
the importance of Sood’s goal—that we listen to each other
when an important story is being told.
Beyond humans listening to each other, Sood also wants
computers to be able to listen and to detect the emotions that
are present. If she could do this, it would signal one of the most
important achievements in the effort toward artificial intelligence,
since Sood and others believe that computers need to be
able to understand feelings as well as facts and figures in order
to demonstrate human-like intelligence.
“If machines are so much a part of our lives, our interactions
with them should be compelling and enjoyable,” she says. “If
we’re building these machines that we want to be intelligent,
then they have to be emotionally intelligent.”
But dreams of artificial intelligence are a long way off, and
for now, the most important thing for Sood is to find and establish
strong connections through stories—whether those connections
are between people in person, people on the Internet or
people and their machines.