Min “Ivy” Xing ’15 this summer closely followed a doctor in Texas who has a robust online reputation — his Twitter feed has almost 6,000 followers.
As the physician posted on topics such as Hodgkin lymphoma or metabolite-sensing G protein-coupled receptors, Xing meticulously tracked his tweets and the responses of his followers.
What Xing was trying to get at is how to assess the trust between the doctor and those who follow him. Eventually, her work could expand to enable computers to predict the level of trust between users who are not directly connected in online social networks. That is, if person A is following person B who is following person C, Xing’s model will be able to predict the level of trust that person A has for person C.
This has been a lingering puzzle for Assistant Professor of Computer Science Daniela Oliveira, who hired Xing using part of a National Science Foundation Career Award she received last spring.
“Ivy is working on a topic that has interested me since I finished graduate school: trust in online social networks,” Oliveira explained. Oliveira, who does research on cyber security, says that trust, just like in real life, plays an important role in online behavior, affecting “our decisions regarding visiting a web page, downloading a piece of software, opening a file, etc.”
The value of this research lies in its potential to understand information transmission among a growing number of online social network users. “When one user sends data to another computer, [the computer] has to do a check that takes time and space,” Xing said. She and Oliveira envision a day when a computer’s security mechanisms can take advantage of a person’s high trust value for another user and more efficiently and securely transfer information between them. Alternatively, if the receiver has a low trust value for the sender, the computer can more nimbly protect the receiver.
How Xing Assigns Trust Values to Online Relationships
Oliveira and Xing worked with Twitter because much of its data is publicly available and can be extracted through its application-programming interface, Oliveira said. They selected the doctor based on his strong online reputation, number of tweets — over 11,000 — and his large base of followers.
To start, Oliveira asked Xing to compute the trust value between two people directly connected — in this case, between the doctor and a follower.
To do this, Xing analyzed the doctor’s followers based on five factors. She looked at how often they retweeted his tweets, how many times they replied to the doctor, how many times they mentioned the doctor in their own tweets, how many times they ‘favorited’ his tweets, and the emotional value of their retweets, replies and mentions — whether these were positive, negative or neutral.
Xing, who plans to declare a computer science and math double major next year, had to learn the development platform of Twitter as well as the computer language she needed to manipulate her data. “I basically started from zero; I didn’t know anything,” she said. “The first half of summer was a learning phase, and the second phase was data extraction and programming.”
Continuing this fall, Xing will manually assign a label of trust, untrust or neutral between the follower and the doctor to a subset of the doctor’s followers. Then she will feed these training samples into machine-learning software to create a model to predict the trust values between an unknown sample made up of a follower and the doctor. Following this, she will extend a mathematical model that uses these initial trust values to predict how much people not directly connected can trust one another. “Machine learning is essentially training computers to recognize complex patterns in input data (or training data),” Xing explained. “The software will produce a model that will be able to automatically generate trust values for new sets of users.” Such an idea can be applied to users of Twitter, Facebook and other social network sites.
Xing said she enjoyed working on a project that could improve Internet information transmission and security. “I was very excited to work on this project; I worked on it whenever I could,” she said. She hopes to eventually find a profession that combines math and computer science, such as a cryptologist for the CIA, or by working for NASA or a Silicon Valley company. Xing is from the San Francisco Bat area, having moved there from Guilin, China when she was 12.
Oliveira said she hired Xing for this project because she liked her independence, motivation and her “getting-things-done” attitude as a student in her 101 class. “These are very important attributes in research and in the workplace,” Oliveira said, adding, too, that she’s “always happy to see young women liking and getting involved in computer science.”