CHI - Predicting Tie Strength

Predicting Tie Strength with Social Media
by Eric Gilbert and Karrie Karahalios (University of Illinois at Urbana Champaign)

Summary:
The article discusses the experimental model's ability to predict the relationship strength between members in a Facebook environment. First, "tie strength" is the term defined by social science to measure relationship bonds and/or strength. The weakest tie being an acquaintance to the opposite side classifying a close family member or fiend as a strong tie. Also, research has shown that one's social relationship strengths impact one's mental health, job seeking abilities, and career opportunities.
Next, the article defines the 7 factors involved in determining tie strength: intensity, intimacy, duration, reciprocal services, structural, emotional support, and social distance. By these factors, the researchers formed the following focus for their study:
1) In a social media environment, can one determine tie strength with only these factors?
2) If social media is only used, what are its constraints?

Research Study
In a lab environment, 35 participants who use Facebook were instructed to rate a randomly selected subset of their friend list on 5 specified questions. This produced a dataset of 2184 Facebook friends. Also, the Google GreaseMonkey script produced 74 Facebook variables to aid in predicting the tie strength. All variables were processed as a linear combination of each other and were not standardized.

Results
The study showed that it achieved a Mean Absolute Error of .0994 in tie strength prediction. Also, to improve the model's performance, 10 follow-up interviews were conducted for the difficult to predict friendships. Furthermore, the study demonstrated that tie strength is important in social media and can be studied as a continuous quantity.

Discussion:
It is interesting to see how social media interacts with our social lives. It was surprising how the study demonstrated that the variable for Inbox Thread depth had a negative impact on tie strength. So, the more emails sent on a single issue more likely will decrease the tie strength. I was reflecting to see if this was apparent in my life and it reminds me of project teams. In some classes, you generate lots of emails about the project, but when the course ends more likely so does the communication with that team member.
Also, the study showed that our individual relationships are filtered through our "clique" of existing friends. Again, this seems due to common interests in a friend group would be a natural development for friendship. Overall, the model's results could be used in future social media development and aid in privacy controls.