Socializing online and in the real world may be edging ever closer, but one stark difference remains: it’s much easier to juggle different groups of friends offline than it is online. A new tool developed at Stanford University addresses the problem by automatically working out a person’s different and overlapping friendship groups by analyzing the history of their Facebook or Gmail account.

The tool—called SocialFlows—is available as a Facebook app. Once a Facebook user connects it with their account, they can extract their social groups in two ways. The researchers’ software can either work from the images a person was tagged in over the last two years, or from the pattern of e-mail recipients recorded in the last two years via a Gmail account.

Once SocialFlows has processed the necessary data, it suggests different groups of contacts or friends, some overlapping. An interactive interface allows the user to add, remove, or reshuffle the groups and then save them to either Facebook or Gmail as contact lists or new groups.

Putting different types of contacts into different groups—family or coworkers, for example—can be useful to people who want to control what they share, and with whom, online, explained Diana MacLean, a member of the MobiSocial Lab at Stanford where she and colleagues created SocialFlows.

“The current solution is friends lists,” said MacLean at the Intelligent User Interfaces conference earlier this week. “That does give some granular control over who gets to see our data, but they’re very tedious to create and even more tedious to maintain over time.” As people acquire new contacts, they must keep adding them to groups or perhaps create new ones, and there is currently no way to transfer groups between services, she pointed out.

Social sites actually have all the data they need to tackle these problems, said MacLean. “Your social topology is captured latently in your communication data.” SocialFlows is an attempt to tap into that. The algorithm behind the tool first creates many small groups by linking people who have often appeared in the same photo as the SocialFlows user, or who have been recipients of the same e-mail. Larger groups are then assembled by merging groups that are similar. Some groups become nested inside others, as when the data reveals a pattern of connections between a subset of people within a larger group.

MacLean and colleagues conducted a user study in which 19 volunteers were asked to use SocialFlows on their e-mail to generate and fine-tune groups that could help them achieve specific tasks, for example, notifying close friends in their home town of an upcoming visit, or inviting guests to a birthday party. Study participants also performed the same task using Gmail’s contact manager.

“Six of the 19 found doing the task using Gmail’s contact manager intolerable and gave up,” reported MacLean. A statistical analysis of the results showed that SocialFlows was significantly faster and easier to use, and the groups it generated were significantly more useful than those created with Gmail.

Eric Gilbert, who recently started a research lab at Georgia Tech to analyze social media and explore new types of design for social Web services, says that managing friends lists is “a problem space calling out for a solution.” He previously built a service called WeMeddle that automatically creates lists of a person’s Twitter contacts from their past activity.

Gilbert says SocialFlows does a good job of taming the complexities of creating lists of different types of friends. But creating lists is, in some ways, just a first step. “A remaining challenge is, how do you control how you communicate and message the people in these lists,” he says. Neither SocialFlows nor WeMeddle have taken that next step, says Gilbert, but this may be much more complex than simply deriving groups. “The stream, which lumps everything together, has come to be seen as very central to social media,” he explains. This fact makes Facebook, Twitter, and other services currently unsuited to maintaining separate social circles within a  larger network.

That’s not something MacLean and colleagues plan to take on soon, though. The researchers are focusing on making tweaks to the algorithms and studying how the social groups they identify in people’s data change over time. In the future, SocialFlows may even be able to suggest changes to a person’s friend lists as his or her collection of contacts and patterns of communication evolve. “Ideally, [SocialFlows] would suggest changes to your current topology rather than wiping [out] all of the changes you already made, as it does now,” said MacLean.