When People, not Computers, Sort SMS Data

Posted by CorinneRamey on Sep 17, 2009

Currently, most SMS surveys have questions that ask people to respond to a menu of multiple choice answers.  But Textonic, an open-source tool that helps sort open-ended text responses, seeks to change that.

"I think it's potentially a major shift in terms of the way we do social research," said Thomas Robertson, one of the lead developers on the project.

Textonic, which has yet to be actually used, was developed as part of a graduate class taught by Clay Shirky in the Interactive Telecommunications Program at New York University. The tool is a way of connecting RapidSMS, the data collection platform used by UNICEF, with Amazon Mechanical Turk.

Amazon Mechanical Turk, which Amazon CEO Jeff Bezos has referred to as "artificial artificial intelligence," is a type of crowdsourcing that recruits humans for simple tasks that are hard for computers, yet easy for people.  The system, which opened publicly in 2005, was originally used by Amazon to weed out duplicate product websites that couldn't be found by a computer.

Today, people from across the globe do mundane jobs, called "human intelligence tasks," like image tagging or podcast transcribing in exchange for micropayments.  Ideally, Textonic will link Mechanical Turk with RapidSMS, allowing responses to RapidSMS surveys to be interpreted by humans instead of computers.

"Instead of using a specifically formatted message, we could ask a simple language questions like 'have you eaten' or 'do you feel safe?'" said Robertson.

Mechanical Turk workers would understand that responses such as "the day before today" and "24 hours ago" all mean "yesterday," whereas computers would likely not sort the data correctly, said Christopher Fabian, co-head of the Innovations Team at UNICEF.  Instead of Mechanical Turk workers, the system could also use others to do the same tasks. "We could potentially use groups of people who otherwise don't have ways of making money, and people in most vulnerable populations," said Fabian.

The application is not likely to be further developed until there is a viable use for it, said Fabian.  For example, the system could be used to track what youth are texting about through two UNICEF youth engagement platforms, speakafrica.org and uniteforclimate.org. "There are text messages coming in from kids all over Africa, and it would be great to see how many kids are talking about what," said Fabian. "We could benefit from better analysis."

Photo: Screenshort from Textonic.org

We have a use for that - though we've built our own software

We've got three organizations doing something similar here in Namibia using a modified version of PlaySMS we've dubbed HERSMS that implements a round-robin style of forwarding messages that aren't handled by keywords to a pool of volunteers who can then respond to the query and have it relayed back to the client.

We use the system to create dialogs, support line style, between volunteers and clients about sexual health topics. The software tracks keywords in the conversations and associates them with themes, which are then graphed and reported.

You can read the Peace Corps news release here: http://www.peacecorps.gov/index.cfm?shell=resources.media.press.view&news_id=1469

I should have the code available for download this week here: http://code.google.com/p/hersms/

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd><p><br> <b><i><blockquote>
  • Lines and paragraphs break automatically.

More information about formatting options