For their study, the researchers examined Epinions, a review site that lets members express trust or distrust of other community members; Wikipedia, the user-edited online encyclopedia that lets members vote for or against another's nomination for site "adminship"; and Slashdot, a tech news site that lets participants label each other "Friend" or "Foe."
Specifically, the researchers -- Jure Leskovec of Stanford University and Daniel Huttenlocher and Jon Kleinberg of Cornell University -- relied on the social psychology theories of balance and status. Based on the principles "the enemy of my friend is my enemy," "the friend of my enemy is my enemy," and similar variations, they created an algorithm to calculate the unknown relationship between two individuals based on other relationships within the network.
Beyond simply predicting whether a particular user might attach a positive or negative link to another user, the researchers wanted to answer questions about why such decisions might be made within the network.
"Answers to these questions can help us reason about how negative relationships are used in online systems, and answers that generalize across multiple domains can help to illuminate some of the underlying principals," the researchers wrote in their paper, "Predicting Positive and Negative Links in Online Social Networks." They conclude that a "hidden" relationship between two members of a network can be inferred using an algorithm measuring the positive or negative attitudes of all others in the network.
Thus the research can truly be said to be a kind of sentiment analysis since it tries to gauge the sentiment of two individuals toward each other based on information about other sentiments in the network.
In a recent live chat debate between social media consultant Joe Stanganelli and Web analytics consultant Pierre DeBois, we looked at the question of whether sentiment analysis could provide the amount of data needed to be useful, given the level of accurate measurement currently possible. The researchers found that, using their algorithm at least, they could accurately predict positive and negative relationships between individuals. The results were fairly consistent over several different networks.
How could such an algorithm be used to predict sentiment for your company or organization, especially in online applications? Share your ideas on the boards below.